From d32137e342389b430bdbe4c64f322b2a6446e6a9 Mon Sep 17 00:00:00 2001
From: Thurston Dang <thurston@google.com>
Date: Tue, 8 Jul 2025 16:49:52 +0000
Subject: [PATCH 1/2] [NFCI][msan] Add avx512bw-intrinsics,
 avx512bw-intrinsics-upgrade tests

Forked from llvm/test/CodeGen/X86.
---
 .../X86/avx512bw-intrinsics-upgrade.ll        | 7652 +++++++++++++++++
 .../X86/avx512bw-intrinsics.ll                | 3687 ++++++++
 2 files changed, 11339 insertions(+)
 create mode 100644 llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
 create mode 100644 llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll

diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
new file mode 100644
index 0000000000000..c8e7db81ec753
--- /dev/null
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
@@ -0,0 +1,7652 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt %s -S -mattr=+avx512f -passes=msan 2>&1 | FileCheck %s
+;
+; Forked from llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
+;
+; Strictly handled:
+; - llvm.x86.avx512.dbpsadbw.512
+; - llvm.x86.avx512.packssdw.512, llvm.x86.avx512.packsswb.512
+; - llvm.x86.avx512.packusdw.512, llvm.x86.avx512.packuswb.512
+; - llvm.x86.avx512.pmaddubs.w.512
+; - llvm.x86.avx512.pmaddw.d.512
+;
+; Heuristically handled:
+; - llvm.sadd.sat.v32i16, llvm.sadd.sat.v64i8
+; - llvm.smax.v32i16, llvm.smax.v64i8
+; - llvm.smin.v32i16, llvm.smin.v64i8
+; - llvm.ssub.sat.v32i16, llvm.ssub.sat.v64i8
+; - llvm.uadd.sat.v32i16, llvm.uadd.sat.v64i8
+; - llvm.umax.v32i16, llvm.umax.v64i8
+; - llvm.umin.v32i16, llvm.umin.v64i8
+; - llvm.usub.sat.v32i16, llvm.usub.sat.v64i8
+; - llvm.x86.avx512.pavg.b.512, llvm.x86.avx512.pavg.w.512
+; - llvm.x86.avx512.permvar.hi.512
+; - llvm.x86.avx512.pmul.hr.sw.512, llvm.x86.avx512.pmulhu.w.512, llvm.x86.avx512.pmulh.w.512
+; - llvm.x86.avx512.pshuf.b.512
+; - llvm.x86.avx512.psllv.w.512, llvm.x86.avx512.psrav.w.512,  llvm.x86.avx512.psrlv.w.512
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+declare i32 @llvm.x86.avx512.kunpck.wd(i32, i32)
+
+define i32 @test_int_x86_avx512_kunpck_wd(i32 %x0, i32 %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_kunpck_wd(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32 [[TMP1]] to <32 x i1>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[X0:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X1:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i1> [[TMP3]], <32 x i1> [[TMP3]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <32 x i1> [[TMP4]], <32 x i1> [[TMP4]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <32 x i1> [[TMP5]], <32 x i1> [[TMP5]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <32 x i1> [[TMP6]], <32 x i1> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = shufflevector <16 x i1> [[_MSPROP1]], <16 x i1> [[_MSPROP]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <16 x i1> [[TMP8]], <16 x i1> [[TMP7]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[_MSPROP2]] to i32
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i1> [[TMP9]] to i32
+; CHECK-NEXT:    store i32 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP11]]
+;
+  %res = call i32 @llvm.x86.avx512.kunpck.wd(i32 %x0, i32 %x1)
+  ret i32 %res
+}
+
+declare i64 @llvm.x86.avx512.kunpck.dq(i64, i64)
+
+define i64 @test_int_x86_avx512_kunpck_qd(i64 %x0, i64 %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_kunpck_qd(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64 [[TMP1]] to <64 x i1>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i64 [[X0:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[X1:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i1> [[TMP3]], <64 x i1> [[TMP3]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <64 x i1> [[TMP4]], <64 x i1> [[TMP4]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <64 x i1> [[TMP5]], <64 x i1> [[TMP5]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <64 x i1> [[TMP6]], <64 x i1> [[TMP6]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = shufflevector <32 x i1> [[_MSPROP1]], <32 x i1> [[_MSPROP]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <32 x i1> [[TMP8]], <32 x i1> [[TMP7]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[_MSPROP2]] to i64
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <64 x i1> [[TMP9]] to i64
+; CHECK-NEXT:    store i64 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP11]]
+;
+  %res = call i64 @llvm.x86.avx512.kunpck.dq(i64 %x0, i64 %x1)
+  ret i64 %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pbroadcast.b.gpr.512(i8, <64 x i8>, i64)
+
+define { <64 x i8>, <64 x i8>, <64 x i8> } @test_int_x86_avx512_mask_pbroadcast_b_gpr_512(i8 %x0, <64 x i8> %x1, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pbroadcast_b_gpr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i8, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <64 x i8> splat (i8 -1), i8 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT3:%.*]] = insertelement <64 x i8> poison, i8 [[X0:%.*]], i64 0
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = shufflevector <64 x i8> [[_MSPROP]], <64 x i8> splat (i8 -1), <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT4:%.*]] = shufflevector <64 x i8> [[DOTSPLATINSERT3]], <64 x i8> poison, <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = insertelement <64 x i8> splat (i8 -1), i8 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT1:%.*]] = insertelement <64 x i8> poison, i8 [[X0]], i64 0
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = shufflevector <64 x i8> [[_MSPROP6]], <64 x i8> splat (i8 -1), <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT2:%.*]] = shufflevector <64 x i8> [[DOTSPLATINSERT1]], <64 x i8> poison, <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[_MSPROP7]], <64 x i8> [[TMP3]]
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[DOTSPLAT2]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], [[_MSPROP7]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP4]], <64 x i8> [[TMP9]], <64 x i8> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[DOTSPLAT2]], <64 x i8> [[X1]]
+; CHECK-NEXT:    [[_MSPROP8:%.*]] = insertelement <64 x i8> splat (i8 -1), i8 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <64 x i8> poison, i8 [[X0]], i64 0
+; CHECK-NEXT:    [[_MSPROP9:%.*]] = shufflevector <64 x i8> [[_MSPROP8]], <64 x i8> splat (i8 -1), <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <64 x i8> [[DOTSPLATINSERT]], <64 x i8> poison, <64 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[_MSPROP9]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i8> [[DOTSPLAT]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[_MSPROP9]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i8> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT10:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP16]], <64 x i8> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[DOTSPLAT]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } { <64 x i8> splat (i8 -1), <64 x i8> splat (i8 -1), <64 x i8> splat (i8 -1) }, <64 x i8> [[_MSPROP5]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } poison, <64 x i8> [[DOTSPLAT4]], 0
+; CHECK-NEXT:    [[TMP19:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP18]], <64 x i8> [[_MSPROP_SELECT]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[RES3]], <64 x i8> [[TMP10]], 1
+; CHECK-NEXT:    [[TMP20:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP19]], <64 x i8> [[_MSPROP_SELECT10]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[RES4]], <64 x i8> [[TMP17]], 2
+; CHECK-NEXT:    store { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP20]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <64 x i8>, <64 x i8>, <64 x i8> } [[RES5]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pbroadcast.b.gpr.512(i8 %x0, <64 x i8> %x1, i64 -1)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.pbroadcast.b.gpr.512(i8 %x0, <64 x i8> %x1, i64 %mask)
+  %res2 = call <64 x i8> @llvm.x86.avx512.mask.pbroadcast.b.gpr.512(i8 %x0, <64 x i8> zeroinitializer, i64 %mask)
+  %res3 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } poison, <64 x i8>  %res, 0
+  %res4 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> }  %res3, <64 x i8> %res1, 1
+  %res5 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> }  %res4, <64 x i8> %res2, 2
+  ret { <64 x i8>, <64 x i8>, <64 x i8> } %res5
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pbroadcast.w.gpr.512(i16, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_pbroadcast_w_gpr_512(i16 %x0, <32 x i16> %x1, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pbroadcast_w_gpr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i16, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <32 x i16> splat (i16 -1), i16 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT3:%.*]] = insertelement <32 x i16> poison, i16 [[X0:%.*]], i64 0
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = shufflevector <32 x i16> [[_MSPROP]], <32 x i16> splat (i16 -1), <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT4:%.*]] = shufflevector <32 x i16> [[DOTSPLATINSERT3]], <32 x i16> poison, <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = insertelement <32 x i16> splat (i16 -1), i16 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT1:%.*]] = insertelement <32 x i16> poison, i16 [[X0]], i64 0
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = shufflevector <32 x i16> [[_MSPROP6]], <32 x i16> splat (i16 -1), <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT2:%.*]] = shufflevector <32 x i16> [[DOTSPLATINSERT1]], <32 x i16> poison, <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[_MSPROP7]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[DOTSPLAT2]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[_MSPROP7]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP4]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[DOTSPLAT2]], <32 x i16> [[X1]]
+; CHECK-NEXT:    [[_MSPROP8:%.*]] = insertelement <32 x i16> splat (i16 -1), i16 [[TMP1]], i64 0
+; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <32 x i16> poison, i16 [[X0]], i64 0
+; CHECK-NEXT:    [[_MSPROP9:%.*]] = shufflevector <32 x i16> [[_MSPROP8]], <32 x i16> splat (i16 -1), <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <32 x i16> [[DOTSPLATINSERT]], <32 x i16> poison, <32 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[DOTSPLAT]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP9]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT10:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[DOTSPLAT]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP5]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[DOTSPLAT4]], 0
+; CHECK-NEXT:    [[TMP19:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP18]], <32 x i16> [[_MSPROP_SELECT]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP10]], 1
+; CHECK-NEXT:    [[TMP20:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP19]], <32 x i16> [[_MSPROP_SELECT10]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP17]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP20]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pbroadcast.w.gpr.512(i16 %x0, <32 x i16> %x1, i32 -1)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.pbroadcast.w.gpr.512(i16 %x0, <32 x i16> %x1, i32 %mask)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.pbroadcast.w.gpr.512(i16 %x0, <32 x i16> zeroinitializer, i32 %mask)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+  }
+
+declare void @llvm.x86.avx512.mask.storeu.b.512(ptr, <64 x i8>, i64)
+
+define void @test_int_x86_avx512_mask_storeu_b_512(ptr %ptr1, ptr %ptr2, <64 x i8> %x1, i64 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_storeu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP1]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[X2:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR1:%.*]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    call void @llvm.masked.store.v64i8.p0(<64 x i8> [[TMP2]], ptr [[TMP9]], i32 1, <64 x i1> [[TMP6]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[TMP5]] to i64
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i64 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1:![0-9]+]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7:[0-9]+]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    call void @llvm.masked.store.v64i8.p0(<64 x i8> [[X1:%.*]], ptr [[PTR1]], i32 1, <64 x i1> [[TMP6]])
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP2]], label [[TMP13:%.*]], label [[TMP14:%.*]], !prof [[PROF1]]
+; CHECK:       13:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       14:
+; CHECK-NEXT:    [[TMP15:%.*]] = ptrtoint ptr [[PTR2:%.*]] to i64
+; CHECK-NEXT:    [[TMP16:%.*]] = xor i64 [[TMP15]], 87960930222080
+; CHECK-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[TMP16]] to ptr
+; CHECK-NEXT:    store <64 x i8> [[TMP2]], ptr [[TMP17]], align 1
+; CHECK-NEXT:    store <64 x i8> [[X1]], ptr [[PTR2]], align 1
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.x86.avx512.mask.storeu.b.512(ptr %ptr1, <64 x i8> %x1, i64 %x2)
+  call void @llvm.x86.avx512.mask.storeu.b.512(ptr %ptr2, <64 x i8> %x1, i64 -1)
+  ret void
+}
+
+declare void @llvm.x86.avx512.mask.storeu.w.512(ptr, <32 x i16>, i32)
+
+define void @test_int_x86_avx512_mask_storeu_w_512(ptr %ptr1, ptr %ptr2, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_storeu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP1]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR1:%.*]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    call void @llvm.masked.store.v32i16.p0(<32 x i16> [[TMP2]], ptr [[TMP9]], i32 1, <32 x i1> [[TMP6]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[TMP5]] to i32
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i32 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    call void @llvm.masked.store.v32i16.p0(<32 x i16> [[X1:%.*]], ptr [[PTR1]], i32 1, <32 x i1> [[TMP6]])
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP2]], label [[TMP13:%.*]], label [[TMP14:%.*]], !prof [[PROF1]]
+; CHECK:       13:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       14:
+; CHECK-NEXT:    [[TMP15:%.*]] = ptrtoint ptr [[PTR2:%.*]] to i64
+; CHECK-NEXT:    [[TMP16:%.*]] = xor i64 [[TMP15]], 87960930222080
+; CHECK-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[TMP16]] to ptr
+; CHECK-NEXT:    store <32 x i16> [[TMP2]], ptr [[TMP17]], align 1
+; CHECK-NEXT:    store <32 x i16> [[X1]], ptr [[PTR2]], align 1
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.x86.avx512.mask.storeu.w.512(ptr %ptr1, <32 x i16> %x1, i32 %x2)
+  call void @llvm.x86.avx512.mask.storeu.w.512(ptr %ptr2, <32 x i16> %x1, i32 -1)
+  ret void
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.loadu.w.512(ptr, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_loadu_w_512(ptr %ptr, ptr %ptr2, <32 x i16> %x1, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_loadu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[TMP6:%.*]] = load <32 x i16>, ptr [[PTR:%.*]], align 1
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 1
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[PTR2:%.*]] to i64
+; CHECK-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], 87960930222080
+; CHECK-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP13]] to ptr
+; CHECK-NEXT:    [[_MSMASKEDLD:%.*]] = call <32 x i16> @llvm.masked.load.v32i16.p0(ptr [[TMP14]], i32 1, <32 x i1> [[TMP11]], <32 x i16> [[_MSLD]])
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i1> [[TMP10]] to i32
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i32 [[TMP15]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP16:%.*]], label [[TMP17:%.*]], !prof [[PROF1]]
+; CHECK:       16:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       17:
+; CHECK-NEXT:    [[TMP18:%.*]] = call <32 x i16> @llvm.masked.load.v32i16.p0(ptr [[PTR2]], i32 1, <32 x i1> [[TMP11]], <32 x i16> [[TMP6]])
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP21:%.*]] = ptrtoint ptr [[PTR]] to i64
+; CHECK-NEXT:    [[TMP22:%.*]] = xor i64 [[TMP21]], 87960930222080
+; CHECK-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP22]] to ptr
+; CHECK-NEXT:    [[_MSMASKEDLD1:%.*]] = call <32 x i16> @llvm.masked.load.v32i16.p0(ptr [[TMP23]], i32 1, <32 x i1> [[TMP20]], <32 x i16> zeroinitializer)
+; CHECK-NEXT:    [[_MSCMP4:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <32 x i1> [[TMP19]] to i32
+; CHECK-NEXT:    [[_MSCMP5:%.*]] = icmp ne i32 [[TMP24]], 0
+; CHECK-NEXT:    [[_MSOR6:%.*]] = or i1 [[_MSCMP4]], [[_MSCMP5]]
+; CHECK-NEXT:    br i1 [[_MSOR6]], label [[TMP25:%.*]], label [[TMP26:%.*]], !prof [[PROF1]]
+; CHECK:       25:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       26:
+; CHECK-NEXT:    [[TMP27:%.*]] = call <32 x i16> @llvm.masked.load.v32i16.p0(ptr [[PTR]], i32 1, <32 x i1> [[TMP20]], <32 x i16> zeroinitializer)
+; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSLD]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP6]], 0
+; CHECK-NEXT:    [[TMP29:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP28]], <32 x i16> [[_MSMASKEDLD]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP18]], 1
+; CHECK-NEXT:    [[TMP30:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP29]], <32 x i16> [[_MSMASKEDLD1]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP27]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP30]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.loadu.w.512(ptr %ptr, <32 x i16> %x1, i32 -1)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.loadu.w.512(ptr %ptr2, <32 x i16> %res, i32 %mask)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.loadu.w.512(ptr %ptr, <32 x i16> zeroinitializer, i32 %mask)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.loadu.b.512(ptr, <64 x i8>, i64)
+
+define { <64 x i8>, <64 x i8>, <64 x i8> } @test_int_x86_avx512_mask_loadu_b_512(ptr %ptr, ptr %ptr2, <64 x i8> %x1, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_loadu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[TMP6:%.*]] = load <64 x i8>, ptr [[PTR:%.*]], align 1
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP9]], align 1
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = ptrtoint ptr [[PTR2:%.*]] to i64
+; CHECK-NEXT:    [[TMP13:%.*]] = xor i64 [[TMP12]], 87960930222080
+; CHECK-NEXT:    [[TMP14:%.*]] = inttoptr i64 [[TMP13]] to ptr
+; CHECK-NEXT:    [[_MSMASKEDLD:%.*]] = call <64 x i8> @llvm.masked.load.v64i8.p0(ptr [[TMP14]], i32 1, <64 x i1> [[TMP11]], <64 x i8> [[_MSLD]])
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <64 x i1> [[TMP10]] to i64
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i64 [[TMP15]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP16:%.*]], label [[TMP17:%.*]], !prof [[PROF1]]
+; CHECK:       16:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       17:
+; CHECK-NEXT:    [[TMP18:%.*]] = call <64 x i8> @llvm.masked.load.v64i8.p0(ptr [[PTR2]], i32 1, <64 x i1> [[TMP11]], <64 x i8> [[TMP6]])
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP21:%.*]] = ptrtoint ptr [[PTR]] to i64
+; CHECK-NEXT:    [[TMP22:%.*]] = xor i64 [[TMP21]], 87960930222080
+; CHECK-NEXT:    [[TMP23:%.*]] = inttoptr i64 [[TMP22]] to ptr
+; CHECK-NEXT:    [[_MSMASKEDLD1:%.*]] = call <64 x i8> @llvm.masked.load.v64i8.p0(ptr [[TMP23]], i32 1, <64 x i1> [[TMP20]], <64 x i8> zeroinitializer)
+; CHECK-NEXT:    [[_MSCMP4:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <64 x i1> [[TMP19]] to i64
+; CHECK-NEXT:    [[_MSCMP5:%.*]] = icmp ne i64 [[TMP24]], 0
+; CHECK-NEXT:    [[_MSOR6:%.*]] = or i1 [[_MSCMP4]], [[_MSCMP5]]
+; CHECK-NEXT:    br i1 [[_MSOR6]], label [[TMP25:%.*]], label [[TMP26:%.*]], !prof [[PROF1]]
+; CHECK:       25:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       26:
+; CHECK-NEXT:    [[TMP27:%.*]] = call <64 x i8> @llvm.masked.load.v64i8.p0(ptr [[PTR]], i32 1, <64 x i1> [[TMP20]], <64 x i8> zeroinitializer)
+; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } { <64 x i8> splat (i8 -1), <64 x i8> splat (i8 -1), <64 x i8> splat (i8 -1) }, <64 x i8> [[_MSLD]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } poison, <64 x i8> [[TMP6]], 0
+; CHECK-NEXT:    [[TMP29:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP28]], <64 x i8> [[_MSMASKEDLD]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[RES3]], <64 x i8> [[TMP18]], 1
+; CHECK-NEXT:    [[TMP30:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP29]], <64 x i8> [[_MSMASKEDLD1]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } [[RES4]], <64 x i8> [[TMP27]], 2
+; CHECK-NEXT:    store { <64 x i8>, <64 x i8>, <64 x i8> } [[TMP30]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <64 x i8>, <64 x i8>, <64 x i8> } [[RES5]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.loadu.b.512(ptr %ptr, <64 x i8> %x1, i64 -1)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.loadu.b.512(ptr %ptr2, <64 x i8> %res, i64 %mask)
+  %res2 = call <64 x i8> @llvm.x86.avx512.mask.loadu.b.512(ptr %ptr, <64 x i8> zeroinitializer, i64 %mask)
+  %res3 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> } poison, <64 x i8>  %res, 0
+  %res4 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> }  %res3, <64 x i8> %res1, 1
+  %res5 = insertvalue { <64 x i8>, <64 x i8>, <64 x i8> }  %res4, <64 x i8> %res2, 2
+  ret { <64 x i8>, <64 x i8>, <64 x i8> } %res5
+}
+
+declare <8 x i64> @llvm.x86.avx512.psll.dq.512(<8 x i64>, i32)
+
+define { <8 x i64>, <8 x i64> } @test_int_x86_avx512_psll_dq_512(<8 x i64> %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psll_dq_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i64>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i64> [[TMP1]] to <64 x i8>
+; CHECK-NEXT:    [[CAST2:%.*]] = bitcast <8 x i64> [[X0:%.*]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[TMP2]], <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[CAST2]], <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <64 x i8> [[_MSPROP]] to <8 x i64>
+; CHECK-NEXT:    [[CAST3:%.*]] = bitcast <64 x i8> [[TMP3]] to <8 x i64>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i64> [[TMP1]] to <64 x i8>
+; CHECK-NEXT:    [[CAST:%.*]] = bitcast <8 x i64> [[X0]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[TMP5]], <64 x i32> <i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 88, i32 89, i32 90, i32 91, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 104, i32 105, i32 106, i32 107, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119, i32 120, i32 121, i32 122, i32 123>
+; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[CAST]], <64 x i32> <i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 88, i32 89, i32 90, i32 91, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 104, i32 105, i32 106, i32 107, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119, i32 120, i32 121, i32 122, i32 123>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <64 x i8> [[_MSPROP4]] to <8 x i64>
+; CHECK-NEXT:    [[CAST1:%.*]] = bitcast <64 x i8> [[TMP6]] to <8 x i64>
+; CHECK-NEXT:    [[TMP8:%.*]] = insertvalue { <8 x i64>, <8 x i64> } { <8 x i64> splat (i64 -1), <8 x i64> splat (i64 -1) }, <8 x i64> [[TMP4]], 0
+; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64> [[CAST3]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = insertvalue { <8 x i64>, <8 x i64> } [[TMP8]], <8 x i64> [[TMP7]], 1
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <8 x i64>, <8 x i64> } [[RES2]], <8 x i64> [[CAST1]], 1
+; CHECK-NEXT:    store { <8 x i64>, <8 x i64> } [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <8 x i64>, <8 x i64> } [[RES3]]
+;
+  %res = call <8 x i64> @llvm.x86.avx512.psll.dq.512(<8 x i64> %x0, i32 8)
+  %res1 = call <8 x i64> @llvm.x86.avx512.psll.dq.512(<8 x i64> %x0, i32 4)
+  %res2 = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64>  %res, 0
+  %res3 = insertvalue { <8 x i64>, <8 x i64> }  %res2, <8 x i64> %res1, 1
+  ret { <8 x i64>, <8 x i64> } %res3
+}
+
+define <8 x i64> @test_int_x86_avx512_psll_load_dq_512(ptr %p0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psll_load_dq_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP2:%.*]], label [[TMP3:%.*]], !prof [[PROF1]]
+; CHECK:       2:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       3:
+; CHECK-NEXT:    [[X0:%.*]] = load <8 x i64>, ptr [[P0:%.*]], align 64
+; CHECK-NEXT:    [[TMP4:%.*]] = ptrtoint ptr [[P0]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = xor i64 [[TMP4]], 87960930222080
+; CHECK-NEXT:    [[TMP6:%.*]] = inttoptr i64 [[TMP5]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i64>, ptr [[TMP6]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i64> [[_MSLD]] to <64 x i8>
+; CHECK-NEXT:    [[CAST:%.*]] = bitcast <8 x i64> [[X0]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[TMP7]], <64 x i32> <i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 88, i32 89, i32 90, i32 91, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 104, i32 105, i32 106, i32 107, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119, i32 120, i32 121, i32 122, i32 123>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <64 x i8> zeroinitializer, <64 x i8> [[CAST]], <64 x i32> <i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 88, i32 89, i32 90, i32 91, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 104, i32 105, i32 106, i32 107, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119, i32 120, i32 121, i32 122, i32 123>
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <64 x i8> [[_MSPROP]] to <8 x i64>
+; CHECK-NEXT:    [[CAST1:%.*]] = bitcast <64 x i8> [[TMP8]] to <8 x i64>
+; CHECK-NEXT:    store <8 x i64> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <8 x i64> [[CAST1]]
+;
+  %x0 = load <8 x i64>, ptr%p0
+  %res = call <8 x i64> @llvm.x86.avx512.psll.dq.512(<8 x i64> %x0, i32 4)
+  ret <8 x i64> %res
+}
+
+declare <8 x i64> @llvm.x86.avx512.psrl.dq.512(<8 x i64>, i32)
+
+define { <8 x i64>, <8 x i64> } @test_int_x86_avx512_psrl_dq_512(<8 x i64> %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrl_dq_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i64>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i64> [[TMP1]] to <64 x i8>
+; CHECK-NEXT:    [[CAST2:%.*]] = bitcast <8 x i64> [[X0:%.*]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP2]], <64 x i8> zeroinitializer, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <64 x i8> [[CAST2]], <64 x i8> zeroinitializer, <64 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 84, i32 85, i32 86, i32 87, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 100, i32 101, i32 102, i32 103, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115, i32 116, i32 117, i32 118, i32 119>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <64 x i8> [[_MSPROP]] to <8 x i64>
+; CHECK-NEXT:    [[CAST3:%.*]] = bitcast <64 x i8> [[TMP3]] to <8 x i64>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i64> [[TMP1]] to <64 x i8>
+; CHECK-NEXT:    [[CAST:%.*]] = bitcast <8 x i64> [[X0]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = shufflevector <64 x i8> [[TMP5]], <64 x i8> zeroinitializer, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115>
+; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <64 x i8> [[CAST]], <64 x i8> zeroinitializer, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <64 x i8> [[_MSPROP4]] to <8 x i64>
+; CHECK-NEXT:    [[CAST1:%.*]] = bitcast <64 x i8> [[TMP6]] to <8 x i64>
+; CHECK-NEXT:    [[TMP8:%.*]] = insertvalue { <8 x i64>, <8 x i64> } { <8 x i64> splat (i64 -1), <8 x i64> splat (i64 -1) }, <8 x i64> [[TMP4]], 0
+; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64> [[CAST3]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = insertvalue { <8 x i64>, <8 x i64> } [[TMP8]], <8 x i64> [[TMP7]], 1
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <8 x i64>, <8 x i64> } [[RES2]], <8 x i64> [[CAST1]], 1
+; CHECK-NEXT:    store { <8 x i64>, <8 x i64> } [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <8 x i64>, <8 x i64> } [[RES3]]
+;
+  %res = call <8 x i64> @llvm.x86.avx512.psrl.dq.512(<8 x i64> %x0, i32 8)
+  %res1 = call <8 x i64> @llvm.x86.avx512.psrl.dq.512(<8 x i64> %x0, i32 4)
+  %res2 = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64>  %res, 0
+  %res3 = insertvalue { <8 x i64>, <8 x i64> }  %res2, <8 x i64> %res1, 1
+  ret { <8 x i64>, <8 x i64> } %res3
+}
+
+define <8 x i64> @test_int_x86_avx512_psrl_load_dq_512(ptr %p0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrl_load_dq_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP2:%.*]], label [[TMP3:%.*]], !prof [[PROF1]]
+; CHECK:       2:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       3:
+; CHECK-NEXT:    [[X0:%.*]] = load <8 x i64>, ptr [[P0:%.*]], align 64
+; CHECK-NEXT:    [[TMP4:%.*]] = ptrtoint ptr [[P0]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = xor i64 [[TMP4]], 87960930222080
+; CHECK-NEXT:    [[TMP6:%.*]] = inttoptr i64 [[TMP5]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i64>, ptr [[TMP6]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i64> [[_MSLD]] to <64 x i8>
+; CHECK-NEXT:    [[CAST:%.*]] = bitcast <8 x i64> [[X0]] to <64 x i8>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP7]], <64 x i8> zeroinitializer, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <64 x i8> [[CAST]], <64 x i8> zeroinitializer, <64 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 82, i32 83, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 98, i32 99, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113, i32 114, i32 115>
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <64 x i8> [[_MSPROP]] to <8 x i64>
+; CHECK-NEXT:    [[CAST1:%.*]] = bitcast <64 x i8> [[TMP8]] to <8 x i64>
+; CHECK-NEXT:    store <8 x i64> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <8 x i64> [[CAST1]]
+;
+  %x0 = load <8 x i64>, ptr%p0
+  %res = call <8 x i64> @llvm.x86.avx512.psrl.dq.512(<8 x i64> %x0, i32 4)
+  ret <8 x i64> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.palignr.512(<64 x i8>, <64 x i8>, i32, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_palignr_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_palignr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    [[PALIGNR:%.*]] = shufflevector <64 x i8> [[X1:%.*]], <64 x i8> [[X0:%.*]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[PALIGNR]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.palignr.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <64 x i8> %x3, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_palignr_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x3, i64 %x4) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_palignr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    [[PALIGNR:%.*]] = shufflevector <64 x i8> [[X1:%.*]], <64 x i8> [[X0:%.*]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[X4:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[PALIGNR]], [[X3:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[PALIGNR]], <64 x i8> [[X3]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.palignr.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <64 x i8> %x3, i64 %x4)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_maskz_palignr_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x4) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_palignr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    [[PALIGNR:%.*]] = shufflevector <64 x i8> [[X1:%.*]], <64 x i8> [[X0:%.*]], <64 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 80, i32 81, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 96, i32 97, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 112, i32 113>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[X4:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[PALIGNR]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP4]], <64 x i8> [[TMP9]], <64 x i8> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[PALIGNR]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP10]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.palignr.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <64 x i8> zeroinitializer, i64 %x4)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pshufh.w.512(<32 x i16>, i32, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pshufh_w_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshufh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufh.w.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pshufh_w_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pshufh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufh.w.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_pshufh_w_512(<32 x i16> %x0, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pshufh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 4, i32 4, i32 8, i32 9, i32 10, i32 11, i32 15, i32 12, i32 12, i32 12, i32 16, i32 17, i32 18, i32 19, i32 23, i32 20, i32 20, i32 20, i32 24, i32 25, i32 26, i32 27, i32 31, i32 28, i32 28, i32 28>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP4]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP3]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP10]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufh.w.512(<32 x i16> %x0, i32 3, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pshufl.w.512(<32 x i16>, i32, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pshufl_w_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshufl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufl.w.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pshufl_w_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pshufl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufl.w.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_pshufl_w_512(<32 x i16> %x0, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pshufl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP1]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X0]], <32 x i32> <i32 3, i32 0, i32 0, i32 0, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 8, i32 8, i32 12, i32 13, i32 14, i32 15, i32 19, i32 16, i32 16, i32 16, i32 20, i32 21, i32 22, i32 23, i32 27, i32 24, i32 24, i32 24, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP4]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP3]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP10]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pshufl.w.512(<32 x i16> %x0, i32 3, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define i64 @test_pcmpeq_b(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_pcmpeq_b(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[TMP4]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <64 x i8> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP]] to i64
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <64 x i1> [[TMP9]] to i64
+; CHECK-NEXT:    store i64 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP11]]
+;
+  %res = call i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8> %a, <64 x i8> %b, i64 -1)
+  ret i64 %res
+}
+
+define i64 @test_mask_pcmpeq_b(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_pcmpeq_b(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <64 x i8> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <64 x i8> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[TMP5]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <64 x i8> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <64 x i8> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <64 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <64 x i1> [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <64 x i1> [[TMP18]] to i64
+; CHECK-NEXT:    store i64 [[TMP19]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP20]]
+;
+  %res = call i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8> %a, <64 x i8> %b, i64 %mask)
+  ret i64 %res
+}
+
+declare i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8>, <64 x i8>, i64)
+
+define i32 @test_pcmpeq_w(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_pcmpeq_w(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <32 x i16> [[TMP4]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <32 x i16> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <32 x i16> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP]] to i32
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i1> [[TMP9]] to i32
+; CHECK-NEXT:    store i32 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP11]]
+;
+  %res = call i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16> %a, <32 x i16> %b, i32 -1)
+  ret i32 %res
+}
+
+define i32 @test_mask_pcmpeq_w(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_pcmpeq_w(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <32 x i16> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <32 x i16> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[TMP5]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <32 x i16> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <32 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i1> [[TMP17]] to i32
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <32 x i1> [[TMP18]] to i32
+; CHECK-NEXT:    store i32 [[TMP19]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP20]]
+;
+  %res = call i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16> %a, <32 x i16> %b, i32 %mask)
+  ret i32 %res
+}
+
+declare i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16>, <32 x i16>, i32)
+
+define i64 @test_pcmpgt_b(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_pcmpgt_b(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A:%.*]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP5:%.*]] = and <64 x i8> [[TMP3]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = or <64 x i8> [[TMP3]], [[TMP1]]
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[B:%.*]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP9:%.*]] = and <64 x i8> [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP7]], [[TMP2]]
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ugt <64 x i8> [[TMP5]], [[TMP10]]
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ugt <64 x i8> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i1> [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp sgt <64 x i8> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <64 x i1> [[TMP13]] to i64
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast <64 x i1> [[TMP14]] to i64
+; CHECK-NEXT:    store i64 [[TMP15]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP16]]
+;
+  %res = call i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8> %a, <64 x i8> %b, i64 -1)
+  ret i64 %res
+}
+
+define i64 @test_mask_pcmpgt_b(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_pcmpgt_b(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <64 x i8> [[A:%.*]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP5:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP6:%.*]] = and <64 x i8> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <64 x i8> [[TMP4]], [[TMP1]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[B:%.*]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP10:%.*]] = and <64 x i8> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP8]], [[TMP2]]
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ugt <64 x i8> [[TMP6]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = icmp ugt <64 x i8> [[TMP7]], [[TMP10]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i1> [[TMP12]], [[TMP13]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt <64 x i8> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = and <64 x i1> [[TMP14]], [[TMP16]]
+; CHECK-NEXT:    [[TMP19:%.*]] = and <64 x i1> [[TMP15]], [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <64 x i1> [[TMP14]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = or <64 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <64 x i1> [[TMP21]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = and <64 x i1> [[TMP15]], [[TMP17]]
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <64 x i1> [[TMP22]] to i64
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <64 x i1> [[TMP23]] to i64
+; CHECK-NEXT:    store i64 [[TMP24]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP25]]
+;
+  %res = call i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8> %a, <64 x i8> %b, i64 %mask)
+  ret i64 %res
+}
+
+declare i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8>, <64 x i8>, i64)
+
+define i32 @test_pcmpgt_w(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_pcmpgt_w(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A:%.*]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP5:%.*]] = and <32 x i16> [[TMP3]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = or <32 x i16> [[TMP3]], [[TMP1]]
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[B:%.*]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP9:%.*]] = and <32 x i16> [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP7]], [[TMP2]]
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ugt <32 x i16> [[TMP5]], [[TMP10]]
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ugt <32 x i16> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i1> [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp sgt <32 x i16> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i1> [[TMP13]] to i32
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast <32 x i1> [[TMP14]] to i32
+; CHECK-NEXT:    store i32 [[TMP15]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP16]]
+;
+  %res = call i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16> %a, <32 x i16> %b, i32 -1)
+  ret i32 %res
+}
+
+define i32 @test_mask_pcmpgt_w(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_pcmpgt_w(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <32 x i16> [[A:%.*]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP5:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP6:%.*]] = and <32 x i16> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <32 x i16> [[TMP4]], [[TMP1]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[B:%.*]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i16> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP8]], [[TMP2]]
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ugt <32 x i16> [[TMP6]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = icmp ugt <32 x i16> [[TMP7]], [[TMP10]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i1> [[TMP12]], [[TMP13]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt <32 x i16> [[A]], [[B]]
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = and <32 x i1> [[TMP14]], [[TMP16]]
+; CHECK-NEXT:    [[TMP19:%.*]] = and <32 x i1> [[TMP15]], [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <32 x i1> [[TMP14]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = or <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <32 x i1> [[TMP21]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = and <32 x i1> [[TMP15]], [[TMP17]]
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <32 x i1> [[TMP22]] to i32
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <32 x i1> [[TMP23]] to i32
+; CHECK-NEXT:    store i32 [[TMP24]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP25]]
+;
+  %res = call i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16> %a, <32 x i16> %b, i32 %mask)
+  ret i32 %res
+}
+
+declare i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16>, <32 x i16>, i32)
+
+declare <64 x i8> @llvm.x86.avx512.mask.punpckhb.w.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_punpckhb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_punpckhb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 8, i32 72, i32 9, i32 73, i32 10, i32 74, i32 11, i32 75, i32 12, i32 76, i32 13, i32 77, i32 14, i32 78, i32 15, i32 79, i32 24, i32 88, i32 25, i32 89, i32 26, i32 90, i32 27, i32 91, i32 28, i32 92, i32 29, i32 93, i32 30, i32 94, i32 31, i32 95, i32 40, i32 104, i32 41, i32 105, i32 42, i32 106, i32 43, i32 107, i32 44, i32 108, i32 45, i32 109, i32 46, i32 110, i32 47, i32 111, i32 56, i32 120, i32 57, i32 121, i32 58, i32 122, i32 59, i32 123, i32 60, i32 124, i32 61, i32 125, i32 62, i32 126, i32 63, i32 127>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i32> <i32 8, i32 72, i32 9, i32 73, i32 10, i32 74, i32 11, i32 75, i32 12, i32 76, i32 13, i32 77, i32 14, i32 78, i32 15, i32 79, i32 24, i32 88, i32 25, i32 89, i32 26, i32 90, i32 27, i32 91, i32 28, i32 92, i32 29, i32 93, i32 30, i32 94, i32 31, i32 95, i32 40, i32 104, i32 41, i32 105, i32 42, i32 106, i32 43, i32 107, i32 44, i32 108, i32 45, i32 109, i32 46, i32 110, i32 47, i32 111, i32 56, i32 120, i32 57, i32 121, i32 58, i32 122, i32 59, i32 123, i32 60, i32 124, i32 61, i32 125, i32 62, i32 126, i32 63, i32 127>
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.punpckhb.w.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_punpckhb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_punpckhb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 8, i32 72, i32 9, i32 73, i32 10, i32 74, i32 11, i32 75, i32 12, i32 76, i32 13, i32 77, i32 14, i32 78, i32 15, i32 79, i32 24, i32 88, i32 25, i32 89, i32 26, i32 90, i32 27, i32 91, i32 28, i32 92, i32 29, i32 93, i32 30, i32 94, i32 31, i32 95, i32 40, i32 104, i32 41, i32 105, i32 42, i32 106, i32 43, i32 107, i32 44, i32 108, i32 45, i32 109, i32 46, i32 110, i32 47, i32 111, i32 56, i32 120, i32 57, i32 121, i32 58, i32 122, i32 59, i32 123, i32 60, i32 124, i32 61, i32 125, i32 62, i32 126, i32 63, i32 127>
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i32> <i32 8, i32 72, i32 9, i32 73, i32 10, i32 74, i32 11, i32 75, i32 12, i32 76, i32 13, i32 77, i32 14, i32 78, i32 15, i32 79, i32 24, i32 88, i32 25, i32 89, i32 26, i32 90, i32 27, i32 91, i32 28, i32 92, i32 29, i32 93, i32 30, i32 94, i32 31, i32 95, i32 40, i32 104, i32 41, i32 105, i32 42, i32 106, i32 43, i32 107, i32 44, i32 108, i32 45, i32 109, i32 46, i32 110, i32 47, i32 111, i32 56, i32 120, i32 57, i32 121, i32 58, i32 122, i32 59, i32 123, i32 60, i32 124, i32 61, i32 125, i32 62, i32 126, i32 63, i32 127>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.punpckhb.w.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.punpcklb.w.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_punpcklb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_punpcklb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 0, i32 64, i32 1, i32 65, i32 2, i32 66, i32 3, i32 67, i32 4, i32 68, i32 5, i32 69, i32 6, i32 70, i32 7, i32 71, i32 16, i32 80, i32 17, i32 81, i32 18, i32 82, i32 19, i32 83, i32 20, i32 84, i32 21, i32 85, i32 22, i32 86, i32 23, i32 87, i32 32, i32 96, i32 33, i32 97, i32 34, i32 98, i32 35, i32 99, i32 36, i32 100, i32 37, i32 101, i32 38, i32 102, i32 39, i32 103, i32 48, i32 112, i32 49, i32 113, i32 50, i32 114, i32 51, i32 115, i32 52, i32 116, i32 53, i32 117, i32 54, i32 118, i32 55, i32 119>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i32> <i32 0, i32 64, i32 1, i32 65, i32 2, i32 66, i32 3, i32 67, i32 4, i32 68, i32 5, i32 69, i32 6, i32 70, i32 7, i32 71, i32 16, i32 80, i32 17, i32 81, i32 18, i32 82, i32 19, i32 83, i32 20, i32 84, i32 21, i32 85, i32 22, i32 86, i32 23, i32 87, i32 32, i32 96, i32 33, i32 97, i32 34, i32 98, i32 35, i32 99, i32 36, i32 100, i32 37, i32 101, i32 38, i32 102, i32 39, i32 103, i32 48, i32 112, i32 49, i32 113, i32 50, i32 114, i32 51, i32 115, i32 52, i32 116, i32 53, i32 117, i32 54, i32 118, i32 55, i32 119>
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.punpcklb.w.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_punpcklb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_punpcklb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <64 x i8> [[TMP1]], <64 x i8> [[TMP2]], <64 x i32> <i32 0, i32 64, i32 1, i32 65, i32 2, i32 66, i32 3, i32 67, i32 4, i32 68, i32 5, i32 69, i32 6, i32 70, i32 7, i32 71, i32 16, i32 80, i32 17, i32 81, i32 18, i32 82, i32 19, i32 83, i32 20, i32 84, i32 21, i32 85, i32 22, i32 86, i32 23, i32 87, i32 32, i32 96, i32 33, i32 97, i32 34, i32 98, i32 35, i32 99, i32 36, i32 100, i32 37, i32 101, i32 38, i32 102, i32 39, i32 103, i32 48, i32 112, i32 49, i32 113, i32 50, i32 114, i32 51, i32 115, i32 52, i32 116, i32 53, i32 117, i32 54, i32 118, i32 55, i32 119>
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], <64 x i32> <i32 0, i32 64, i32 1, i32 65, i32 2, i32 66, i32 3, i32 67, i32 4, i32 68, i32 5, i32 69, i32 6, i32 70, i32 7, i32 71, i32 16, i32 80, i32 17, i32 81, i32 18, i32 82, i32 19, i32 83, i32 20, i32 84, i32 21, i32 85, i32 22, i32 86, i32 23, i32 87, i32 32, i32 96, i32 33, i32 97, i32 34, i32 98, i32 35, i32 99, i32 36, i32 100, i32 37, i32 101, i32 38, i32 102, i32 39, i32 103, i32 48, i32 112, i32 49, i32 113, i32 50, i32 114, i32 51, i32 115, i32 52, i32 116, i32 53, i32 117, i32 54, i32 118, i32 55, i32 119>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.punpcklb.w.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.punpckhw.d.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_punpckhw_d_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_punpckhw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP2]], <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 14, i32 46, i32 15, i32 47, i32 20, i32 52, i32 21, i32 53, i32 22, i32 54, i32 23, i32 55, i32 28, i32 60, i32 29, i32 61, i32 30, i32 62, i32 31, i32 63>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]], <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 14, i32 46, i32 15, i32 47, i32 20, i32 52, i32 21, i32 53, i32 22, i32 54, i32 23, i32 55, i32 28, i32 60, i32 29, i32 61, i32 30, i32 62, i32 31, i32 63>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.punpckhw.d.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_punpckhw_d_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_punpckhw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP2]], <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 14, i32 46, i32 15, i32 47, i32 20, i32 52, i32 21, i32 53, i32 22, i32 54, i32 23, i32 55, i32 28, i32 60, i32 29, i32 61, i32 30, i32 62, i32 31, i32 63>
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]], <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 14, i32 46, i32 15, i32 47, i32 20, i32 52, i32 21, i32 53, i32 22, i32 54, i32 23, i32 55, i32 28, i32 60, i32 29, i32 61, i32 30, i32 62, i32 31, i32 63>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.punpckhw.d.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.punpcklw.d.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_punpcklw_d_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_punpcklw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP2]], <32 x i32> <i32 0, i32 32, i32 1, i32 33, i32 2, i32 34, i32 3, i32 35, i32 8, i32 40, i32 9, i32 41, i32 10, i32 42, i32 11, i32 43, i32 16, i32 48, i32 17, i32 49, i32 18, i32 50, i32 19, i32 51, i32 24, i32 56, i32 25, i32 57, i32 26, i32 58, i32 27, i32 59>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]], <32 x i32> <i32 0, i32 32, i32 1, i32 33, i32 2, i32 34, i32 3, i32 35, i32 8, i32 40, i32 9, i32 41, i32 10, i32 42, i32 11, i32 43, i32 16, i32 48, i32 17, i32 49, i32 18, i32 50, i32 19, i32 51, i32 24, i32 56, i32 25, i32 57, i32 26, i32 58, i32 27, i32 59>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.punpcklw.d.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_punpcklw_d_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_punpcklw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i16> [[TMP1]], <32 x i16> [[TMP2]], <32 x i32> <i32 0, i32 32, i32 1, i32 33, i32 2, i32 34, i32 3, i32 35, i32 8, i32 40, i32 9, i32 41, i32 10, i32 42, i32 11, i32 43, i32 16, i32 48, i32 17, i32 49, i32 18, i32 50, i32 19, i32 51, i32 24, i32 56, i32 25, i32 57, i32 26, i32 58, i32 27, i32 59>
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]], <32 x i32> <i32 0, i32 32, i32 1, i32 33, i32 2, i32 34, i32 3, i32 35, i32 8, i32 40, i32 9, i32 41, i32 10, i32 42, i32 11, i32 43, i32 16, i32 48, i32 17, i32 49, i32 18, i32 50, i32 19, i32 51, i32 24, i32 56, i32 25, i32 57, i32 26, i32 58, i32 27, i32 59>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.punpcklw.d.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pmaxs_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaxs_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.smax.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pmaxs_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.smax.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmaxs_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaxs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.smax.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmaxs_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.smax.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pmaxu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaxu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.umax.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pmaxu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.umax.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmaxu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaxu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.umax.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmaxu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.umax.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pmins_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmins_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.smin.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pmins_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.smin.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmins_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmins_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.smin.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmins_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.smin.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pminu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pminu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.umin.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pminu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.umin.v64i8(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pminu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pminu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.umin.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pminu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.umin.v32i16(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmovzxb.w.512(<32 x i8>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmovzxb_w_512(<32 x i8> %x0, <32 x i16> %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmovzxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = zext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP3:%.*]] = zext <32 x i8> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovzxb.w.512(<32 x i8> %x0, <32 x i16> %x1, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmovzxb_w_512(<32 x i8> %x0, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovzxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = zext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = zext <32 x i8> [[TMP4]] to <32 x i16>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP1]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovzxb.w.512(<32 x i8> %x0, <32 x i16> %x1, i32 %x2)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_pmovzxb_w_512(<32 x i8> %x0, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmovzxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = zext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = zext <32 x i8> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP1]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovzxb.w.512(<32 x i8> %x0, <32 x i16> zeroinitializer, i32 %x2)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmovsxb.w.512(<32 x i8>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmovsxb_w_512(<32 x i8> %x0, <32 x i16> %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmovsxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = sext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP3:%.*]] = sext <32 x i8> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovsxb.w.512(<32 x i8> %x0, <32 x i16> %x1, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmovsxb_w_512(<32 x i8> %x0, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovsxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = sext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = sext <32 x i8> [[TMP4]] to <32 x i16>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP1]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovsxb.w.512(<32 x i8> %x0, <32 x i16> %x1, i32 %x2)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_pmovsxb_w_512(<32 x i8> %x0, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmovsxb_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = shufflevector <32 x i8> [[TMP1]], <32 x i8> splat (i8 -1), <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i8> [[X0:%.*]], <32 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = sext <32 x i8> [[_MSPROP]] to <32 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = sext <32 x i8> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP1]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmovsxb.w.512(<32 x i8> %x0, <32 x i16> zeroinitializer, i32 %x2)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psrl.w.512(<32 x i16>, <8 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psrl_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP10]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrl.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psrl_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP16:%.*]] = xor <32 x i16> [[TMP12]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], [[TMP11]]
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP18]], <32 x i16> [[TMP15]]
+; CHECK-NEXT:    [[TMP19:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP12]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP19]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrl.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psrl_w_512(<32 x i16> %x0, <8 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = xor <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP10]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP17]], <32 x i16> [[TMP14]]
+; CHECK-NEXT:    [[TMP18:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP11]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP18]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrl.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psrl.wi.512(<32 x i16>, i32, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_psrl_wi_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrl_wi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 3)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[X0:%.*]], i32 3)
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP9:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[TMP6]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP5]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i16> [[TMP11]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP12]], <32 x i16> [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP6]], <32 x i16> [[X2]]
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 4)
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[X0]], i32 4)
+; CHECK-NEXT:    [[TMP17:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 5)
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[X0]], i32 5)
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i32 [[X3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP22:%.*]] = select <32 x i1> [[TMP21]], <32 x i16> [[TMP18]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP23:%.*]] = xor <32 x i16> [[TMP19]], zeroinitializer
+; CHECK-NEXT:    [[TMP24:%.*]] = or <32 x i16> [[TMP23]], [[TMP18]]
+; CHECK-NEXT:    [[TMP25:%.*]] = or <32 x i16> [[TMP24]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT1:%.*]] = select <32 x i1> [[TMP20]], <32 x i16> [[TMP25]], <32 x i16> [[TMP22]]
+; CHECK-NEXT:    [[TMP26:%.*]] = select <32 x i1> [[TMP21]], <32 x i16> [[TMP19]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP_SELECT]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP13]], 0
+; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP27]], <32 x i16> [[TMP15]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP16]], 1
+; CHECK-NEXT:    [[TMP29:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP28]], <32 x i16> [[_MSPROP_SELECT1]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP26]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP29]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrl.wi.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.psrl.wi.512(<32 x i16> %x0, i32 4, <32 x i16> %x2, i32 -1)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.psrl.wi.512(<32 x i16> %x0, i32 5, <32 x i16> zeroinitializer, i32 %x3)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psra.w.512(<32 x i16>, <8 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psra_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP10]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psra.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psra_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP16:%.*]] = xor <32 x i16> [[TMP12]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], [[TMP11]]
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP18]], <32 x i16> [[TMP15]]
+; CHECK-NEXT:    [[TMP19:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP12]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP19]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psra.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psra_w_512(<32 x i16> %x0, <8 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = xor <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP10]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP17]], <32 x i16> [[TMP14]]
+; CHECK-NEXT:    [[TMP18:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP11]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP18]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psra.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psra.wi.512(<32 x i16>, i32, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_psra_wi_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psra_wi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 3)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[X0:%.*]], i32 3)
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP9:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[TMP6]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP5]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i16> [[TMP11]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP12]], <32 x i16> [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP6]], <32 x i16> [[X2]]
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 4)
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[X0]], i32 4)
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i32 [[X3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = select <32 x i1> [[TMP18]], <32 x i16> [[TMP15]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = xor <32 x i16> [[TMP16]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = or <32 x i16> [[TMP20]], [[TMP15]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <32 x i16> [[TMP21]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT1:%.*]] = select <32 x i1> [[TMP17]], <32 x i16> [[TMP22]], <32 x i16> [[TMP19]]
+; CHECK-NEXT:    [[TMP23:%.*]] = select <32 x i1> [[TMP18]], <32 x i16> [[TMP16]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP24:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 5)
+; CHECK-NEXT:    [[TMP25:%.*]] = or <32 x i16> [[TMP24]], zeroinitializer
+; CHECK-NEXT:    [[TMP26:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[X0]], i32 5)
+; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP_SELECT]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP13]], 0
+; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP27]], <32 x i16> [[_MSPROP_SELECT1]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP23]], 1
+; CHECK-NEXT:    [[TMP29:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP28]], <32 x i16> [[TMP25]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP26]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP29]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psra.wi.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.psra.wi.512(<32 x i16> %x0, i32 4, <32 x i16> zeroinitializer, i32 %x3)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.psra.wi.512(<32 x i16> %x0, i32 5, <32 x i16> %x2, i32 -1)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psll.w.512(<32 x i16>, <8 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psll_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP10]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psll.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psll_w_512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP16:%.*]] = xor <32 x i16> [[TMP12]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], [[TMP11]]
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP18]], <32 x i16> [[TMP15]]
+; CHECK-NEXT:    [[TMP19:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP12]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP19]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psll.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psll_w_512(<32 x i16> %x0, <8 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[X0:%.*]], <8 x i16> [[X1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP14:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = xor <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP10]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[TMP16]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP17]], <32 x i16> [[TMP14]]
+; CHECK-NEXT:    [[TMP18:%.*]] = select <32 x i1> [[TMP13]], <32 x i16> [[TMP11]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP18]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psll.w.512(<32 x i16> %x0, <8 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psll.wi.512(<32 x i16>, i32, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_psll_wi_512(<32 x i16> %x0, i32 %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psll_wi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 3)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[X0:%.*]], i32 3)
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP9:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[TMP6]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP5]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i16> [[TMP11]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP12]], <32 x i16> [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP8]], <32 x i16> [[TMP6]], <32 x i16> [[X2]]
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 4)
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[X0]], i32 4)
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i32 [[X3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = select <32 x i1> [[TMP18]], <32 x i16> [[TMP15]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = xor <32 x i16> [[TMP16]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = or <32 x i16> [[TMP20]], [[TMP15]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <32 x i16> [[TMP21]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT1:%.*]] = select <32 x i1> [[TMP17]], <32 x i16> [[TMP22]], <32 x i16> [[TMP19]]
+; CHECK-NEXT:    [[TMP23:%.*]] = select <32 x i1> [[TMP18]], <32 x i16> [[TMP16]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP24:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 5)
+; CHECK-NEXT:    [[TMP25:%.*]] = or <32 x i16> [[TMP24]], zeroinitializer
+; CHECK-NEXT:    [[TMP26:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[X0]], i32 5)
+; CHECK-NEXT:    [[TMP27:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP_SELECT]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP13]], 0
+; CHECK-NEXT:    [[TMP28:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP27]], <32 x i16> [[_MSPROP_SELECT1]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP23]], 1
+; CHECK-NEXT:    [[TMP29:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP28]], <32 x i16> [[TMP25]], 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP26]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP29]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psll.wi.512(<32 x i16> %x0, i32 3, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.psll.wi.512(<32 x i16> %x0, i32 4, <32 x i16> zeroinitializer, i32 %x3)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.psll.wi.512(<32 x i16> %x0, i32 5, <32 x i16> %x2, i32 -1)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pshuf.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pshuf_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshuf_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pshuf_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pshuf_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+
+declare <64 x i8> @llvm.x86.avx512.cvtmask2b.512(i64)
+
+define <64 x i8> @test_int_x86_avx512_cvtmask2b_512(i64 %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_cvtmask2b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i64 [[TMP1]] to <64 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64 [[X0:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = sext <64 x i1> [[TMP2]] to <64 x i8>
+; CHECK-NEXT:    [[VPMOVM2:%.*]] = sext <64 x i1> [[TMP3]] to <64 x i8>
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[VPMOVM2]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.cvtmask2b.512(i64 %x0)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.cvtmask2w.512(i32)
+
+define <32 x i16> @test_int_x86_avx512_cvtmask2w_512(i32 %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_cvtmask2w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i32 [[TMP1]] to <32 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32 [[X0:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[_MSPROP:%.*]] = sext <32 x i1> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    [[VPMOVM2:%.*]] = sext <32 x i1> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[VPMOVM2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.cvtmask2w.512(i32 %x0)
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_mask_packs_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i16> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP9]], <32 x i16> [[TMP14]], <32 x i16> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP8]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP15]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rm_512(<16 x i32> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmbk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmbkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32>, <16 x i32>, <32 x i16>, i32)
+
+define <64 x i8> @test_mask_packs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP7]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <64 x i8> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP9]], <64 x i8> [[TMP14]], <64 x i8> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP8]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP15]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packs_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packs_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <64 x i8> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <64 x i8> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP20]], <64 x i8> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> [[TMP14]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP21]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packs_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <64 x i8> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP14]], <64 x i8> [[TMP19]], <64 x i8> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP13]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP20]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.packsswb.512(<32 x i16>, <32 x i16>, <64 x i8>, i64)
+
+
+define <32 x i16> @test_mask_packus_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i16> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP9]], <32 x i16> [[TMP14]], <32 x i16> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP8]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP15]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rm_512(<16 x i32> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmbk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmbkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32>, <16 x i32>, <32 x i16>, i32)
+
+define <64 x i8> @test_mask_packus_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP7]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packus_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packus_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <64 x i8> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP9]], <64 x i8> [[TMP14]], <64 x i8> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP8]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP15]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packus_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packus_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <64 x i8> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <64 x i8> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP20]], <64 x i8> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> [[TMP14]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP21]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_packus_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <64 x i8> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP14]], <64 x i8> [[TMP19]], <64 x i8> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP13]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP20]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16> %a, <32 x i16> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.packuswb.512(<32 x i16>, <32 x i16>, <64 x i8>, i64)
+
+define i64 @test_cmp_b_512(<64 x i8> %a0, <64 x i8> %a1) nounwind #0 {
+; CHECK-LABEL: @test_cmp_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[TMP4]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <64 x i8> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP]] to i64
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <64 x i1> [[TMP9]] to i64
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i8> [[TMP12]], [[TMP13]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP12]], [[TMP1]]
+; CHECK-NEXT:    [[TMP16:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP18:%.*]] = and <64 x i8> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP16]], [[TMP2]]
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ult <64 x i8> [[TMP14]], [[TMP19]]
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ult <64 x i8> [[TMP15]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = xor <64 x i1> [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = icmp slt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <64 x i1> [[TMP22]] to i64
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <64 x i1> [[TMP23]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP10]], [[TMP24]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i64 [[TMP11]], [[TMP25]]
+; CHECK-NEXT:    [[TMP26:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP27:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP28:%.*]] = and <64 x i8> [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = or <64 x i8> [[TMP26]], [[TMP1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP31:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP32:%.*]] = and <64 x i8> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP33:%.*]] = or <64 x i8> [[TMP30]], [[TMP2]]
+; CHECK-NEXT:    [[TMP34:%.*]] = icmp ule <64 x i8> [[TMP28]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ule <64 x i8> [[TMP29]], [[TMP32]]
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <64 x i1> [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = icmp sle <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <64 x i1> [[TMP36]] to i64
+; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <64 x i1> [[TMP37]] to i64
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], [[TMP38]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i64 [[RET1]], [[TMP39]]
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i64 [[_MSPROP1]], 0
+; CHECK-NEXT:    [[RET3:%.*]] = add i64 [[RET2]], 0
+; CHECK-NEXT:    [[TMP40:%.*]] = xor <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP41:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <64 x i8> [[TMP41]], zeroinitializer
+; CHECK-NEXT:    [[TMP43:%.*]] = xor <64 x i8> [[TMP41]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP44:%.*]] = and <64 x i8> [[TMP43]], [[TMP40]]
+; CHECK-NEXT:    [[TMP45:%.*]] = icmp eq <64 x i8> [[TMP44]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <64 x i1> [[TMP42]], [[TMP45]]
+; CHECK-NEXT:    [[TMP46:%.*]] = icmp ne <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP47:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP3]] to i64
+; CHECK-NEXT:    [[TMP48:%.*]] = bitcast <64 x i1> [[TMP46]] to i64
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i64 [[_MSPROP2]], [[TMP47]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i64 [[RET3]], [[TMP48]]
+; CHECK-NEXT:    [[TMP49:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP50:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP51:%.*]] = and <64 x i8> [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = or <64 x i8> [[TMP49]], [[TMP1]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP54:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP55:%.*]] = and <64 x i8> [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = or <64 x i8> [[TMP53]], [[TMP2]]
+; CHECK-NEXT:    [[TMP57:%.*]] = icmp uge <64 x i8> [[TMP51]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = icmp uge <64 x i8> [[TMP52]], [[TMP55]]
+; CHECK-NEXT:    [[TMP59:%.*]] = xor <64 x i1> [[TMP57]], [[TMP58]]
+; CHECK-NEXT:    [[TMP60:%.*]] = icmp sge <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP61:%.*]] = bitcast <64 x i1> [[TMP59]] to i64
+; CHECK-NEXT:    [[TMP62:%.*]] = bitcast <64 x i1> [[TMP60]] to i64
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i64 [[_MSPROP4]], [[TMP61]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i64 [[RET4]], [[TMP62]]
+; CHECK-NEXT:    [[TMP63:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP64:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP65:%.*]] = and <64 x i8> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP66:%.*]] = or <64 x i8> [[TMP63]], [[TMP1]]
+; CHECK-NEXT:    [[TMP67:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP68:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP69:%.*]] = and <64 x i8> [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = or <64 x i8> [[TMP67]], [[TMP2]]
+; CHECK-NEXT:    [[TMP71:%.*]] = icmp ugt <64 x i8> [[TMP65]], [[TMP70]]
+; CHECK-NEXT:    [[TMP72:%.*]] = icmp ugt <64 x i8> [[TMP66]], [[TMP69]]
+; CHECK-NEXT:    [[TMP73:%.*]] = xor <64 x i1> [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = icmp sgt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP75:%.*]] = bitcast <64 x i1> [[TMP73]] to i64
+; CHECK-NEXT:    [[TMP76:%.*]] = bitcast <64 x i1> [[TMP74]] to i64
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i64 [[_MSPROP5]], [[TMP75]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i64 [[RET5]], [[TMP76]]
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i64 [[_MSPROP6]], 0
+; CHECK-NEXT:    [[RET7:%.*]] = add i64 [[RET6]], -1
+; CHECK-NEXT:    store i64 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RET7]]
+;
+  %res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 -1)
+  %res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 -1)
+  %ret1 = add i64 %res0, %res1
+  %res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 -1)
+  %ret2 = add i64 %ret1, %res2
+  %res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 -1)
+  %ret3 = add i64 %ret2, %res3
+  %res4 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 4, i64 -1)
+  %ret4 = add i64 %ret3, %res4
+  %res5 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 5, i64 -1)
+  %ret5 = add i64 %ret4, %res5
+  %res6 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 6, i64 -1)
+  %ret6 = add i64 %ret5, %res6
+  %res7 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 7, i64 -1)
+  %ret7 = add i64 %ret6, %res7
+  ret i64 %ret7
+}
+
+define i64 @test_mask_cmp_b_512(<64 x i8> %a0, <64 x i8> %a1, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_cmp_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <64 x i8> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <64 x i8> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[TMP5]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <64 x i8> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <64 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <64 x i1> [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <64 x i1> [[TMP18]] to i64
+; CHECK-NEXT:    [[TMP21:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP22:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP23:%.*]] = and <64 x i8> [[TMP21]], [[TMP22]]
+; CHECK-NEXT:    [[TMP24:%.*]] = or <64 x i8> [[TMP21]], [[TMP1]]
+; CHECK-NEXT:    [[TMP25:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP26:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP27:%.*]] = and <64 x i8> [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = or <64 x i8> [[TMP25]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = icmp ult <64 x i8> [[TMP23]], [[TMP28]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ult <64 x i8> [[TMP24]], [[TMP27]]
+; CHECK-NEXT:    [[TMP31:%.*]] = xor <64 x i1> [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = icmp slt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP34:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP35:%.*]] = and <64 x i1> [[TMP31]], [[TMP33]]
+; CHECK-NEXT:    [[TMP36:%.*]] = and <64 x i1> [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP37:%.*]] = and <64 x i1> [[TMP31]], [[TMP34]]
+; CHECK-NEXT:    [[TMP38:%.*]] = or <64 x i1> [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP39:%.*]] = or <64 x i1> [[TMP38]], [[TMP37]]
+; CHECK-NEXT:    [[TMP40:%.*]] = and <64 x i1> [[TMP32]], [[TMP34]]
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <64 x i1> [[TMP39]] to i64
+; CHECK-NEXT:    [[TMP42:%.*]] = bitcast <64 x i1> [[TMP40]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP19]], [[TMP41]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i64 [[TMP20]], [[TMP42]]
+; CHECK-NEXT:    [[TMP43:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP44:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP45:%.*]] = and <64 x i8> [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = or <64 x i8> [[TMP43]], [[TMP1]]
+; CHECK-NEXT:    [[TMP47:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP48:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP49:%.*]] = and <64 x i8> [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = or <64 x i8> [[TMP47]], [[TMP2]]
+; CHECK-NEXT:    [[TMP51:%.*]] = icmp ule <64 x i8> [[TMP45]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = icmp ule <64 x i8> [[TMP46]], [[TMP49]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <64 x i1> [[TMP51]], [[TMP52]]
+; CHECK-NEXT:    [[TMP54:%.*]] = icmp sle <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP55:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP56:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP57:%.*]] = and <64 x i1> [[TMP53]], [[TMP55]]
+; CHECK-NEXT:    [[TMP58:%.*]] = and <64 x i1> [[TMP54]], [[TMP55]]
+; CHECK-NEXT:    [[TMP59:%.*]] = and <64 x i1> [[TMP53]], [[TMP56]]
+; CHECK-NEXT:    [[TMP60:%.*]] = or <64 x i1> [[TMP57]], [[TMP58]]
+; CHECK-NEXT:    [[TMP61:%.*]] = or <64 x i1> [[TMP60]], [[TMP59]]
+; CHECK-NEXT:    [[TMP62:%.*]] = and <64 x i1> [[TMP54]], [[TMP56]]
+; CHECK-NEXT:    [[TMP63:%.*]] = bitcast <64 x i1> [[TMP61]] to i64
+; CHECK-NEXT:    [[TMP64:%.*]] = bitcast <64 x i1> [[TMP62]] to i64
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], [[TMP63]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i64 [[RET1]], [[TMP64]]
+; CHECK-NEXT:    [[TMP65:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP66:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP67:%.*]] = and <64 x i1> zeroinitializer, [[TMP65]]
+; CHECK-NEXT:    [[TMP68:%.*]] = and <64 x i1> zeroinitializer, [[TMP65]]
+; CHECK-NEXT:    [[TMP69:%.*]] = and <64 x i1> zeroinitializer, [[TMP66]]
+; CHECK-NEXT:    [[TMP70:%.*]] = or <64 x i1> [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP71:%.*]] = or <64 x i1> [[TMP70]], [[TMP69]]
+; CHECK-NEXT:    [[TMP72:%.*]] = and <64 x i1> zeroinitializer, [[TMP66]]
+; CHECK-NEXT:    [[TMP73:%.*]] = bitcast <64 x i1> [[TMP71]] to i64
+; CHECK-NEXT:    [[TMP74:%.*]] = bitcast <64 x i1> [[TMP72]] to i64
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i64 [[_MSPROP1]], [[TMP73]]
+; CHECK-NEXT:    [[RET3:%.*]] = add i64 [[RET2]], [[TMP74]]
+; CHECK-NEXT:    [[TMP75:%.*]] = xor <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP76:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP77:%.*]] = icmp ne <64 x i8> [[TMP76]], zeroinitializer
+; CHECK-NEXT:    [[TMP78:%.*]] = xor <64 x i8> [[TMP76]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP79:%.*]] = and <64 x i8> [[TMP78]], [[TMP75]]
+; CHECK-NEXT:    [[TMP80:%.*]] = icmp eq <64 x i8> [[TMP79]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <64 x i1> [[TMP77]], [[TMP80]]
+; CHECK-NEXT:    [[TMP81:%.*]] = icmp ne <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP82:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP83:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP84:%.*]] = and <64 x i1> [[_MSPROP_ICMP3]], [[TMP82]]
+; CHECK-NEXT:    [[TMP85:%.*]] = and <64 x i1> [[TMP81]], [[TMP82]]
+; CHECK-NEXT:    [[TMP86:%.*]] = and <64 x i1> [[_MSPROP_ICMP3]], [[TMP83]]
+; CHECK-NEXT:    [[TMP87:%.*]] = or <64 x i1> [[TMP84]], [[TMP85]]
+; CHECK-NEXT:    [[TMP88:%.*]] = or <64 x i1> [[TMP87]], [[TMP86]]
+; CHECK-NEXT:    [[TMP89:%.*]] = and <64 x i1> [[TMP81]], [[TMP83]]
+; CHECK-NEXT:    [[TMP90:%.*]] = bitcast <64 x i1> [[TMP88]] to i64
+; CHECK-NEXT:    [[TMP91:%.*]] = bitcast <64 x i1> [[TMP89]] to i64
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i64 [[_MSPROP2]], [[TMP90]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i64 [[RET3]], [[TMP91]]
+; CHECK-NEXT:    [[TMP92:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP93:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP94:%.*]] = and <64 x i8> [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = or <64 x i8> [[TMP92]], [[TMP1]]
+; CHECK-NEXT:    [[TMP96:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP97:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP98:%.*]] = and <64 x i8> [[TMP96]], [[TMP97]]
+; CHECK-NEXT:    [[TMP99:%.*]] = or <64 x i8> [[TMP96]], [[TMP2]]
+; CHECK-NEXT:    [[TMP100:%.*]] = icmp uge <64 x i8> [[TMP94]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = icmp uge <64 x i8> [[TMP95]], [[TMP98]]
+; CHECK-NEXT:    [[TMP102:%.*]] = xor <64 x i1> [[TMP100]], [[TMP101]]
+; CHECK-NEXT:    [[TMP103:%.*]] = icmp sge <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP104:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = and <64 x i1> [[TMP102]], [[TMP104]]
+; CHECK-NEXT:    [[TMP107:%.*]] = and <64 x i1> [[TMP103]], [[TMP104]]
+; CHECK-NEXT:    [[TMP108:%.*]] = and <64 x i1> [[TMP102]], [[TMP105]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <64 x i1> [[TMP106]], [[TMP107]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <64 x i1> [[TMP109]], [[TMP108]]
+; CHECK-NEXT:    [[TMP111:%.*]] = and <64 x i1> [[TMP103]], [[TMP105]]
+; CHECK-NEXT:    [[TMP112:%.*]] = bitcast <64 x i1> [[TMP110]] to i64
+; CHECK-NEXT:    [[TMP113:%.*]] = bitcast <64 x i1> [[TMP111]] to i64
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i64 [[_MSPROP4]], [[TMP112]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i64 [[RET4]], [[TMP113]]
+; CHECK-NEXT:    [[TMP114:%.*]] = xor <64 x i8> [[A0]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP115:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP116:%.*]] = and <64 x i8> [[TMP114]], [[TMP115]]
+; CHECK-NEXT:    [[TMP117:%.*]] = or <64 x i8> [[TMP114]], [[TMP1]]
+; CHECK-NEXT:    [[TMP118:%.*]] = xor <64 x i8> [[A1]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP119:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP120:%.*]] = and <64 x i8> [[TMP118]], [[TMP119]]
+; CHECK-NEXT:    [[TMP121:%.*]] = or <64 x i8> [[TMP118]], [[TMP2]]
+; CHECK-NEXT:    [[TMP122:%.*]] = icmp ugt <64 x i8> [[TMP116]], [[TMP121]]
+; CHECK-NEXT:    [[TMP123:%.*]] = icmp ugt <64 x i8> [[TMP117]], [[TMP120]]
+; CHECK-NEXT:    [[TMP124:%.*]] = xor <64 x i1> [[TMP122]], [[TMP123]]
+; CHECK-NEXT:    [[TMP125:%.*]] = icmp sgt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP126:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP127:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP128:%.*]] = and <64 x i1> [[TMP124]], [[TMP126]]
+; CHECK-NEXT:    [[TMP129:%.*]] = and <64 x i1> [[TMP125]], [[TMP126]]
+; CHECK-NEXT:    [[TMP130:%.*]] = and <64 x i1> [[TMP124]], [[TMP127]]
+; CHECK-NEXT:    [[TMP131:%.*]] = or <64 x i1> [[TMP128]], [[TMP129]]
+; CHECK-NEXT:    [[TMP132:%.*]] = or <64 x i1> [[TMP131]], [[TMP130]]
+; CHECK-NEXT:    [[TMP133:%.*]] = and <64 x i1> [[TMP125]], [[TMP127]]
+; CHECK-NEXT:    [[TMP134:%.*]] = bitcast <64 x i1> [[TMP132]] to i64
+; CHECK-NEXT:    [[TMP135:%.*]] = bitcast <64 x i1> [[TMP133]] to i64
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i64 [[_MSPROP5]], [[TMP134]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i64 [[RET5]], [[TMP135]]
+; CHECK-NEXT:    [[TMP136:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP137:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP138:%.*]] = and <64 x i1> zeroinitializer, [[TMP136]]
+; CHECK-NEXT:    [[TMP139:%.*]] = and <64 x i1> splat (i1 true), [[TMP136]]
+; CHECK-NEXT:    [[TMP140:%.*]] = and <64 x i1> zeroinitializer, [[TMP137]]
+; CHECK-NEXT:    [[TMP141:%.*]] = or <64 x i1> [[TMP138]], [[TMP139]]
+; CHECK-NEXT:    [[TMP142:%.*]] = or <64 x i1> [[TMP141]], [[TMP140]]
+; CHECK-NEXT:    [[TMP143:%.*]] = and <64 x i1> splat (i1 true), [[TMP137]]
+; CHECK-NEXT:    [[TMP144:%.*]] = bitcast <64 x i1> [[TMP142]] to i64
+; CHECK-NEXT:    [[TMP145:%.*]] = bitcast <64 x i1> [[TMP143]] to i64
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i64 [[_MSPROP6]], [[TMP144]]
+; CHECK-NEXT:    [[RET7:%.*]] = add i64 [[RET6]], [[TMP145]]
+; CHECK-NEXT:    store i64 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RET7]]
+;
+  %res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask)
+  %res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask)
+  %ret1 = add i64 %res0, %res1
+  %res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask)
+  %ret2 = add i64 %ret1, %res2
+  %res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask)
+  %ret3 = add i64 %ret2, %res3
+  %res4 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 4, i64 %mask)
+  %ret4 = add i64 %ret3, %res4
+  %res5 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 5, i64 %mask)
+  %ret5 = add i64 %ret4, %res5
+  %res6 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 6, i64 %mask)
+  %ret6 = add i64 %ret5, %res6
+  %res7 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 7, i64 %mask)
+  %ret7 = add i64 %ret6, %res7
+  ret i64 %ret7
+}
+
+declare i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8>, <64 x i8>, i32, i64) nounwind readnone
+
+define i64 @test_ucmp_b_512(<64 x i8> %a0, <64 x i8> %a1) nounwind #0 {
+; CHECK-LABEL: @test_ucmp_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[TMP4]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <64 x i8> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP]] to i64
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <64 x i1> [[TMP9]] to i64
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i8> [[A0]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP15:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP16:%.*]] = and <64 x i8> [[A1]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ult <64 x i8> [[TMP13]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ult <64 x i8> [[TMP14]], [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = xor <64 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ult <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <64 x i1> [[TMP20]] to i64
+; CHECK-NEXT:    [[TMP23:%.*]] = bitcast <64 x i1> [[TMP21]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP10]], [[TMP22]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i64 [[TMP11]], [[TMP23]]
+; CHECK-NEXT:    [[TMP24:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP25:%.*]] = and <64 x i8> [[A0]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP27:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP28:%.*]] = and <64 x i8> [[A1]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ule <64 x i8> [[TMP25]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = icmp ule <64 x i8> [[TMP26]], [[TMP28]]
+; CHECK-NEXT:    [[TMP32:%.*]] = xor <64 x i1> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP33:%.*]] = icmp ule <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP34:%.*]] = bitcast <64 x i1> [[TMP32]] to i64
+; CHECK-NEXT:    [[TMP35:%.*]] = bitcast <64 x i1> [[TMP33]] to i64
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], [[TMP34]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i64 [[RET1]], [[TMP35]]
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i64 [[_MSPROP1]], 0
+; CHECK-NEXT:    [[RET3:%.*]] = add i64 [[RET2]], 0
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP37:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp ne <64 x i8> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[TMP39:%.*]] = xor <64 x i8> [[TMP37]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP40:%.*]] = and <64 x i8> [[TMP39]], [[TMP36]]
+; CHECK-NEXT:    [[TMP41:%.*]] = icmp eq <64 x i8> [[TMP40]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <64 x i1> [[TMP38]], [[TMP41]]
+; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP43:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP3]] to i64
+; CHECK-NEXT:    [[TMP44:%.*]] = bitcast <64 x i1> [[TMP42]] to i64
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i64 [[_MSPROP2]], [[TMP43]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i64 [[RET3]], [[TMP44]]
+; CHECK-NEXT:    [[TMP45:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP46:%.*]] = and <64 x i8> [[A0]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP48:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP49:%.*]] = and <64 x i8> [[A1]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP51:%.*]] = icmp uge <64 x i8> [[TMP46]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = icmp uge <64 x i8> [[TMP47]], [[TMP49]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <64 x i1> [[TMP51]], [[TMP52]]
+; CHECK-NEXT:    [[TMP54:%.*]] = icmp uge <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP55:%.*]] = bitcast <64 x i1> [[TMP53]] to i64
+; CHECK-NEXT:    [[TMP56:%.*]] = bitcast <64 x i1> [[TMP54]] to i64
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i64 [[_MSPROP4]], [[TMP55]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i64 [[RET4]], [[TMP56]]
+; CHECK-NEXT:    [[TMP57:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP58:%.*]] = and <64 x i8> [[A0]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP60:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP61:%.*]] = and <64 x i8> [[A1]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP63:%.*]] = icmp ugt <64 x i8> [[TMP58]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = icmp ugt <64 x i8> [[TMP59]], [[TMP61]]
+; CHECK-NEXT:    [[TMP65:%.*]] = xor <64 x i1> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP66:%.*]] = icmp ugt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP67:%.*]] = bitcast <64 x i1> [[TMP65]] to i64
+; CHECK-NEXT:    [[TMP68:%.*]] = bitcast <64 x i1> [[TMP66]] to i64
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i64 [[_MSPROP5]], [[TMP67]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i64 [[RET5]], [[TMP68]]
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i64 [[_MSPROP6]], 0
+; CHECK-NEXT:    [[RET7:%.*]] = add i64 [[RET6]], -1
+; CHECK-NEXT:    store i64 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RET7]]
+;
+  %res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 -1)
+  %res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 -1)
+  %ret1 = add i64 %res0, %res1
+  %res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 -1)
+  %ret2 = add i64 %ret1, %res2
+  %res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 -1)
+  %ret3 = add i64 %ret2, %res3
+  %res4 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 4, i64 -1)
+  %ret4 = add i64 %ret3, %res4
+  %res5 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 5, i64 -1)
+  %ret5 = add i64 %ret4, %res5
+  %res6 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 6, i64 -1)
+  %ret6 = add i64 %ret5, %res6
+  %res7 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 7, i64 -1)
+  %ret7 = add i64 %ret6, %res7
+  ret i64 %ret7
+}
+
+define i64 @test_mask_x86_avx512_ucmp_b_512(<64 x i8> %a0, <64 x i8> %a1, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_x86_avx512_ucmp_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <64 x i8> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <64 x i8> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[TMP5]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <64 x i8> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <64 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <64 x i1> [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <64 x i1> [[TMP18]] to i64
+; CHECK-NEXT:    [[TMP21:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP22:%.*]] = and <64 x i8> [[A0]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP24:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP25:%.*]] = and <64 x i8> [[A1]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP27:%.*]] = icmp ult <64 x i8> [[TMP22]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = icmp ult <64 x i8> [[TMP23]], [[TMP25]]
+; CHECK-NEXT:    [[TMP29:%.*]] = xor <64 x i1> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ult <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP31:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP32:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP33:%.*]] = and <64 x i1> [[TMP29]], [[TMP31]]
+; CHECK-NEXT:    [[TMP34:%.*]] = and <64 x i1> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP35:%.*]] = and <64 x i1> [[TMP29]], [[TMP32]]
+; CHECK-NEXT:    [[TMP36:%.*]] = or <64 x i1> [[TMP33]], [[TMP34]]
+; CHECK-NEXT:    [[TMP37:%.*]] = or <64 x i1> [[TMP36]], [[TMP35]]
+; CHECK-NEXT:    [[TMP38:%.*]] = and <64 x i1> [[TMP30]], [[TMP32]]
+; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <64 x i1> [[TMP37]] to i64
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <64 x i1> [[TMP38]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP19]], [[TMP39]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i64 [[TMP20]], [[TMP40]]
+; CHECK-NEXT:    [[TMP41:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP42:%.*]] = and <64 x i8> [[A0]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP44:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP45:%.*]] = and <64 x i8> [[A1]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP47:%.*]] = icmp ule <64 x i8> [[TMP42]], [[TMP46]]
+; CHECK-NEXT:    [[TMP48:%.*]] = icmp ule <64 x i8> [[TMP43]], [[TMP45]]
+; CHECK-NEXT:    [[TMP49:%.*]] = xor <64 x i1> [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = icmp ule <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP51:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP52:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP53:%.*]] = and <64 x i1> [[TMP49]], [[TMP51]]
+; CHECK-NEXT:    [[TMP54:%.*]] = and <64 x i1> [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP55:%.*]] = and <64 x i1> [[TMP49]], [[TMP52]]
+; CHECK-NEXT:    [[TMP56:%.*]] = or <64 x i1> [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP57:%.*]] = or <64 x i1> [[TMP56]], [[TMP55]]
+; CHECK-NEXT:    [[TMP58:%.*]] = and <64 x i1> [[TMP50]], [[TMP52]]
+; CHECK-NEXT:    [[TMP59:%.*]] = bitcast <64 x i1> [[TMP57]] to i64
+; CHECK-NEXT:    [[TMP60:%.*]] = bitcast <64 x i1> [[TMP58]] to i64
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i64 [[_MSPROP]], [[TMP59]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i64 [[RET1]], [[TMP60]]
+; CHECK-NEXT:    [[TMP61:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP62:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP63:%.*]] = and <64 x i1> zeroinitializer, [[TMP61]]
+; CHECK-NEXT:    [[TMP64:%.*]] = and <64 x i1> zeroinitializer, [[TMP61]]
+; CHECK-NEXT:    [[TMP65:%.*]] = and <64 x i1> zeroinitializer, [[TMP62]]
+; CHECK-NEXT:    [[TMP66:%.*]] = or <64 x i1> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP67:%.*]] = or <64 x i1> [[TMP66]], [[TMP65]]
+; CHECK-NEXT:    [[TMP68:%.*]] = and <64 x i1> zeroinitializer, [[TMP62]]
+; CHECK-NEXT:    [[TMP69:%.*]] = bitcast <64 x i1> [[TMP67]] to i64
+; CHECK-NEXT:    [[TMP70:%.*]] = bitcast <64 x i1> [[TMP68]] to i64
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i64 [[_MSPROP1]], [[TMP69]]
+; CHECK-NEXT:    [[RET3:%.*]] = add i64 [[RET2]], [[TMP70]]
+; CHECK-NEXT:    [[TMP71:%.*]] = xor <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP72:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP73:%.*]] = icmp ne <64 x i8> [[TMP72]], zeroinitializer
+; CHECK-NEXT:    [[TMP74:%.*]] = xor <64 x i8> [[TMP72]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP75:%.*]] = and <64 x i8> [[TMP74]], [[TMP71]]
+; CHECK-NEXT:    [[TMP76:%.*]] = icmp eq <64 x i8> [[TMP75]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <64 x i1> [[TMP73]], [[TMP76]]
+; CHECK-NEXT:    [[TMP77:%.*]] = icmp ne <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP78:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP79:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP80:%.*]] = and <64 x i1> [[_MSPROP_ICMP3]], [[TMP78]]
+; CHECK-NEXT:    [[TMP81:%.*]] = and <64 x i1> [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP82:%.*]] = and <64 x i1> [[_MSPROP_ICMP3]], [[TMP79]]
+; CHECK-NEXT:    [[TMP83:%.*]] = or <64 x i1> [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP84:%.*]] = or <64 x i1> [[TMP83]], [[TMP82]]
+; CHECK-NEXT:    [[TMP85:%.*]] = and <64 x i1> [[TMP77]], [[TMP79]]
+; CHECK-NEXT:    [[TMP86:%.*]] = bitcast <64 x i1> [[TMP84]] to i64
+; CHECK-NEXT:    [[TMP87:%.*]] = bitcast <64 x i1> [[TMP85]] to i64
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i64 [[_MSPROP2]], [[TMP86]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i64 [[RET3]], [[TMP87]]
+; CHECK-NEXT:    [[TMP88:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP89:%.*]] = and <64 x i8> [[A0]], [[TMP88]]
+; CHECK-NEXT:    [[TMP90:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP91:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP92:%.*]] = and <64 x i8> [[A1]], [[TMP91]]
+; CHECK-NEXT:    [[TMP93:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP94:%.*]] = icmp uge <64 x i8> [[TMP89]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = icmp uge <64 x i8> [[TMP90]], [[TMP92]]
+; CHECK-NEXT:    [[TMP96:%.*]] = xor <64 x i1> [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = icmp uge <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP98:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP99:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP100:%.*]] = and <64 x i1> [[TMP96]], [[TMP98]]
+; CHECK-NEXT:    [[TMP101:%.*]] = and <64 x i1> [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP102:%.*]] = and <64 x i1> [[TMP96]], [[TMP99]]
+; CHECK-NEXT:    [[TMP103:%.*]] = or <64 x i1> [[TMP100]], [[TMP101]]
+; CHECK-NEXT:    [[TMP104:%.*]] = or <64 x i1> [[TMP103]], [[TMP102]]
+; CHECK-NEXT:    [[TMP105:%.*]] = and <64 x i1> [[TMP97]], [[TMP99]]
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast <64 x i1> [[TMP104]] to i64
+; CHECK-NEXT:    [[TMP107:%.*]] = bitcast <64 x i1> [[TMP105]] to i64
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i64 [[_MSPROP4]], [[TMP106]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i64 [[RET4]], [[TMP107]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP109:%.*]] = and <64 x i8> [[A0]], [[TMP108]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <64 x i8> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP111:%.*]] = xor <64 x i8> [[TMP2]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP112:%.*]] = and <64 x i8> [[A1]], [[TMP111]]
+; CHECK-NEXT:    [[TMP113:%.*]] = or <64 x i8> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP114:%.*]] = icmp ugt <64 x i8> [[TMP109]], [[TMP113]]
+; CHECK-NEXT:    [[TMP115:%.*]] = icmp ugt <64 x i8> [[TMP110]], [[TMP112]]
+; CHECK-NEXT:    [[TMP116:%.*]] = xor <64 x i1> [[TMP114]], [[TMP115]]
+; CHECK-NEXT:    [[TMP117:%.*]] = icmp ugt <64 x i8> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP118:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP119:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP120:%.*]] = and <64 x i1> [[TMP116]], [[TMP118]]
+; CHECK-NEXT:    [[TMP121:%.*]] = and <64 x i1> [[TMP117]], [[TMP118]]
+; CHECK-NEXT:    [[TMP122:%.*]] = and <64 x i1> [[TMP116]], [[TMP119]]
+; CHECK-NEXT:    [[TMP123:%.*]] = or <64 x i1> [[TMP120]], [[TMP121]]
+; CHECK-NEXT:    [[TMP124:%.*]] = or <64 x i1> [[TMP123]], [[TMP122]]
+; CHECK-NEXT:    [[TMP125:%.*]] = and <64 x i1> [[TMP117]], [[TMP119]]
+; CHECK-NEXT:    [[TMP126:%.*]] = bitcast <64 x i1> [[TMP124]] to i64
+; CHECK-NEXT:    [[TMP127:%.*]] = bitcast <64 x i1> [[TMP125]] to i64
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i64 [[_MSPROP5]], [[TMP126]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i64 [[RET5]], [[TMP127]]
+; CHECK-NEXT:    [[TMP128:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP129:%.*]] = bitcast i64 [[MASK]] to <64 x i1>
+; CHECK-NEXT:    [[TMP130:%.*]] = and <64 x i1> zeroinitializer, [[TMP128]]
+; CHECK-NEXT:    [[TMP131:%.*]] = and <64 x i1> splat (i1 true), [[TMP128]]
+; CHECK-NEXT:    [[TMP132:%.*]] = and <64 x i1> zeroinitializer, [[TMP129]]
+; CHECK-NEXT:    [[TMP133:%.*]] = or <64 x i1> [[TMP130]], [[TMP131]]
+; CHECK-NEXT:    [[TMP134:%.*]] = or <64 x i1> [[TMP133]], [[TMP132]]
+; CHECK-NEXT:    [[TMP135:%.*]] = and <64 x i1> splat (i1 true), [[TMP129]]
+; CHECK-NEXT:    [[TMP136:%.*]] = bitcast <64 x i1> [[TMP134]] to i64
+; CHECK-NEXT:    [[TMP137:%.*]] = bitcast <64 x i1> [[TMP135]] to i64
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i64 [[_MSPROP6]], [[TMP136]]
+; CHECK-NEXT:    [[RET7:%.*]] = add i64 [[RET6]], [[TMP137]]
+; CHECK-NEXT:    store i64 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RET7]]
+;
+  %res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask)
+  %res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask)
+  %ret1 = add i64 %res0, %res1
+  %res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask)
+  %ret2 = add i64 %ret1, %res2
+  %res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask)
+  %ret3 = add i64 %ret2, %res3
+  %res4 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 4, i64 %mask)
+  %ret4 = add i64 %ret3, %res4
+  %res5 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 5, i64 %mask)
+  %ret5 = add i64 %ret4, %res5
+  %res6 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 6, i64 %mask)
+  %ret6 = add i64 %ret5, %res6
+  %res7 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 7, i64 %mask)
+  %ret7 = add i64 %ret6, %res7
+  ret i64 %ret7
+}
+
+declare i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8>, <64 x i8>, i32, i64) nounwind readnone
+
+define i32 @test_cmp_w_512(<32 x i16> %a0, <32 x i16> %a1) nounwind #0 {
+; CHECK-LABEL: @test_cmp_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <32 x i16> [[TMP4]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <32 x i16> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <32 x i16> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP]] to i32
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i1> [[TMP9]] to i32
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i16> [[TMP12]], [[TMP13]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP12]], [[TMP1]]
+; CHECK-NEXT:    [[TMP16:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP18:%.*]] = and <32 x i16> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP16]], [[TMP2]]
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ult <32 x i16> [[TMP14]], [[TMP19]]
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ult <32 x i16> [[TMP15]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = xor <32 x i1> [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = icmp slt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP24:%.*]] = bitcast <32 x i1> [[TMP22]] to i32
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <32 x i1> [[TMP23]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP10]], [[TMP24]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i32 [[TMP11]], [[TMP25]]
+; CHECK-NEXT:    [[TMP26:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP27:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP28:%.*]] = and <32 x i16> [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = or <32 x i16> [[TMP26]], [[TMP1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP31:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP32:%.*]] = and <32 x i16> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP33:%.*]] = or <32 x i16> [[TMP30]], [[TMP2]]
+; CHECK-NEXT:    [[TMP34:%.*]] = icmp ule <32 x i16> [[TMP28]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ule <32 x i16> [[TMP29]], [[TMP32]]
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <32 x i1> [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = icmp sle <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <32 x i1> [[TMP36]] to i32
+; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <32 x i1> [[TMP37]] to i32
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i32 [[_MSPROP]], [[TMP38]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i32 [[RET1]], [[TMP39]]
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i32 [[_MSPROP1]], 0
+; CHECK-NEXT:    [[RET3:%.*]] = add i32 [[RET2]], 0
+; CHECK-NEXT:    [[TMP40:%.*]] = xor <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP41:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP41]], zeroinitializer
+; CHECK-NEXT:    [[TMP43:%.*]] = xor <32 x i16> [[TMP41]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP44:%.*]] = and <32 x i16> [[TMP43]], [[TMP40]]
+; CHECK-NEXT:    [[TMP45:%.*]] = icmp eq <32 x i16> [[TMP44]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <32 x i1> [[TMP42]], [[TMP45]]
+; CHECK-NEXT:    [[TMP46:%.*]] = icmp ne <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP47:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP3]] to i32
+; CHECK-NEXT:    [[TMP48:%.*]] = bitcast <32 x i1> [[TMP46]] to i32
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i32 [[_MSPROP2]], [[TMP47]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i32 [[RET3]], [[TMP48]]
+; CHECK-NEXT:    [[TMP49:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP50:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP51:%.*]] = and <32 x i16> [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = or <32 x i16> [[TMP49]], [[TMP1]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP54:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP55:%.*]] = and <32 x i16> [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = or <32 x i16> [[TMP53]], [[TMP2]]
+; CHECK-NEXT:    [[TMP57:%.*]] = icmp uge <32 x i16> [[TMP51]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = icmp uge <32 x i16> [[TMP52]], [[TMP55]]
+; CHECK-NEXT:    [[TMP59:%.*]] = xor <32 x i1> [[TMP57]], [[TMP58]]
+; CHECK-NEXT:    [[TMP60:%.*]] = icmp sge <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP61:%.*]] = bitcast <32 x i1> [[TMP59]] to i32
+; CHECK-NEXT:    [[TMP62:%.*]] = bitcast <32 x i1> [[TMP60]] to i32
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i32 [[_MSPROP4]], [[TMP61]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i32 [[RET4]], [[TMP62]]
+; CHECK-NEXT:    [[TMP63:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP64:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP65:%.*]] = and <32 x i16> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP66:%.*]] = or <32 x i16> [[TMP63]], [[TMP1]]
+; CHECK-NEXT:    [[TMP67:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP68:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP69:%.*]] = and <32 x i16> [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = or <32 x i16> [[TMP67]], [[TMP2]]
+; CHECK-NEXT:    [[TMP71:%.*]] = icmp ugt <32 x i16> [[TMP65]], [[TMP70]]
+; CHECK-NEXT:    [[TMP72:%.*]] = icmp ugt <32 x i16> [[TMP66]], [[TMP69]]
+; CHECK-NEXT:    [[TMP73:%.*]] = xor <32 x i1> [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = icmp sgt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP75:%.*]] = bitcast <32 x i1> [[TMP73]] to i32
+; CHECK-NEXT:    [[TMP76:%.*]] = bitcast <32 x i1> [[TMP74]] to i32
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i32 [[_MSPROP5]], [[TMP75]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i32 [[RET5]], [[TMP76]]
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i32 [[_MSPROP6]], 0
+; CHECK-NEXT:    [[RET7:%.*]] = add i32 [[RET6]], -1
+; CHECK-NEXT:    store i32 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RET7]]
+;
+  %res0 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 0, i32 -1)
+  %res1 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 1, i32 -1)
+  %ret1 = add i32 %res0, %res1
+  %res2 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 2, i32 -1)
+  %ret2 = add i32 %ret1, %res2
+  %res3 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 3, i32 -1)
+  %ret3 = add i32 %ret2, %res3
+  %res4 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 4, i32 -1)
+  %ret4 = add i32 %ret3, %res4
+  %res5 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 5, i32 -1)
+  %ret5 = add i32 %ret4, %res5
+  %res6 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 6, i32 -1)
+  %ret6 = add i32 %ret5, %res6
+  %res7 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 7, i32 -1)
+  %ret7 = add i32 %ret6, %res7
+  ret i32 %ret7
+}
+
+define i32 @test_mask_cmp_w_512(<32 x i16> %a0, <32 x i16> %a1, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_cmp_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <32 x i16> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <32 x i16> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[TMP5]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <32 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i1> [[TMP17]] to i32
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <32 x i1> [[TMP18]] to i32
+; CHECK-NEXT:    [[TMP21:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP22:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP23:%.*]] = and <32 x i16> [[TMP21]], [[TMP22]]
+; CHECK-NEXT:    [[TMP24:%.*]] = or <32 x i16> [[TMP21]], [[TMP1]]
+; CHECK-NEXT:    [[TMP25:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP26:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP27:%.*]] = and <32 x i16> [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = or <32 x i16> [[TMP25]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = icmp ult <32 x i16> [[TMP23]], [[TMP28]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ult <32 x i16> [[TMP24]], [[TMP27]]
+; CHECK-NEXT:    [[TMP31:%.*]] = xor <32 x i1> [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = icmp slt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP34:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP35:%.*]] = and <32 x i1> [[TMP31]], [[TMP33]]
+; CHECK-NEXT:    [[TMP36:%.*]] = and <32 x i1> [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP37:%.*]] = and <32 x i1> [[TMP31]], [[TMP34]]
+; CHECK-NEXT:    [[TMP38:%.*]] = or <32 x i1> [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP39:%.*]] = or <32 x i1> [[TMP38]], [[TMP37]]
+; CHECK-NEXT:    [[TMP40:%.*]] = and <32 x i1> [[TMP32]], [[TMP34]]
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <32 x i1> [[TMP39]] to i32
+; CHECK-NEXT:    [[TMP42:%.*]] = bitcast <32 x i1> [[TMP40]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP19]], [[TMP41]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i32 [[TMP20]], [[TMP42]]
+; CHECK-NEXT:    [[TMP43:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP44:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP45:%.*]] = and <32 x i16> [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = or <32 x i16> [[TMP43]], [[TMP1]]
+; CHECK-NEXT:    [[TMP47:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP48:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP49:%.*]] = and <32 x i16> [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = or <32 x i16> [[TMP47]], [[TMP2]]
+; CHECK-NEXT:    [[TMP51:%.*]] = icmp ule <32 x i16> [[TMP45]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = icmp ule <32 x i16> [[TMP46]], [[TMP49]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <32 x i1> [[TMP51]], [[TMP52]]
+; CHECK-NEXT:    [[TMP54:%.*]] = icmp sle <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP55:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP56:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP57:%.*]] = and <32 x i1> [[TMP53]], [[TMP55]]
+; CHECK-NEXT:    [[TMP58:%.*]] = and <32 x i1> [[TMP54]], [[TMP55]]
+; CHECK-NEXT:    [[TMP59:%.*]] = and <32 x i1> [[TMP53]], [[TMP56]]
+; CHECK-NEXT:    [[TMP60:%.*]] = or <32 x i1> [[TMP57]], [[TMP58]]
+; CHECK-NEXT:    [[TMP61:%.*]] = or <32 x i1> [[TMP60]], [[TMP59]]
+; CHECK-NEXT:    [[TMP62:%.*]] = and <32 x i1> [[TMP54]], [[TMP56]]
+; CHECK-NEXT:    [[TMP63:%.*]] = bitcast <32 x i1> [[TMP61]] to i32
+; CHECK-NEXT:    [[TMP64:%.*]] = bitcast <32 x i1> [[TMP62]] to i32
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i32 [[_MSPROP]], [[TMP63]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i32 [[RET1]], [[TMP64]]
+; CHECK-NEXT:    [[TMP65:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP66:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP67:%.*]] = and <32 x i1> zeroinitializer, [[TMP65]]
+; CHECK-NEXT:    [[TMP68:%.*]] = and <32 x i1> zeroinitializer, [[TMP65]]
+; CHECK-NEXT:    [[TMP69:%.*]] = and <32 x i1> zeroinitializer, [[TMP66]]
+; CHECK-NEXT:    [[TMP70:%.*]] = or <32 x i1> [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP71:%.*]] = or <32 x i1> [[TMP70]], [[TMP69]]
+; CHECK-NEXT:    [[TMP72:%.*]] = and <32 x i1> zeroinitializer, [[TMP66]]
+; CHECK-NEXT:    [[TMP73:%.*]] = bitcast <32 x i1> [[TMP71]] to i32
+; CHECK-NEXT:    [[TMP74:%.*]] = bitcast <32 x i1> [[TMP72]] to i32
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i32 [[_MSPROP1]], [[TMP73]]
+; CHECK-NEXT:    [[RET3:%.*]] = add i32 [[RET2]], [[TMP74]]
+; CHECK-NEXT:    [[TMP75:%.*]] = xor <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP76:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP77:%.*]] = icmp ne <32 x i16> [[TMP76]], zeroinitializer
+; CHECK-NEXT:    [[TMP78:%.*]] = xor <32 x i16> [[TMP76]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP79:%.*]] = and <32 x i16> [[TMP78]], [[TMP75]]
+; CHECK-NEXT:    [[TMP80:%.*]] = icmp eq <32 x i16> [[TMP79]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <32 x i1> [[TMP77]], [[TMP80]]
+; CHECK-NEXT:    [[TMP81:%.*]] = icmp ne <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP82:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP83:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP84:%.*]] = and <32 x i1> [[_MSPROP_ICMP3]], [[TMP82]]
+; CHECK-NEXT:    [[TMP85:%.*]] = and <32 x i1> [[TMP81]], [[TMP82]]
+; CHECK-NEXT:    [[TMP86:%.*]] = and <32 x i1> [[_MSPROP_ICMP3]], [[TMP83]]
+; CHECK-NEXT:    [[TMP87:%.*]] = or <32 x i1> [[TMP84]], [[TMP85]]
+; CHECK-NEXT:    [[TMP88:%.*]] = or <32 x i1> [[TMP87]], [[TMP86]]
+; CHECK-NEXT:    [[TMP89:%.*]] = and <32 x i1> [[TMP81]], [[TMP83]]
+; CHECK-NEXT:    [[TMP90:%.*]] = bitcast <32 x i1> [[TMP88]] to i32
+; CHECK-NEXT:    [[TMP91:%.*]] = bitcast <32 x i1> [[TMP89]] to i32
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i32 [[_MSPROP2]], [[TMP90]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i32 [[RET3]], [[TMP91]]
+; CHECK-NEXT:    [[TMP92:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP93:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP94:%.*]] = and <32 x i16> [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = or <32 x i16> [[TMP92]], [[TMP1]]
+; CHECK-NEXT:    [[TMP96:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP97:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP98:%.*]] = and <32 x i16> [[TMP96]], [[TMP97]]
+; CHECK-NEXT:    [[TMP99:%.*]] = or <32 x i16> [[TMP96]], [[TMP2]]
+; CHECK-NEXT:    [[TMP100:%.*]] = icmp uge <32 x i16> [[TMP94]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = icmp uge <32 x i16> [[TMP95]], [[TMP98]]
+; CHECK-NEXT:    [[TMP102:%.*]] = xor <32 x i1> [[TMP100]], [[TMP101]]
+; CHECK-NEXT:    [[TMP103:%.*]] = icmp sge <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP104:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = and <32 x i1> [[TMP102]], [[TMP104]]
+; CHECK-NEXT:    [[TMP107:%.*]] = and <32 x i1> [[TMP103]], [[TMP104]]
+; CHECK-NEXT:    [[TMP108:%.*]] = and <32 x i1> [[TMP102]], [[TMP105]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i1> [[TMP106]], [[TMP107]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i1> [[TMP109]], [[TMP108]]
+; CHECK-NEXT:    [[TMP111:%.*]] = and <32 x i1> [[TMP103]], [[TMP105]]
+; CHECK-NEXT:    [[TMP112:%.*]] = bitcast <32 x i1> [[TMP110]] to i32
+; CHECK-NEXT:    [[TMP113:%.*]] = bitcast <32 x i1> [[TMP111]] to i32
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i32 [[_MSPROP4]], [[TMP112]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i32 [[RET4]], [[TMP113]]
+; CHECK-NEXT:    [[TMP114:%.*]] = xor <32 x i16> [[A0]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP115:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP116:%.*]] = and <32 x i16> [[TMP114]], [[TMP115]]
+; CHECK-NEXT:    [[TMP117:%.*]] = or <32 x i16> [[TMP114]], [[TMP1]]
+; CHECK-NEXT:    [[TMP118:%.*]] = xor <32 x i16> [[A1]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP119:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP120:%.*]] = and <32 x i16> [[TMP118]], [[TMP119]]
+; CHECK-NEXT:    [[TMP121:%.*]] = or <32 x i16> [[TMP118]], [[TMP2]]
+; CHECK-NEXT:    [[TMP122:%.*]] = icmp ugt <32 x i16> [[TMP116]], [[TMP121]]
+; CHECK-NEXT:    [[TMP123:%.*]] = icmp ugt <32 x i16> [[TMP117]], [[TMP120]]
+; CHECK-NEXT:    [[TMP124:%.*]] = xor <32 x i1> [[TMP122]], [[TMP123]]
+; CHECK-NEXT:    [[TMP125:%.*]] = icmp sgt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP126:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP127:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP128:%.*]] = and <32 x i1> [[TMP124]], [[TMP126]]
+; CHECK-NEXT:    [[TMP129:%.*]] = and <32 x i1> [[TMP125]], [[TMP126]]
+; CHECK-NEXT:    [[TMP130:%.*]] = and <32 x i1> [[TMP124]], [[TMP127]]
+; CHECK-NEXT:    [[TMP131:%.*]] = or <32 x i1> [[TMP128]], [[TMP129]]
+; CHECK-NEXT:    [[TMP132:%.*]] = or <32 x i1> [[TMP131]], [[TMP130]]
+; CHECK-NEXT:    [[TMP133:%.*]] = and <32 x i1> [[TMP125]], [[TMP127]]
+; CHECK-NEXT:    [[TMP134:%.*]] = bitcast <32 x i1> [[TMP132]] to i32
+; CHECK-NEXT:    [[TMP135:%.*]] = bitcast <32 x i1> [[TMP133]] to i32
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i32 [[_MSPROP5]], [[TMP134]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i32 [[RET5]], [[TMP135]]
+; CHECK-NEXT:    [[TMP136:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP137:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP138:%.*]] = and <32 x i1> zeroinitializer, [[TMP136]]
+; CHECK-NEXT:    [[TMP139:%.*]] = and <32 x i1> splat (i1 true), [[TMP136]]
+; CHECK-NEXT:    [[TMP140:%.*]] = and <32 x i1> zeroinitializer, [[TMP137]]
+; CHECK-NEXT:    [[TMP141:%.*]] = or <32 x i1> [[TMP138]], [[TMP139]]
+; CHECK-NEXT:    [[TMP142:%.*]] = or <32 x i1> [[TMP141]], [[TMP140]]
+; CHECK-NEXT:    [[TMP143:%.*]] = and <32 x i1> splat (i1 true), [[TMP137]]
+; CHECK-NEXT:    [[TMP144:%.*]] = bitcast <32 x i1> [[TMP142]] to i32
+; CHECK-NEXT:    [[TMP145:%.*]] = bitcast <32 x i1> [[TMP143]] to i32
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i32 [[_MSPROP6]], [[TMP144]]
+; CHECK-NEXT:    [[RET7:%.*]] = add i32 [[RET6]], [[TMP145]]
+; CHECK-NEXT:    store i32 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RET7]]
+;
+  %res0 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 0, i32 %mask)
+  %res1 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 1, i32 %mask)
+  %ret1 = add i32 %res0, %res1
+  %res2 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 2, i32 %mask)
+  %ret2 = add i32 %ret1, %res2
+  %res3 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 3, i32 %mask)
+  %ret3 = add i32 %ret2, %res3
+  %res4 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 4, i32 %mask)
+  %ret4 = add i32 %ret3, %res4
+  %res5 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 5, i32 %mask)
+  %ret5 = add i32 %ret4, %res5
+  %res6 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 6, i32 %mask)
+  %ret6 = add i32 %ret5, %res6
+  %res7 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 7, i32 %mask)
+  %ret7 = add i32 %ret6, %res7
+  ret i32 %ret7
+}
+
+declare i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16>, <32 x i16>, i32, i32) nounwind readnone
+
+define i32 @test_ucmp_w_512(<32 x i16> %a0, <32 x i16> %a1) nounwind #0 {
+; CHECK-LABEL: @test_ucmp_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <32 x i16> [[TMP4]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <32 x i16> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <32 x i16> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP]] to i32
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i1> [[TMP9]] to i32
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i16> [[A0]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP15:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP16:%.*]] = and <32 x i16> [[A1]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ult <32 x i16> [[TMP13]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ult <32 x i16> [[TMP14]], [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = xor <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ult <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <32 x i1> [[TMP20]] to i32
+; CHECK-NEXT:    [[TMP23:%.*]] = bitcast <32 x i1> [[TMP21]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP10]], [[TMP22]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i32 [[TMP11]], [[TMP23]]
+; CHECK-NEXT:    [[TMP24:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP25:%.*]] = and <32 x i16> [[A0]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP27:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP28:%.*]] = and <32 x i16> [[A1]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ule <32 x i16> [[TMP25]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = icmp ule <32 x i16> [[TMP26]], [[TMP28]]
+; CHECK-NEXT:    [[TMP32:%.*]] = xor <32 x i1> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP33:%.*]] = icmp ule <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP34:%.*]] = bitcast <32 x i1> [[TMP32]] to i32
+; CHECK-NEXT:    [[TMP35:%.*]] = bitcast <32 x i1> [[TMP33]] to i32
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i32 [[_MSPROP]], [[TMP34]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i32 [[RET1]], [[TMP35]]
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i32 [[_MSPROP1]], 0
+; CHECK-NEXT:    [[RET3:%.*]] = add i32 [[RET2]], 0
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP37:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[TMP39:%.*]] = xor <32 x i16> [[TMP37]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP40:%.*]] = and <32 x i16> [[TMP39]], [[TMP36]]
+; CHECK-NEXT:    [[TMP41:%.*]] = icmp eq <32 x i16> [[TMP40]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <32 x i1> [[TMP38]], [[TMP41]]
+; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP43:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP3]] to i32
+; CHECK-NEXT:    [[TMP44:%.*]] = bitcast <32 x i1> [[TMP42]] to i32
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i32 [[_MSPROP2]], [[TMP43]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i32 [[RET3]], [[TMP44]]
+; CHECK-NEXT:    [[TMP45:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP46:%.*]] = and <32 x i16> [[A0]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP48:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP49:%.*]] = and <32 x i16> [[A1]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP51:%.*]] = icmp uge <32 x i16> [[TMP46]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = icmp uge <32 x i16> [[TMP47]], [[TMP49]]
+; CHECK-NEXT:    [[TMP53:%.*]] = xor <32 x i1> [[TMP51]], [[TMP52]]
+; CHECK-NEXT:    [[TMP54:%.*]] = icmp uge <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP55:%.*]] = bitcast <32 x i1> [[TMP53]] to i32
+; CHECK-NEXT:    [[TMP56:%.*]] = bitcast <32 x i1> [[TMP54]] to i32
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i32 [[_MSPROP4]], [[TMP55]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i32 [[RET4]], [[TMP56]]
+; CHECK-NEXT:    [[TMP57:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP58:%.*]] = and <32 x i16> [[A0]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP60:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP61:%.*]] = and <32 x i16> [[A1]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP63:%.*]] = icmp ugt <32 x i16> [[TMP58]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = icmp ugt <32 x i16> [[TMP59]], [[TMP61]]
+; CHECK-NEXT:    [[TMP65:%.*]] = xor <32 x i1> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP66:%.*]] = icmp ugt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP67:%.*]] = bitcast <32 x i1> [[TMP65]] to i32
+; CHECK-NEXT:    [[TMP68:%.*]] = bitcast <32 x i1> [[TMP66]] to i32
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i32 [[_MSPROP5]], [[TMP67]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i32 [[RET5]], [[TMP68]]
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i32 [[_MSPROP6]], 0
+; CHECK-NEXT:    [[RET7:%.*]] = add i32 [[RET6]], -1
+; CHECK-NEXT:    store i32 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RET7]]
+;
+  %res0 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 0, i32 -1)
+  %res1 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 1, i32 -1)
+  %ret1 = add i32 %res0, %res1
+  %res2 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 2, i32 -1)
+  %ret2 = add i32 %ret1, %res2
+  %res3 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 3, i32 -1)
+  %ret3 = add i32 %ret2, %res3
+  %res4 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 4, i32 -1)
+  %ret4 = add i32 %ret3, %res4
+  %res5 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 5, i32 -1)
+  %ret5 = add i32 %ret4, %res5
+  %res6 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 6, i32 -1)
+  %ret6 = add i32 %ret5, %res6
+  %res7 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 7, i32 -1)
+  %ret7 = add i32 %ret6, %res7
+  ret i32 %ret7
+}
+
+define i32 @test_mask_ucmp_w_512(<32 x i16> %a0, <32 x i16> %a1, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_ucmp_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <32 x i16> [[A0:%.*]], [[A1:%.*]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <32 x i16> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[TMP5]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP8:%.*]] = and <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP6]], [[TMP9]]
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = and <32 x i1> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i1> [[TMP17]] to i32
+; CHECK-NEXT:    [[TMP20:%.*]] = bitcast <32 x i1> [[TMP18]] to i32
+; CHECK-NEXT:    [[TMP21:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP22:%.*]] = and <32 x i16> [[A0]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP24:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP25:%.*]] = and <32 x i16> [[A1]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP27:%.*]] = icmp ult <32 x i16> [[TMP22]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = icmp ult <32 x i16> [[TMP23]], [[TMP25]]
+; CHECK-NEXT:    [[TMP29:%.*]] = xor <32 x i1> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP30:%.*]] = icmp ult <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP31:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP32:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP33:%.*]] = and <32 x i1> [[TMP29]], [[TMP31]]
+; CHECK-NEXT:    [[TMP34:%.*]] = and <32 x i1> [[TMP30]], [[TMP31]]
+; CHECK-NEXT:    [[TMP35:%.*]] = and <32 x i1> [[TMP29]], [[TMP32]]
+; CHECK-NEXT:    [[TMP36:%.*]] = or <32 x i1> [[TMP33]], [[TMP34]]
+; CHECK-NEXT:    [[TMP37:%.*]] = or <32 x i1> [[TMP36]], [[TMP35]]
+; CHECK-NEXT:    [[TMP38:%.*]] = and <32 x i1> [[TMP30]], [[TMP32]]
+; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <32 x i1> [[TMP37]] to i32
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <32 x i1> [[TMP38]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP19]], [[TMP39]]
+; CHECK-NEXT:    [[RET1:%.*]] = add i32 [[TMP20]], [[TMP40]]
+; CHECK-NEXT:    [[TMP41:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP42:%.*]] = and <32 x i16> [[A0]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP44:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP45:%.*]] = and <32 x i16> [[A1]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP47:%.*]] = icmp ule <32 x i16> [[TMP42]], [[TMP46]]
+; CHECK-NEXT:    [[TMP48:%.*]] = icmp ule <32 x i16> [[TMP43]], [[TMP45]]
+; CHECK-NEXT:    [[TMP49:%.*]] = xor <32 x i1> [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = icmp ule <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP51:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP52:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP53:%.*]] = and <32 x i1> [[TMP49]], [[TMP51]]
+; CHECK-NEXT:    [[TMP54:%.*]] = and <32 x i1> [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP55:%.*]] = and <32 x i1> [[TMP49]], [[TMP52]]
+; CHECK-NEXT:    [[TMP56:%.*]] = or <32 x i1> [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP57:%.*]] = or <32 x i1> [[TMP56]], [[TMP55]]
+; CHECK-NEXT:    [[TMP58:%.*]] = and <32 x i1> [[TMP50]], [[TMP52]]
+; CHECK-NEXT:    [[TMP59:%.*]] = bitcast <32 x i1> [[TMP57]] to i32
+; CHECK-NEXT:    [[TMP60:%.*]] = bitcast <32 x i1> [[TMP58]] to i32
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or i32 [[_MSPROP]], [[TMP59]]
+; CHECK-NEXT:    [[RET2:%.*]] = add i32 [[RET1]], [[TMP60]]
+; CHECK-NEXT:    [[TMP61:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP62:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP63:%.*]] = and <32 x i1> zeroinitializer, [[TMP61]]
+; CHECK-NEXT:    [[TMP64:%.*]] = and <32 x i1> zeroinitializer, [[TMP61]]
+; CHECK-NEXT:    [[TMP65:%.*]] = and <32 x i1> zeroinitializer, [[TMP62]]
+; CHECK-NEXT:    [[TMP66:%.*]] = or <32 x i1> [[TMP63]], [[TMP64]]
+; CHECK-NEXT:    [[TMP67:%.*]] = or <32 x i1> [[TMP66]], [[TMP65]]
+; CHECK-NEXT:    [[TMP68:%.*]] = and <32 x i1> zeroinitializer, [[TMP62]]
+; CHECK-NEXT:    [[TMP69:%.*]] = bitcast <32 x i1> [[TMP67]] to i32
+; CHECK-NEXT:    [[TMP70:%.*]] = bitcast <32 x i1> [[TMP68]] to i32
+; CHECK-NEXT:    [[_MSPROP2:%.*]] = or i32 [[_MSPROP1]], [[TMP69]]
+; CHECK-NEXT:    [[RET3:%.*]] = add i32 [[RET2]], [[TMP70]]
+; CHECK-NEXT:    [[TMP71:%.*]] = xor <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP72:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP73:%.*]] = icmp ne <32 x i16> [[TMP72]], zeroinitializer
+; CHECK-NEXT:    [[TMP74:%.*]] = xor <32 x i16> [[TMP72]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP75:%.*]] = and <32 x i16> [[TMP74]], [[TMP71]]
+; CHECK-NEXT:    [[TMP76:%.*]] = icmp eq <32 x i16> [[TMP75]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP3:%.*]] = and <32 x i1> [[TMP73]], [[TMP76]]
+; CHECK-NEXT:    [[TMP77:%.*]] = icmp ne <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP78:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP79:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP80:%.*]] = and <32 x i1> [[_MSPROP_ICMP3]], [[TMP78]]
+; CHECK-NEXT:    [[TMP81:%.*]] = and <32 x i1> [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP82:%.*]] = and <32 x i1> [[_MSPROP_ICMP3]], [[TMP79]]
+; CHECK-NEXT:    [[TMP83:%.*]] = or <32 x i1> [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP84:%.*]] = or <32 x i1> [[TMP83]], [[TMP82]]
+; CHECK-NEXT:    [[TMP85:%.*]] = and <32 x i1> [[TMP77]], [[TMP79]]
+; CHECK-NEXT:    [[TMP86:%.*]] = bitcast <32 x i1> [[TMP84]] to i32
+; CHECK-NEXT:    [[TMP87:%.*]] = bitcast <32 x i1> [[TMP85]] to i32
+; CHECK-NEXT:    [[_MSPROP4:%.*]] = or i32 [[_MSPROP2]], [[TMP86]]
+; CHECK-NEXT:    [[RET4:%.*]] = add i32 [[RET3]], [[TMP87]]
+; CHECK-NEXT:    [[TMP88:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP89:%.*]] = and <32 x i16> [[A0]], [[TMP88]]
+; CHECK-NEXT:    [[TMP90:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP91:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP92:%.*]] = and <32 x i16> [[A1]], [[TMP91]]
+; CHECK-NEXT:    [[TMP93:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP94:%.*]] = icmp uge <32 x i16> [[TMP89]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = icmp uge <32 x i16> [[TMP90]], [[TMP92]]
+; CHECK-NEXT:    [[TMP96:%.*]] = xor <32 x i1> [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = icmp uge <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP98:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP99:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP100:%.*]] = and <32 x i1> [[TMP96]], [[TMP98]]
+; CHECK-NEXT:    [[TMP101:%.*]] = and <32 x i1> [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP102:%.*]] = and <32 x i1> [[TMP96]], [[TMP99]]
+; CHECK-NEXT:    [[TMP103:%.*]] = or <32 x i1> [[TMP100]], [[TMP101]]
+; CHECK-NEXT:    [[TMP104:%.*]] = or <32 x i1> [[TMP103]], [[TMP102]]
+; CHECK-NEXT:    [[TMP105:%.*]] = and <32 x i1> [[TMP97]], [[TMP99]]
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast <32 x i1> [[TMP104]] to i32
+; CHECK-NEXT:    [[TMP107:%.*]] = bitcast <32 x i1> [[TMP105]] to i32
+; CHECK-NEXT:    [[_MSPROP5:%.*]] = or i32 [[_MSPROP4]], [[TMP106]]
+; CHECK-NEXT:    [[RET5:%.*]] = add i32 [[RET4]], [[TMP107]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP109:%.*]] = and <32 x i16> [[A0]], [[TMP108]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[A0]], [[TMP1]]
+; CHECK-NEXT:    [[TMP111:%.*]] = xor <32 x i16> [[TMP2]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP112:%.*]] = and <32 x i16> [[A1]], [[TMP111]]
+; CHECK-NEXT:    [[TMP113:%.*]] = or <32 x i16> [[A1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP114:%.*]] = icmp ugt <32 x i16> [[TMP109]], [[TMP113]]
+; CHECK-NEXT:    [[TMP115:%.*]] = icmp ugt <32 x i16> [[TMP110]], [[TMP112]]
+; CHECK-NEXT:    [[TMP116:%.*]] = xor <32 x i1> [[TMP114]], [[TMP115]]
+; CHECK-NEXT:    [[TMP117:%.*]] = icmp ugt <32 x i16> [[A0]], [[A1]]
+; CHECK-NEXT:    [[TMP118:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP119:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP120:%.*]] = and <32 x i1> [[TMP116]], [[TMP118]]
+; CHECK-NEXT:    [[TMP121:%.*]] = and <32 x i1> [[TMP117]], [[TMP118]]
+; CHECK-NEXT:    [[TMP122:%.*]] = and <32 x i1> [[TMP116]], [[TMP119]]
+; CHECK-NEXT:    [[TMP123:%.*]] = or <32 x i1> [[TMP120]], [[TMP121]]
+; CHECK-NEXT:    [[TMP124:%.*]] = or <32 x i1> [[TMP123]], [[TMP122]]
+; CHECK-NEXT:    [[TMP125:%.*]] = and <32 x i1> [[TMP117]], [[TMP119]]
+; CHECK-NEXT:    [[TMP126:%.*]] = bitcast <32 x i1> [[TMP124]] to i32
+; CHECK-NEXT:    [[TMP127:%.*]] = bitcast <32 x i1> [[TMP125]] to i32
+; CHECK-NEXT:    [[_MSPROP6:%.*]] = or i32 [[_MSPROP5]], [[TMP126]]
+; CHECK-NEXT:    [[RET6:%.*]] = add i32 [[RET5]], [[TMP127]]
+; CHECK-NEXT:    [[TMP128:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP129:%.*]] = bitcast i32 [[MASK]] to <32 x i1>
+; CHECK-NEXT:    [[TMP130:%.*]] = and <32 x i1> zeroinitializer, [[TMP128]]
+; CHECK-NEXT:    [[TMP131:%.*]] = and <32 x i1> splat (i1 true), [[TMP128]]
+; CHECK-NEXT:    [[TMP132:%.*]] = and <32 x i1> zeroinitializer, [[TMP129]]
+; CHECK-NEXT:    [[TMP133:%.*]] = or <32 x i1> [[TMP130]], [[TMP131]]
+; CHECK-NEXT:    [[TMP134:%.*]] = or <32 x i1> [[TMP133]], [[TMP132]]
+; CHECK-NEXT:    [[TMP135:%.*]] = and <32 x i1> splat (i1 true), [[TMP129]]
+; CHECK-NEXT:    [[TMP136:%.*]] = bitcast <32 x i1> [[TMP134]] to i32
+; CHECK-NEXT:    [[TMP137:%.*]] = bitcast <32 x i1> [[TMP135]] to i32
+; CHECK-NEXT:    [[_MSPROP7:%.*]] = or i32 [[_MSPROP6]], [[TMP136]]
+; CHECK-NEXT:    [[RET7:%.*]] = add i32 [[RET6]], [[TMP137]]
+; CHECK-NEXT:    store i32 [[_MSPROP7]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RET7]]
+;
+  %res0 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 0, i32 %mask)
+  %res1 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 1, i32 %mask)
+  %ret1 = add i32 %res0, %res1
+  %res2 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 2, i32 %mask)
+  %ret2 = add i32 %ret1, %res2
+  %res3 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 3, i32 %mask)
+  %ret3 = add i32 %ret2, %res3
+  %res4 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 4, i32 %mask)
+  %ret4 = add i32 %ret3, %res4
+  %res5 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 5, i32 %mask)
+  %ret5 = add i32 %ret4, %res5
+  %res6 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 6, i32 %mask)
+  %ret6 = add i32 %ret5, %res6
+  %res7 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i32 7, i32 %mask)
+  %ret7 = add i32 %ret6, %res7
+  ret i32 %ret7
+}
+
+declare i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16>, <32 x i16>, i32, i32) nounwind readnone
+
+
+declare <64 x i8> @llvm.x86.avx512.mask.pavg.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @mm512_avg_epu8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @mm512_avg_epu8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @mm512_mask_avg_epu8(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) nounwind #0 {
+; CHECK-LABEL: @mm512_mask_avg_epu8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  ret <64 x i8> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pavg.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @mm512_avg_epu16(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @mm512_avg_epu16(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @mm512_mask_avg_epu16(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @mm512_mask_avg_epu16(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pabs.w.512(<32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pabs_w_512(<32 x i16> %x0, <32 x i16> %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pabs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = call <32 x i16> @llvm.abs.v32i16(<32 x i16> [[X0:%.*]], i1 false)
+; CHECK-NEXT:    store <32 x i16> [[TMP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pabs.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pabs_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pabs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.abs.v32i16(<32 x i16> [[X0:%.*]], i1 false)
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP1]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pabs.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
+  ret <32 x i16> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pabs.b.512(<64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_int_x86_avx512_pabs_b_512(<64 x i8> %x0, <64 x i8> %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pabs_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = call <64 x i8> @llvm.abs.v64i8(<64 x i8> [[X0:%.*]], i1 false)
+; CHECK-NEXT:    store <64 x i8> [[TMP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP2]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pabs.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pabs_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pabs_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <64 x i8> @llvm.abs.v64i8(<64 x i8> [[X0:%.*]], i1 false)
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP2]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[X2:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP1]], <64 x i8> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP4]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[TMP1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP4]], <64 x i8> [[X1]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pabs.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
+  ret <64 x i8> %res
+}
+
+declare i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8>, <64 x i8>, i64)
+
+define i64 @test_int_x86_avx512_ptestm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_ptestm_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = and <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = and <64 x i8> [[X0:%.*]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and <64 x i8> [[TMP1]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <64 x i8> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <64 x i8> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP11]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i8> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i64 [[X2:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP17]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <64 x i1> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <64 x i1> [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <64 x i1> [[TMP22]], [[TMP21]]
+; CHECK-NEXT:    [[TMP24:%.*]] = and <64 x i1> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <64 x i1> [[TMP23]] to i64
+; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <64 x i1> [[TMP24]] to i64
+; CHECK-NEXT:    [[TMP27:%.*]] = and <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP28:%.*]] = and <64 x i8> [[X0]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = and <64 x i8> [[TMP1]], [[X1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = or <64 x i8> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP31:%.*]] = or <64 x i8> [[TMP30]], [[TMP29]]
+; CHECK-NEXT:    [[TMP32:%.*]] = and <64 x i8> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = xor <64 x i8> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP34:%.*]] = or <64 x i8> [[TMP31]], zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <64 x i8> [[TMP34]], zeroinitializer
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <64 x i8> [[TMP34]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP37:%.*]] = and <64 x i8> [[TMP36]], [[TMP33]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp eq <64 x i8> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <64 x i1> [[TMP35]], [[TMP38]]
+; CHECK-NEXT:    [[TMP39:%.*]] = icmp ne <64 x i8> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP1]] to i64
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <64 x i1> [[TMP39]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP25]], [[TMP40]]
+; CHECK-NEXT:    [[RES2:%.*]] = add i64 [[TMP26]], [[TMP41]]
+; CHECK-NEXT:    store i64 [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES2]]
+;
+  %res = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
+  %res1 = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
+  %res2 = add i64 %res, %res1
+  ret i64 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16>, <32 x i16>, i32)
+
+define i32 @test_int_x86_avx512_ptestm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_ptestm_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = and <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = and <32 x i16> [[X0:%.*]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and <32 x i16> [[TMP1]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <32 x i16> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <32 x i16> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP11]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP17]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <32 x i1> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <32 x i1> [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <32 x i1> [[TMP22]], [[TMP21]]
+; CHECK-NEXT:    [[TMP24:%.*]] = and <32 x i1> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <32 x i1> [[TMP23]] to i32
+; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <32 x i1> [[TMP24]] to i32
+; CHECK-NEXT:    [[TMP27:%.*]] = and <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP28:%.*]] = and <32 x i16> [[X0]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = and <32 x i16> [[TMP1]], [[X1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = or <32 x i16> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP31:%.*]] = or <32 x i16> [[TMP30]], [[TMP29]]
+; CHECK-NEXT:    [[TMP32:%.*]] = and <32 x i16> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = xor <32 x i16> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP34:%.*]] = or <32 x i16> [[TMP31]], zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP34]], zeroinitializer
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <32 x i16> [[TMP34]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP37:%.*]] = and <32 x i16> [[TMP36]], [[TMP33]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp eq <32 x i16> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <32 x i1> [[TMP35]], [[TMP38]]
+; CHECK-NEXT:    [[TMP39:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP1]] to i32
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <32 x i1> [[TMP39]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP25]], [[TMP40]]
+; CHECK-NEXT:    [[RES2:%.*]] = add i32 [[TMP26]], [[TMP41]]
+; CHECK-NEXT:    store i32 [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES2]]
+;
+  %res = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
+  %res1 = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
+  %res2 = add i32 %res, %res1
+  ret i32 %res2
+}
+
+declare i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8>, <64 x i8>, i64 %x2)
+
+define i64 @test_int_x86_avx512_ptestnm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_ptestnm_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = and <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = and <64 x i8> [[X0:%.*]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and <64 x i8> [[TMP1]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <64 x i8> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <64 x i8> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP11]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i8> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i64 [[X2:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP17]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <64 x i1> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = and <64 x i1> [[_MSPROP_ICMP]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <64 x i1> [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <64 x i1> [[TMP22]], [[TMP21]]
+; CHECK-NEXT:    [[TMP24:%.*]] = and <64 x i1> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <64 x i1> [[TMP23]] to i64
+; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <64 x i1> [[TMP24]] to i64
+; CHECK-NEXT:    [[TMP27:%.*]] = and <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP28:%.*]] = and <64 x i8> [[X0]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = and <64 x i8> [[TMP1]], [[X1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = or <64 x i8> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP31:%.*]] = or <64 x i8> [[TMP30]], [[TMP29]]
+; CHECK-NEXT:    [[TMP32:%.*]] = and <64 x i8> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = xor <64 x i8> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP34:%.*]] = or <64 x i8> [[TMP31]], zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <64 x i8> [[TMP34]], zeroinitializer
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <64 x i8> [[TMP34]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP37:%.*]] = and <64 x i8> [[TMP36]], [[TMP33]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp eq <64 x i8> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <64 x i1> [[TMP35]], [[TMP38]]
+; CHECK-NEXT:    [[TMP39:%.*]] = icmp eq <64 x i8> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP1]] to i64
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <64 x i1> [[TMP39]] to i64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i64 [[TMP25]], [[TMP40]]
+; CHECK-NEXT:    [[RES2:%.*]] = add i64 [[TMP26]], [[TMP41]]
+; CHECK-NEXT:    store i64 [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[RES2]]
+;
+  %res = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
+  %res1 = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
+  %res2 = add i64 %res, %res1
+  ret i64 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16>, <32 x i16>, i32 %x2)
+
+define i32 @test_int_x86_avx512_ptestnm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_ptestnm_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = and <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = and <32 x i16> [[X0:%.*]], [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and <32 x i16> [[TMP1]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP7:%.*]] = or <32 x i16> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <32 x i16> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP11]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP19:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP17]]
+; CHECK-NEXT:    [[TMP20:%.*]] = and <32 x i1> [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = and <32 x i1> [[_MSPROP_ICMP]], [[TMP18]]
+; CHECK-NEXT:    [[TMP22:%.*]] = or <32 x i1> [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP23:%.*]] = or <32 x i1> [[TMP22]], [[TMP21]]
+; CHECK-NEXT:    [[TMP24:%.*]] = and <32 x i1> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[TMP25:%.*]] = bitcast <32 x i1> [[TMP23]] to i32
+; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <32 x i1> [[TMP24]] to i32
+; CHECK-NEXT:    [[TMP27:%.*]] = and <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP28:%.*]] = and <32 x i16> [[X0]], [[TMP2]]
+; CHECK-NEXT:    [[TMP29:%.*]] = and <32 x i16> [[TMP1]], [[X1]]
+; CHECK-NEXT:    [[TMP30:%.*]] = or <32 x i16> [[TMP27]], [[TMP28]]
+; CHECK-NEXT:    [[TMP31:%.*]] = or <32 x i16> [[TMP30]], [[TMP29]]
+; CHECK-NEXT:    [[TMP32:%.*]] = and <32 x i16> [[X0]], [[X1]]
+; CHECK-NEXT:    [[TMP33:%.*]] = xor <32 x i16> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP34:%.*]] = or <32 x i16> [[TMP31]], zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP34]], zeroinitializer
+; CHECK-NEXT:    [[TMP36:%.*]] = xor <32 x i16> [[TMP34]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP37:%.*]] = and <32 x i16> [[TMP36]], [[TMP33]]
+; CHECK-NEXT:    [[TMP38:%.*]] = icmp eq <32 x i16> [[TMP37]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <32 x i1> [[TMP35]], [[TMP38]]
+; CHECK-NEXT:    [[TMP39:%.*]] = icmp eq <32 x i16> [[TMP32]], zeroinitializer
+; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP1]] to i32
+; CHECK-NEXT:    [[TMP41:%.*]] = bitcast <32 x i1> [[TMP39]] to i32
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or i32 [[TMP25]], [[TMP40]]
+; CHECK-NEXT:    [[RES2:%.*]] = add i32 [[TMP26]], [[TMP41]]
+; CHECK-NEXT:    store i32 [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES2]]
+;
+  %res = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
+  %res1 = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
+  %res2 = add i32 %res, %res1
+  ret i32 %res2
+}
+
+declare i64 @llvm.x86.avx512.cvtb2mask.512(<64 x i8>)
+
+define i64 @test_int_x86_avx512_cvtb2mask_512(<64 x i8> %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_cvtb2mask_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = xor <64 x i8> [[X0:%.*]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[TMP1]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP4:%.*]] = and <64 x i8> [[TMP2]], [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <64 x i8> [[TMP2]], [[TMP1]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ult <64 x i8> [[TMP4]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ult <64 x i8> [[TMP5]], splat (i8 -128)
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i1> [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp slt <64 x i8> [[X0]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <64 x i1> [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <64 x i1> [[TMP9]] to i64
+; CHECK-NEXT:    store i64 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i64 [[TMP11]]
+;
+  %res = call i64 @llvm.x86.avx512.cvtb2mask.512(<64 x i8> %x0)
+  ret i64 %res
+}
+
+declare i32 @llvm.x86.avx512.cvtw2mask.512(<32 x i16>)
+
+define i32 @test_int_x86_avx512_cvtw2mask_512(<32 x i16> %x0) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_cvtw2mask_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = xor <32 x i16> [[X0:%.*]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[TMP1]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP4:%.*]] = and <32 x i16> [[TMP2]], [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP2]], [[TMP1]]
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ult <32 x i16> [[TMP4]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ult <32 x i16> [[TMP5]], splat (i16 -32768)
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i1> [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp slt <32 x i16> [[X0]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i1> [[TMP8]] to i32
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i1> [[TMP9]] to i32
+; CHECK-NEXT:    store i32 [[TMP10]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP11]]
+;
+  %res = call i32 @llvm.x86.avx512.cvtw2mask.512(<32 x i16> %x0)
+  ret i32 %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmulhu.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmulhu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulhu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulhu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulhu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmulh.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmulh_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulh_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmul.hr.sw.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmulhr_sw_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulhr_sw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulhr_sw_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulhr_sw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmaddubs.w.512(<64 x i8>, <64 x i8>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_pmaddubs_w_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaddubs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmaddubs_w_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaddubs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.pmaddw.d.512(<32 x i16>, <32 x i16>, <16 x i32>, i16)
+
+define <16 x i32> @test_int_x86_avx512_pmaddw_d_512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaddw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <16 x i32> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <16 x i32> [[TMP7]]
+;
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2, i16 -1)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_int_x86_avx512_mask_pmaddw_d_512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2, i16 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaddw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i16, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i16 [[TMP3]] to <16 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i16 [[X3:%.*]] to <16 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> zeroinitializer, <16 x i32> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <16 x i32> [[TMP9]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <16 x i32> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <16 x i32> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP10]], <16 x i32> [[TMP15]], <16 x i32> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> [[TMP9]], <16 x i32> [[X2]]
+; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <16 x i32> [[TMP16]]
+;
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2, i16 %x3)
+  ret <16 x i32> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.permvar.hi.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.vpermt2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = and i16 [[TMP4]], 31
+; CHECK-NEXT:    [[TMP6:%.*]] = or i16 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = and i16 [[TMP7]], 31
+; CHECK-NEXT:    [[TMP9:%.*]] = or i16 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP11:%.*]] = and i16 [[TMP10]], 31
+; CHECK-NEXT:    [[TMP12:%.*]] = or i16 [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP14:%.*]] = and i16 [[TMP13]], 31
+; CHECK-NEXT:    [[TMP15:%.*]] = or i16 [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP17:%.*]] = and i16 [[TMP16]], 31
+; CHECK-NEXT:    [[TMP18:%.*]] = or i16 [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP20:%.*]] = and i16 [[TMP19]], 31
+; CHECK-NEXT:    [[TMP21:%.*]] = or i16 [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP23:%.*]] = and i16 [[TMP22]], 31
+; CHECK-NEXT:    [[TMP24:%.*]] = or i16 [[TMP22]], [[TMP23]]
+; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP26:%.*]] = and i16 [[TMP25]], 31
+; CHECK-NEXT:    [[TMP27:%.*]] = or i16 [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP29:%.*]] = and i16 [[TMP28]], 31
+; CHECK-NEXT:    [[TMP30:%.*]] = or i16 [[TMP28]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP32:%.*]] = and i16 [[TMP31]], 31
+; CHECK-NEXT:    [[TMP33:%.*]] = or i16 [[TMP31]], [[TMP32]]
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP35:%.*]] = and i16 [[TMP34]], 31
+; CHECK-NEXT:    [[TMP36:%.*]] = or i16 [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP38:%.*]] = and i16 [[TMP37]], 31
+; CHECK-NEXT:    [[TMP39:%.*]] = or i16 [[TMP37]], [[TMP38]]
+; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP41:%.*]] = and i16 [[TMP40]], 31
+; CHECK-NEXT:    [[TMP42:%.*]] = or i16 [[TMP40]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP44:%.*]] = and i16 [[TMP43]], 31
+; CHECK-NEXT:    [[TMP45:%.*]] = or i16 [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP47:%.*]] = and i16 [[TMP46]], 31
+; CHECK-NEXT:    [[TMP48:%.*]] = or i16 [[TMP46]], [[TMP47]]
+; CHECK-NEXT:    [[TMP49:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP50:%.*]] = and i16 [[TMP49]], 31
+; CHECK-NEXT:    [[TMP51:%.*]] = or i16 [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP53:%.*]] = and i16 [[TMP52]], 31
+; CHECK-NEXT:    [[TMP54:%.*]] = or i16 [[TMP52]], [[TMP53]]
+; CHECK-NEXT:    [[TMP55:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP56:%.*]] = and i16 [[TMP55]], 31
+; CHECK-NEXT:    [[TMP57:%.*]] = or i16 [[TMP55]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP59:%.*]] = and i16 [[TMP58]], 31
+; CHECK-NEXT:    [[TMP60:%.*]] = or i16 [[TMP58]], [[TMP59]]
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP62:%.*]] = and i16 [[TMP61]], 31
+; CHECK-NEXT:    [[TMP63:%.*]] = or i16 [[TMP61]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP65:%.*]] = and i16 [[TMP64]], 31
+; CHECK-NEXT:    [[TMP66:%.*]] = or i16 [[TMP64]], [[TMP65]]
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP68:%.*]] = and i16 [[TMP67]], 31
+; CHECK-NEXT:    [[TMP69:%.*]] = or i16 [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP71:%.*]] = and i16 [[TMP70]], 31
+; CHECK-NEXT:    [[TMP72:%.*]] = or i16 [[TMP70]], [[TMP71]]
+; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP74:%.*]] = and i16 [[TMP73]], 31
+; CHECK-NEXT:    [[TMP75:%.*]] = or i16 [[TMP73]], [[TMP74]]
+; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP77:%.*]] = and i16 [[TMP76]], 31
+; CHECK-NEXT:    [[TMP78:%.*]] = or i16 [[TMP76]], [[TMP77]]
+; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP80:%.*]] = and i16 [[TMP79]], 31
+; CHECK-NEXT:    [[TMP81:%.*]] = or i16 [[TMP79]], [[TMP80]]
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP83:%.*]] = and i16 [[TMP82]], 31
+; CHECK-NEXT:    [[TMP84:%.*]] = or i16 [[TMP82]], [[TMP83]]
+; CHECK-NEXT:    [[TMP85:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP86:%.*]] = and i16 [[TMP85]], 31
+; CHECK-NEXT:    [[TMP87:%.*]] = or i16 [[TMP85]], [[TMP86]]
+; CHECK-NEXT:    [[TMP88:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP89:%.*]] = and i16 [[TMP88]], 31
+; CHECK-NEXT:    [[TMP90:%.*]] = or i16 [[TMP88]], [[TMP89]]
+; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP92:%.*]] = and i16 [[TMP91]], 31
+; CHECK-NEXT:    [[TMP93:%.*]] = or i16 [[TMP91]], [[TMP92]]
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP95:%.*]] = and i16 [[TMP94]], 31
+; CHECK-NEXT:    [[TMP96:%.*]] = or i16 [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP98:%.*]] = and i16 [[TMP97]], 31
+; CHECK-NEXT:    [[TMP99:%.*]] = or i16 [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP100:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP99]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP101:%.*]], label [[TMP102:%.*]], !prof [[PROF1]]
+; CHECK:       101:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       102:
+; CHECK-NEXT:    [[TMP103:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[TMP100]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP103]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.vpermt2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> [[TMP1]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], [[X1]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], [[TMP1]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.vpermt2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.maskz.vpermt2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_maskz_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], zeroinitializer
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.maskz.vpermt2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.vpermi2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_vpermi2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_vpermi2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = and i16 [[TMP4]], 31
+; CHECK-NEXT:    [[TMP6:%.*]] = or i16 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = and i16 [[TMP7]], 31
+; CHECK-NEXT:    [[TMP9:%.*]] = or i16 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP11:%.*]] = and i16 [[TMP10]], 31
+; CHECK-NEXT:    [[TMP12:%.*]] = or i16 [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP14:%.*]] = and i16 [[TMP13]], 31
+; CHECK-NEXT:    [[TMP15:%.*]] = or i16 [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP17:%.*]] = and i16 [[TMP16]], 31
+; CHECK-NEXT:    [[TMP18:%.*]] = or i16 [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP20:%.*]] = and i16 [[TMP19]], 31
+; CHECK-NEXT:    [[TMP21:%.*]] = or i16 [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP23:%.*]] = and i16 [[TMP22]], 31
+; CHECK-NEXT:    [[TMP24:%.*]] = or i16 [[TMP22]], [[TMP23]]
+; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP26:%.*]] = and i16 [[TMP25]], 31
+; CHECK-NEXT:    [[TMP27:%.*]] = or i16 [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP29:%.*]] = and i16 [[TMP28]], 31
+; CHECK-NEXT:    [[TMP30:%.*]] = or i16 [[TMP28]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP32:%.*]] = and i16 [[TMP31]], 31
+; CHECK-NEXT:    [[TMP33:%.*]] = or i16 [[TMP31]], [[TMP32]]
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP35:%.*]] = and i16 [[TMP34]], 31
+; CHECK-NEXT:    [[TMP36:%.*]] = or i16 [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP38:%.*]] = and i16 [[TMP37]], 31
+; CHECK-NEXT:    [[TMP39:%.*]] = or i16 [[TMP37]], [[TMP38]]
+; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP41:%.*]] = and i16 [[TMP40]], 31
+; CHECK-NEXT:    [[TMP42:%.*]] = or i16 [[TMP40]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP44:%.*]] = and i16 [[TMP43]], 31
+; CHECK-NEXT:    [[TMP45:%.*]] = or i16 [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP47:%.*]] = and i16 [[TMP46]], 31
+; CHECK-NEXT:    [[TMP48:%.*]] = or i16 [[TMP46]], [[TMP47]]
+; CHECK-NEXT:    [[TMP49:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP50:%.*]] = and i16 [[TMP49]], 31
+; CHECK-NEXT:    [[TMP51:%.*]] = or i16 [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP53:%.*]] = and i16 [[TMP52]], 31
+; CHECK-NEXT:    [[TMP54:%.*]] = or i16 [[TMP52]], [[TMP53]]
+; CHECK-NEXT:    [[TMP55:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP56:%.*]] = and i16 [[TMP55]], 31
+; CHECK-NEXT:    [[TMP57:%.*]] = or i16 [[TMP55]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP59:%.*]] = and i16 [[TMP58]], 31
+; CHECK-NEXT:    [[TMP60:%.*]] = or i16 [[TMP58]], [[TMP59]]
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP62:%.*]] = and i16 [[TMP61]], 31
+; CHECK-NEXT:    [[TMP63:%.*]] = or i16 [[TMP61]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP65:%.*]] = and i16 [[TMP64]], 31
+; CHECK-NEXT:    [[TMP66:%.*]] = or i16 [[TMP64]], [[TMP65]]
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP68:%.*]] = and i16 [[TMP67]], 31
+; CHECK-NEXT:    [[TMP69:%.*]] = or i16 [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP71:%.*]] = and i16 [[TMP70]], 31
+; CHECK-NEXT:    [[TMP72:%.*]] = or i16 [[TMP70]], [[TMP71]]
+; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP74:%.*]] = and i16 [[TMP73]], 31
+; CHECK-NEXT:    [[TMP75:%.*]] = or i16 [[TMP73]], [[TMP74]]
+; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP77:%.*]] = and i16 [[TMP76]], 31
+; CHECK-NEXT:    [[TMP78:%.*]] = or i16 [[TMP76]], [[TMP77]]
+; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP80:%.*]] = and i16 [[TMP79]], 31
+; CHECK-NEXT:    [[TMP81:%.*]] = or i16 [[TMP79]], [[TMP80]]
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP83:%.*]] = and i16 [[TMP82]], 31
+; CHECK-NEXT:    [[TMP84:%.*]] = or i16 [[TMP82]], [[TMP83]]
+; CHECK-NEXT:    [[TMP85:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP86:%.*]] = and i16 [[TMP85]], 31
+; CHECK-NEXT:    [[TMP87:%.*]] = or i16 [[TMP85]], [[TMP86]]
+; CHECK-NEXT:    [[TMP88:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP89:%.*]] = and i16 [[TMP88]], 31
+; CHECK-NEXT:    [[TMP90:%.*]] = or i16 [[TMP88]], [[TMP89]]
+; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP92:%.*]] = and i16 [[TMP91]], 31
+; CHECK-NEXT:    [[TMP93:%.*]] = or i16 [[TMP91]], [[TMP92]]
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP95:%.*]] = and i16 [[TMP94]], 31
+; CHECK-NEXT:    [[TMP96:%.*]] = or i16 [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP98:%.*]] = and i16 [[TMP97]], 31
+; CHECK-NEXT:    [[TMP99:%.*]] = or i16 [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP100:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X1:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP99]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP101:%.*]], label [[TMP102:%.*]], !prof [[PROF1]]
+; CHECK:       101:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       102:
+; CHECK-NEXT:    [[TMP103:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[TMP100]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP103]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_vpermi2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_vpermi2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X1:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], [[X1]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8>, <64 x i8>, i32, <32 x i16>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_dbpsadbw_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x3, i32 %x4) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_dbpsadbw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], i32 2)
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[X4:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[X3:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[X3]]
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP4:%.*]] = icmp ne i512 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR5:%.*]] = or i1 [[_MSCMP3]], [[_MSCMP4]]
+; CHECK-NEXT:    br i1 [[_MSOR5]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[TMP21:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0]], <64 x i8> [[X1]], i32 3)
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP23:%.*]] = bitcast i32 [[X4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP24:%.*]] = select <32 x i1> [[TMP23]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP25:%.*]] = xor <32 x i16> [[TMP21]], zeroinitializer
+; CHECK-NEXT:    [[TMP26:%.*]] = or <32 x i16> [[TMP25]], zeroinitializer
+; CHECK-NEXT:    [[TMP27:%.*]] = or <32 x i16> [[TMP26]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT1:%.*]] = select <32 x i1> [[TMP22]], <32 x i16> [[TMP27]], <32 x i16> [[TMP24]]
+; CHECK-NEXT:    [[TMP28:%.*]] = select <32 x i1> [[TMP23]], <32 x i16> [[TMP21]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP6:%.*]] = icmp ne i512 [[TMP29]], 0
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP7:%.*]] = icmp ne i512 [[TMP30]], 0
+; CHECK-NEXT:    [[_MSOR8:%.*]] = or i1 [[_MSCMP6]], [[_MSCMP7]]
+; CHECK-NEXT:    br i1 [[_MSOR8]], label [[TMP31:%.*]], label [[TMP32:%.*]], !prof [[PROF1]]
+; CHECK:       31:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       32:
+; CHECK-NEXT:    [[TMP33:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0]], <64 x i8> [[X1]], i32 4)
+; CHECK-NEXT:    [[TMP34:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP_SELECT]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP16]], 0
+; CHECK-NEXT:    [[TMP35:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP34]], <32 x i16> [[_MSPROP_SELECT1]], 1
+; CHECK-NEXT:    [[RES4:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]], <32 x i16> [[TMP28]], 1
+; CHECK-NEXT:    [[TMP36:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP35]], <32 x i16> zeroinitializer, 2
+; CHECK-NEXT:    [[RES5:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES4]], <32 x i16> [[TMP33]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP36]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES5]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <32 x i16> %x3, i32 %x4)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 3, <32 x i16> zeroinitializer, i32 %x4)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 4, <32 x i16> %x3, i32 -1)
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16>  %res, 0
+  %res4 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res3, <32 x i16> %res1, 1
+  %res5 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res4, <32 x i16> %res2, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res5
+}
+
+define <32 x i16> @test_mask_adds_epu16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epu16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epu16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epu16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epu16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epu16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.paddus.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_mask_subs_epu16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epu16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epu16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epu16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epu16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epu16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <64 x i8> @test_mask_adds_epu8_rr_512(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epu8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epu8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP4]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epu8_rm_512(<64 x i8> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP8]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epu8_rmk_512(<64 x i8> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i8> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i8> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP16]], <64 x i8> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[TMP10]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP17]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epu8_rmkz_512(<64 x i8> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epu8_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.paddus.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_mask_subs_epu8_rr_512(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epu8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epu8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP4]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epu8_rm_512(<64 x i8> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP8]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epu8_rmk_512(<64 x i8> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i8> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i8> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP16]], <64 x i8> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[TMP10]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP17]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epu8_rmkz_512(<64 x i8> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epu8_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <32 x i16> @test_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_adds_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_adds_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_subs_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_subs_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP8]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP10]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP17]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i32 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <64 x i8> @test_mask_adds_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP4]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rm_512(<64 x i8> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP8]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rmk_512(<64 x i8> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i8> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i8> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP16]], <64 x i8> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[TMP10]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP17]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rmkz_512(<64 x i8> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_adds_epi8_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_mask_subs_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP10]], <64 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP4]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP11]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rm_512(<64 x i8> %a, ptr %ptr_b) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP8]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rmk_512(<64 x i8> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP10:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <64 x i8> [[TMP10]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <64 x i8> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP16]], <64 x i8> [[TMP13]]
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP12]], <64 x i8> [[TMP10]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP17]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+  ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rmkz_512(<64 x i8> %a, ptr %ptr_b, i64 %mask) nounwind #0 {
+; CHECK-LABEL: @test_mask_subs_epi8_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR7]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <64 x i8>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP2]], [[_MSLD]]
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> [[A:%.*]], <64 x i8> [[B]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %b = load <64 x i8>, ptr %ptr_b
+  %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+  ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+declare <32 x i16> @llvm.x86.avx512.mask.psrlv32hi(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psrav32.hi(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrav32.hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrav32.hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psrav32.hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psllv32hi(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_int_x86_avx512_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_int_x86_avx512_maskz_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.mask.psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> zeroinitializer, i32 %x3)
+  ret <32 x i16> %res
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmov.wb.512(<32 x i16>, <32 x i8>, i32)
+
+define <32 x i8> @test_int_x86_avx512_pmov_wb_512(<32 x i16> %x0, <32 x i8> %x1) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP2]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmov.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 -1)
+  ret <32 x i8> %res
+}
+
+define <32 x i8> @test_int_x86_avx512_mask_pmov_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i8> [[_MSPROP]], <32 x i8> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i8> [[TMP4]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i8> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[TMP10]], <32 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i8> [[TMP4]], <32 x i8> [[X1]]
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP11]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmov.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2)
+  ret <32 x i8> %res
+}
+
+define <32 x i8> @test_int_x86_avx512_maskz_pmov_wb_512(<32 x i16> %x0, i32 %x2) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[_MSPROP]], <32 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i8> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i8> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP4]], <32 x i8> [[TMP9]], <32 x i8> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[TMP3]], <32 x i8> zeroinitializer
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP10]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmov.wb.512(<32 x i16> %x0, <32 x i8> zeroinitializer, i32 %x2)
+  ret <32 x i8> %res
+}
+
+attributes #0 = { sanitize_memory }
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll
new file mode 100644
index 0000000000000..6f2a6ccbfa702
--- /dev/null
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll
@@ -0,0 +1,3687 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt %s -S -mattr=+avx512f -passes=msan 2>&1 | FileCheck %s
+;
+; Forked from llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
+;
+; Strictly handled:
+; - llvm.x86.avx512.dbpsadbw.512
+; - llvm.x86.avx512.ktestc.d, llvm.x86.avx512.ktestc.q, llvm.x86.avx512.ktestz.d, llvm.x86.avx512.ktestz.q
+; - llvm.x86.avx512.mask.pmovs.wb.512, llvm.x86.avx512.mask.pmovs.wb.mem.512
+; - llvm.x86.avx512.mask.pmovus.wb.512, llvm.x86.avx512.mask.pmovus.wb.mem.512
+; - llvm.x86.avx512.mask.pmov.wb.mem.512
+; - llvm.x86.avx512.packssdw.512, llvm.x86.avx512.packsswb.512
+; - llvm.x86.avx512.packusdw.512, llvm.x86.avx512.packuswb.512
+; - llvm.x86.avx512.pmaddubs.w.512, llvm.x86.avx512.pmaddw.d.512
+; - llvm.x86.avx512.psad.bw.512
+;
+; Heuristically handled:
+; - llvm.x86.avx512.kadd.d, llvm.x86.avx512.kadd.q
+; - llvm.x86.avx512.pavg.b.512, llvm.x86.avx512.pavg.w.512
+; - llvm.x86.avx512.permvar.hi.512
+; - llvm.x86.avx512.pmul.hr.sw.512, llvm.x86.avx512.pmulhu.w.512, llvm.x86.avx512.pmulh.w.512
+; - llvm.x86.avx512.pshuf.b.512
+; - llvm.x86.avx512.psllv.w.512
+; - llvm.x86.avx512.psrav.w.512, llvm.x86.avx512.psrlv.w.512
+
+target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i32 @test_int_x86_avx512_kadd_d(<32 x i16> %A, <32 x i16> %B) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_kadd_d(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = xor <32 x i16> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = or <32 x i16> [[TMP0]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = xor <32 x i16> [[TMP3]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP6:%.*]] = and <32 x i16> [[TMP5]], [[TMP2]]
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq <32 x i16> [[TMP6]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[TMP10]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP10]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i16> [[TMP12]], [[TMP9]]
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <32 x i1> [[TMP11]], [[TMP14]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <32 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i1> [[_MSPROP_ICMP]], [[_MSPROP_ICMP1]]
+; CHECK-NEXT:    [[TMP16:%.*]] = call <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1> [[TMP8]], <32 x i1> [[TMP15]])
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <32 x i1> [[_MSPROP]] to i32
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <32 x i1> [[TMP16]] to i32
+; CHECK-NEXT:    [[TMP19:%.*]] = xor i32 [[TMP18]], 0
+; CHECK-NEXT:    [[TMP20:%.*]] = or i32 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne i32 [[TMP20]], 0
+; CHECK-NEXT:    [[TMP22:%.*]] = xor i32 [[TMP20]], -1
+; CHECK-NEXT:    [[TMP23:%.*]] = and i32 [[TMP22]], [[TMP19]]
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP23]], 0
+; CHECK-NEXT:    [[_MSPROP_ICMP2:%.*]] = and i1 [[TMP21]], [[TMP24]]
+; CHECK-NEXT:    [[TMP25:%.*]] = icmp eq i32 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSPROP3:%.*]] = zext i1 [[_MSPROP_ICMP2]] to i32
+; CHECK-NEXT:    [[TMP26:%.*]] = zext i1 [[TMP25]] to i32
+; CHECK-NEXT:    store i32 [[_MSPROP3]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP26]]
+;
+entry:
+  %0 = icmp ne <32 x i16> %A, zeroinitializer
+  %1 = icmp ne <32 x i16> %B, zeroinitializer
+  %2 = call <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1> %0, <32 x i1> %1)
+  %3 = bitcast <32 x i1> %2 to i32
+  %4 = icmp eq i32 %3, 0
+  %5 = zext i1 %4 to i32
+  ret i32 %5
+}
+declare <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1>, <32 x i1>)
+
+define i32 @test_int_x86_avx512_kadd_q(<64 x i8> %A, <64 x i8> %B) nounwind #0 {
+; CHECK-LABEL: @test_int_x86_avx512_kadd_q(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = xor <64 x i8> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = or <64 x i8> [[TMP0]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = xor <64 x i8> [[TMP3]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP6:%.*]] = and <64 x i8> [[TMP5]], [[TMP2]]
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq <64 x i8> [[TMP6]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <64 x i8> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <64 x i8> [[TMP10]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP10]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP13:%.*]] = and <64 x i8> [[TMP12]], [[TMP9]]
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <64 x i1> [[TMP11]], [[TMP14]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <64 x i8> [[B]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i1> [[_MSPROP_ICMP]], [[_MSPROP_ICMP1]]
+; CHECK-NEXT:    [[TMP16:%.*]] = call <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1> [[TMP8]], <64 x i1> [[TMP15]])
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <64 x i1> [[_MSPROP]] to i64
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i1> [[TMP16]] to i64
+; CHECK-NEXT:    [[TMP19:%.*]] = xor i64 [[TMP18]], 0
+; CHECK-NEXT:    [[TMP20:%.*]] = or i64 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne i64 [[TMP20]], 0
+; CHECK-NEXT:    [[TMP22:%.*]] = xor i64 [[TMP20]], -1
+; CHECK-NEXT:    [[TMP23:%.*]] = and i64 [[TMP22]], [[TMP19]]
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[TMP23]], 0
+; CHECK-NEXT:    [[_MSPROP_ICMP2:%.*]] = and i1 [[TMP21]], [[TMP24]]
+; CHECK-NEXT:    [[TMP25:%.*]] = icmp eq i64 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSPROP3:%.*]] = zext i1 [[_MSPROP_ICMP2]] to i32
+; CHECK-NEXT:    [[TMP26:%.*]] = zext i1 [[TMP25]] to i32
+; CHECK-NEXT:    store i32 [[_MSPROP3]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[TMP26]]
+;
+entry:
+  %0 = icmp ne <64 x i8> %A, zeroinitializer
+  %1 = icmp ne <64 x i8> %B, zeroinitializer
+  %2 = call <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1> %0, <64 x i1> %1)
+  %3 = bitcast <64 x i1> %2 to i64
+  %4 = icmp eq i64 %3, 0
+  %5 = zext i1 %4 to i32
+  ret i32 %5
+}
+declare <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1>, <64 x i1>)
+
+define i32 @test_x86_avx512_ktestc_d(<32 x i16> %A, <32 x i16> %B) #0 {
+; CHECK-LABEL: @test_x86_avx512_ktestc_d(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <32 x i16> [[TMP4]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <32 x i16> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <32 x i16> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP11]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <32 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <32 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP]] to i32
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i32 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP1]] to i32
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i32 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1:![0-9]+]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8:[0-9]+]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[RES:%.*]] = call i32 @llvm.x86.avx512.ktestc.d(<32 x i1> [[TMP9]], <32 x i1> [[TMP16]])
+; CHECK-NEXT:    store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %1 = icmp ne <32 x i16> %A, zeroinitializer
+  %2 = icmp ne <32 x i16> %B, zeroinitializer
+  %res = call i32 @llvm.x86.avx512.ktestc.d(<32 x i1> %1, <32 x i1> %2) ; <i32> [#uses=1]
+  ret i32 %res
+}
+declare i32 @llvm.x86.avx512.ktestc.d(<32 x i1>, <32 x i1>) nounwind readnone
+
+define i32 @test_x86_avx512_ktestz_d(<32 x i16> %A, <32 x i16> %B) #0 {
+; CHECK-LABEL: @test_x86_avx512_ktestz_d(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <32 x i16> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <32 x i16> [[TMP4]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <32 x i16> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <32 x i16> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <32 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <32 x i16> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP11]], splat (i16 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <32 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <32 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP]] to i32
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i32 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <32 x i1> [[_MSPROP_ICMP1]] to i32
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i32 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[RES:%.*]] = call i32 @llvm.x86.avx512.ktestz.d(<32 x i1> [[TMP9]], <32 x i1> [[TMP16]])
+; CHECK-NEXT:    store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %1 = icmp ne <32 x i16> %A, zeroinitializer
+  %2 = icmp ne <32 x i16> %B, zeroinitializer
+  %res = call i32 @llvm.x86.avx512.ktestz.d(<32 x i1> %1, <32 x i1> %2) ; <i32> [#uses=1]
+  ret i32 %res
+}
+declare i32 @llvm.x86.avx512.ktestz.d(<32 x i1>, <32 x i1>) nounwind readnone
+
+define i32 @test_x86_avx512_ktestc_q(<64 x i8> %A, <64 x i8> %B) #0 {
+; CHECK-LABEL: @test_x86_avx512_ktestc_q(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = or <64 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[TMP4]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <64 x i8> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <64 x i8> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <64 x i8> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP11]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i8> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <64 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <64 x i8> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP]] to i64
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP1]] to i64
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[RES:%.*]] = call i32 @llvm.x86.avx512.ktestc.q(<64 x i1> [[TMP9]], <64 x i1> [[TMP16]])
+; CHECK-NEXT:    store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %1 = icmp ne <64 x i8> %A, zeroinitializer
+  %2 = icmp ne <64 x i8> %B, zeroinitializer
+  %res = call i32 @llvm.x86.avx512.ktestc.q(<64 x i1> %1, <64 x i1> %2) ; <i32> [#uses=1]
+  ret i32 %res
+}
+declare i32 @llvm.x86.avx512.ktestc.q(<64 x i1>, <64 x i1>) nounwind readnone
+
+define i32 @test_x86_avx512_ktestz_q(<64 x i8> %A, <64 x i8> %B) #0 {
+; CHECK-LABEL: @test_x86_avx512_ktestz_q(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <64 x i8> [[A:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = or <64 x i8> [[TMP1]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <64 x i8> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[TMP4]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <64 x i8> [[TMP6]], [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP:%.*]] = and <64 x i1> [[TMP5]], [[TMP8]]
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <64 x i8> [[A]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <64 x i8> [[B:%.*]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP11]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP11]], splat (i8 -1)
+; CHECK-NEXT:    [[TMP14:%.*]] = and <64 x i8> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq <64 x i8> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_ICMP1:%.*]] = and <64 x i1> [[TMP12]], [[TMP15]]
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <64 x i8> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP]] to i64
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i1> [[_MSPROP_ICMP1]] to i64
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[RES:%.*]] = call i32 @llvm.x86.avx512.ktestz.q(<64 x i1> [[TMP9]], <64 x i1> [[TMP16]])
+; CHECK-NEXT:    store i32 0, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret i32 [[RES]]
+;
+  %1 = icmp ne <64 x i8> %A, zeroinitializer
+  %2 = icmp ne <64 x i8> %B, zeroinitializer
+  %res = call i32 @llvm.x86.avx512.ktestz.q(<64 x i1> %1, <64 x i1> %2) ; <i32> [#uses=1]
+  ret i32 %res
+}
+declare i32 @llvm.x86.avx512.ktestz.q(<64 x i1>, <64 x i1>) nounwind readnone
+
+define <32 x i16> @test_mask_packs_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packs_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packs_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i16> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP9]], <32 x i16> [[TMP14]], <32 x i16> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP8]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP15]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packs_epi32_rm_512(<16 x i32> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmbk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi32_rmbkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32>, <16 x i32>)
+
+define <64 x i8> @test_mask_packs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP7]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @test_mask_packs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <64 x i8> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP9]], <64 x i8> [[TMP14]], <64 x i8> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP8]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP15]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packs_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @test_mask_packs_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <64 x i8> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <64 x i8> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP20]], <64 x i8> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> [[TMP14]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP21]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packs_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packs_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <64 x i8> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP14]], <64 x i8> [[TMP19]], <64 x i8> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP13]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP20]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
+  ret <64 x i8> %3
+}
+
+declare <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16>, <32 x i16>)
+
+
+define <32 x i16> @test_mask_packus_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packus_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packus_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <16 x i32> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i16> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP9]], <32 x i16> [[TMP14]], <32 x i16> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP8]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP15]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packus_epi32_rm_512(<16 x i32> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <16 x i32>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %b = load <16 x i32>, ptr %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32 x i16> %passThru, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmbk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <32 x i16> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <32 x i16> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP20]], <32 x i16> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <32 x i1> [[TMP16]], <32 x i16> [[TMP14]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i32 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi32_rmbkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[Q:%.*]] = load i32, ptr [[PTR_B:%.*]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
+; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> [[A:%.*]], <16 x i32> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <32 x i16> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <32 x i16> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP14]], <32 x i16> [[TMP19]], <32 x i16> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <32 x i1> [[TMP15]], <32 x i16> [[TMP13]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
+;
+  %q = load i32, ptr %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32>, <16 x i32>)
+
+define <64 x i8> @test_mask_packus_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rr_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP7]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @test_mask_packus_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rrk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <64 x i8> [[TMP9]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <64 x i8> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP15]], <64 x i8> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP11]], <64 x i8> [[TMP9]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP16]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packus_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rrkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[TMP8:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = xor <64 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = or <64 x i8> [[TMP12]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP9]], <64 x i8> [[TMP14]], <64 x i8> [[TMP11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = select <64 x i1> [[TMP10]], <64 x i8> [[TMP8]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP15]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packus_epi16_rm_512(<32 x i16> %a, ptr %ptr_b) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rm_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP7]], align 64
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP8]], 0
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP10:%.*]], label [[TMP11:%.*]], !prof [[PROF1]]
+; CHECK:       10:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       11:
+; CHECK-NEXT:    [[TMP12:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    store <64 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @test_mask_packus_epi16_rmk_512(<32 x i16> %a, ptr %ptr_b, <64 x i8> %passThru, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rmk_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP7:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080
+; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP9]], align 64
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP11]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP12:%.*]], label [[TMP13:%.*]], !prof [[PROF1]]
+; CHECK:       12:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       13:
+; CHECK-NEXT:    [[TMP14:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP17:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> zeroinitializer, <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP18:%.*]] = xor <64 x i8> [[TMP14]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = or <64 x i8> [[TMP19]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP20]], <64 x i8> [[TMP17]]
+; CHECK-NEXT:    [[TMP21:%.*]] = select <64 x i1> [[TMP16]], <64 x i8> [[TMP14]], <64 x i8> [[PASSTHRU]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP21]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
+  ret <64 x i8> %3
+}
+
+define <64 x i8> @test_mask_packus_epi16_rmkz_512(<32 x i16> %a, ptr %ptr_b, i64 %mask) #0 {
+; CHECK-LABEL: @test_mask_packus_epi16_rmkz_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[B:%.*]] = load <32 x i16>, ptr [[PTR_B:%.*]], align 64
+; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PTR_B]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
+; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <32 x i16> [[_MSLD]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP1]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP11:%.*]], label [[TMP12:%.*]], !prof [[PROF1]]
+; CHECK:       11:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       12:
+; CHECK-NEXT:    [[TMP13:%.*]] = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> [[A:%.*]], <32 x i16> [[B]])
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP16:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> zeroinitializer, <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = xor <64 x i8> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = or <64 x i8> [[TMP17]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = or <64 x i8> [[TMP18]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP14]], <64 x i8> [[TMP19]], <64 x i8> [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = select <64 x i1> [[TMP15]], <64 x i8> [[TMP13]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP20]]
+;
+  %b = load <32 x i16>, ptr %ptr_b
+  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i64 %mask to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
+  ret <64 x i8> %3
+}
+
+declare <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16>@test_int_x86_avx512_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = and i16 [[TMP4]], 31
+; CHECK-NEXT:    [[TMP6:%.*]] = or i16 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = and i16 [[TMP7]], 31
+; CHECK-NEXT:    [[TMP9:%.*]] = or i16 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP11:%.*]] = and i16 [[TMP10]], 31
+; CHECK-NEXT:    [[TMP12:%.*]] = or i16 [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP14:%.*]] = and i16 [[TMP13]], 31
+; CHECK-NEXT:    [[TMP15:%.*]] = or i16 [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP17:%.*]] = and i16 [[TMP16]], 31
+; CHECK-NEXT:    [[TMP18:%.*]] = or i16 [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP20:%.*]] = and i16 [[TMP19]], 31
+; CHECK-NEXT:    [[TMP21:%.*]] = or i16 [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP23:%.*]] = and i16 [[TMP22]], 31
+; CHECK-NEXT:    [[TMP24:%.*]] = or i16 [[TMP22]], [[TMP23]]
+; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP26:%.*]] = and i16 [[TMP25]], 31
+; CHECK-NEXT:    [[TMP27:%.*]] = or i16 [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP29:%.*]] = and i16 [[TMP28]], 31
+; CHECK-NEXT:    [[TMP30:%.*]] = or i16 [[TMP28]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP32:%.*]] = and i16 [[TMP31]], 31
+; CHECK-NEXT:    [[TMP33:%.*]] = or i16 [[TMP31]], [[TMP32]]
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP35:%.*]] = and i16 [[TMP34]], 31
+; CHECK-NEXT:    [[TMP36:%.*]] = or i16 [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP38:%.*]] = and i16 [[TMP37]], 31
+; CHECK-NEXT:    [[TMP39:%.*]] = or i16 [[TMP37]], [[TMP38]]
+; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP41:%.*]] = and i16 [[TMP40]], 31
+; CHECK-NEXT:    [[TMP42:%.*]] = or i16 [[TMP40]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP44:%.*]] = and i16 [[TMP43]], 31
+; CHECK-NEXT:    [[TMP45:%.*]] = or i16 [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP47:%.*]] = and i16 [[TMP46]], 31
+; CHECK-NEXT:    [[TMP48:%.*]] = or i16 [[TMP46]], [[TMP47]]
+; CHECK-NEXT:    [[TMP49:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP50:%.*]] = and i16 [[TMP49]], 31
+; CHECK-NEXT:    [[TMP51:%.*]] = or i16 [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP53:%.*]] = and i16 [[TMP52]], 31
+; CHECK-NEXT:    [[TMP54:%.*]] = or i16 [[TMP52]], [[TMP53]]
+; CHECK-NEXT:    [[TMP55:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP56:%.*]] = and i16 [[TMP55]], 31
+; CHECK-NEXT:    [[TMP57:%.*]] = or i16 [[TMP55]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP59:%.*]] = and i16 [[TMP58]], 31
+; CHECK-NEXT:    [[TMP60:%.*]] = or i16 [[TMP58]], [[TMP59]]
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP62:%.*]] = and i16 [[TMP61]], 31
+; CHECK-NEXT:    [[TMP63:%.*]] = or i16 [[TMP61]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP65:%.*]] = and i16 [[TMP64]], 31
+; CHECK-NEXT:    [[TMP66:%.*]] = or i16 [[TMP64]], [[TMP65]]
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP68:%.*]] = and i16 [[TMP67]], 31
+; CHECK-NEXT:    [[TMP69:%.*]] = or i16 [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP71:%.*]] = and i16 [[TMP70]], 31
+; CHECK-NEXT:    [[TMP72:%.*]] = or i16 [[TMP70]], [[TMP71]]
+; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP74:%.*]] = and i16 [[TMP73]], 31
+; CHECK-NEXT:    [[TMP75:%.*]] = or i16 [[TMP73]], [[TMP74]]
+; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP77:%.*]] = and i16 [[TMP76]], 31
+; CHECK-NEXT:    [[TMP78:%.*]] = or i16 [[TMP76]], [[TMP77]]
+; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP80:%.*]] = and i16 [[TMP79]], 31
+; CHECK-NEXT:    [[TMP81:%.*]] = or i16 [[TMP79]], [[TMP80]]
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP83:%.*]] = and i16 [[TMP82]], 31
+; CHECK-NEXT:    [[TMP84:%.*]] = or i16 [[TMP82]], [[TMP83]]
+; CHECK-NEXT:    [[TMP85:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP86:%.*]] = and i16 [[TMP85]], 31
+; CHECK-NEXT:    [[TMP87:%.*]] = or i16 [[TMP85]], [[TMP86]]
+; CHECK-NEXT:    [[TMP88:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP89:%.*]] = and i16 [[TMP88]], 31
+; CHECK-NEXT:    [[TMP90:%.*]] = or i16 [[TMP88]], [[TMP89]]
+; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP92:%.*]] = and i16 [[TMP91]], 31
+; CHECK-NEXT:    [[TMP93:%.*]] = or i16 [[TMP91]], [[TMP92]]
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP95:%.*]] = and i16 [[TMP94]], 31
+; CHECK-NEXT:    [[TMP96:%.*]] = or i16 [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP98:%.*]] = and i16 [[TMP97]], 31
+; CHECK-NEXT:    [[TMP99:%.*]] = or i16 [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP100:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP99]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP101:%.*]], label [[TMP102:%.*]], !prof [[PROF1]]
+; CHECK:       101:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       102:
+; CHECK-NEXT:    [[TMP103:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[TMP100]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP103]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> [[TMP1]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], [[X1]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], [[TMP1]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x1
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_maskz_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_vpermt2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X0:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X1:%.*]], <32 x i16> [[X0]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], zeroinitializer
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>)
+
+define <32 x i16>@test_int_x86_avx512_vpermi2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_vpermi2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = and i16 [[TMP4]], 31
+; CHECK-NEXT:    [[TMP6:%.*]] = or i16 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = and i16 [[TMP7]], 31
+; CHECK-NEXT:    [[TMP9:%.*]] = or i16 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP11:%.*]] = and i16 [[TMP10]], 31
+; CHECK-NEXT:    [[TMP12:%.*]] = or i16 [[TMP10]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP14:%.*]] = and i16 [[TMP13]], 31
+; CHECK-NEXT:    [[TMP15:%.*]] = or i16 [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP17:%.*]] = and i16 [[TMP16]], 31
+; CHECK-NEXT:    [[TMP18:%.*]] = or i16 [[TMP16]], [[TMP17]]
+; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP20:%.*]] = and i16 [[TMP19]], 31
+; CHECK-NEXT:    [[TMP21:%.*]] = or i16 [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP23:%.*]] = and i16 [[TMP22]], 31
+; CHECK-NEXT:    [[TMP24:%.*]] = or i16 [[TMP22]], [[TMP23]]
+; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP26:%.*]] = and i16 [[TMP25]], 31
+; CHECK-NEXT:    [[TMP27:%.*]] = or i16 [[TMP25]], [[TMP26]]
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP29:%.*]] = and i16 [[TMP28]], 31
+; CHECK-NEXT:    [[TMP30:%.*]] = or i16 [[TMP28]], [[TMP29]]
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP32:%.*]] = and i16 [[TMP31]], 31
+; CHECK-NEXT:    [[TMP33:%.*]] = or i16 [[TMP31]], [[TMP32]]
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP35:%.*]] = and i16 [[TMP34]], 31
+; CHECK-NEXT:    [[TMP36:%.*]] = or i16 [[TMP34]], [[TMP35]]
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP38:%.*]] = and i16 [[TMP37]], 31
+; CHECK-NEXT:    [[TMP39:%.*]] = or i16 [[TMP37]], [[TMP38]]
+; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP41:%.*]] = and i16 [[TMP40]], 31
+; CHECK-NEXT:    [[TMP42:%.*]] = or i16 [[TMP40]], [[TMP41]]
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP44:%.*]] = and i16 [[TMP43]], 31
+; CHECK-NEXT:    [[TMP45:%.*]] = or i16 [[TMP43]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP47:%.*]] = and i16 [[TMP46]], 31
+; CHECK-NEXT:    [[TMP48:%.*]] = or i16 [[TMP46]], [[TMP47]]
+; CHECK-NEXT:    [[TMP49:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP50:%.*]] = and i16 [[TMP49]], 31
+; CHECK-NEXT:    [[TMP51:%.*]] = or i16 [[TMP49]], [[TMP50]]
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP53:%.*]] = and i16 [[TMP52]], 31
+; CHECK-NEXT:    [[TMP54:%.*]] = or i16 [[TMP52]], [[TMP53]]
+; CHECK-NEXT:    [[TMP55:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP56:%.*]] = and i16 [[TMP55]], 31
+; CHECK-NEXT:    [[TMP57:%.*]] = or i16 [[TMP55]], [[TMP56]]
+; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP59:%.*]] = and i16 [[TMP58]], 31
+; CHECK-NEXT:    [[TMP60:%.*]] = or i16 [[TMP58]], [[TMP59]]
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP62:%.*]] = and i16 [[TMP61]], 31
+; CHECK-NEXT:    [[TMP63:%.*]] = or i16 [[TMP61]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP65:%.*]] = and i16 [[TMP64]], 31
+; CHECK-NEXT:    [[TMP66:%.*]] = or i16 [[TMP64]], [[TMP65]]
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP68:%.*]] = and i16 [[TMP67]], 31
+; CHECK-NEXT:    [[TMP69:%.*]] = or i16 [[TMP67]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP71:%.*]] = and i16 [[TMP70]], 31
+; CHECK-NEXT:    [[TMP72:%.*]] = or i16 [[TMP70]], [[TMP71]]
+; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP74:%.*]] = and i16 [[TMP73]], 31
+; CHECK-NEXT:    [[TMP75:%.*]] = or i16 [[TMP73]], [[TMP74]]
+; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP77:%.*]] = and i16 [[TMP76]], 31
+; CHECK-NEXT:    [[TMP78:%.*]] = or i16 [[TMP76]], [[TMP77]]
+; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP80:%.*]] = and i16 [[TMP79]], 31
+; CHECK-NEXT:    [[TMP81:%.*]] = or i16 [[TMP79]], [[TMP80]]
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP83:%.*]] = and i16 [[TMP82]], 31
+; CHECK-NEXT:    [[TMP84:%.*]] = or i16 [[TMP82]], [[TMP83]]
+; CHECK-NEXT:    [[TMP85:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP86:%.*]] = and i16 [[TMP85]], 31
+; CHECK-NEXT:    [[TMP87:%.*]] = or i16 [[TMP85]], [[TMP86]]
+; CHECK-NEXT:    [[TMP88:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP89:%.*]] = and i16 [[TMP88]], 31
+; CHECK-NEXT:    [[TMP90:%.*]] = or i16 [[TMP88]], [[TMP89]]
+; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP92:%.*]] = and i16 [[TMP91]], 31
+; CHECK-NEXT:    [[TMP93:%.*]] = or i16 [[TMP91]], [[TMP92]]
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP95:%.*]] = and i16 [[TMP94]], 31
+; CHECK-NEXT:    [[TMP96:%.*]] = or i16 [[TMP94]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP98:%.*]] = and i16 [[TMP97]], 31
+; CHECK-NEXT:    [[TMP99:%.*]] = or i16 [[TMP97]], [[TMP98]]
+; CHECK-NEXT:    [[TMP100:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X1:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP99]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP101:%.*]], label [[TMP102:%.*]], !prof [[PROF1]]
+; CHECK:       101:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       102:
+; CHECK-NEXT:    [[TMP103:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[TMP100]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP103]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_vpermi2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_vpermi2var_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <32 x i16> [[TMP3]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = and i16 [[TMP5]], 31
+; CHECK-NEXT:    [[TMP7:%.*]] = or i16 [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <32 x i16> [[TMP3]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = and i16 [[TMP8]], 31
+; CHECK-NEXT:    [[TMP10:%.*]] = or i16 [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <32 x i16> [[TMP3]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = and i16 [[TMP11]], 31
+; CHECK-NEXT:    [[TMP13:%.*]] = or i16 [[TMP11]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <32 x i16> [[TMP3]], i64 3
+; CHECK-NEXT:    [[TMP15:%.*]] = and i16 [[TMP14]], 31
+; CHECK-NEXT:    [[TMP16:%.*]] = or i16 [[TMP14]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <32 x i16> [[TMP3]], i64 4
+; CHECK-NEXT:    [[TMP18:%.*]] = and i16 [[TMP17]], 31
+; CHECK-NEXT:    [[TMP19:%.*]] = or i16 [[TMP17]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <32 x i16> [[TMP3]], i64 5
+; CHECK-NEXT:    [[TMP21:%.*]] = and i16 [[TMP20]], 31
+; CHECK-NEXT:    [[TMP22:%.*]] = or i16 [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <32 x i16> [[TMP3]], i64 6
+; CHECK-NEXT:    [[TMP24:%.*]] = and i16 [[TMP23]], 31
+; CHECK-NEXT:    [[TMP25:%.*]] = or i16 [[TMP23]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <32 x i16> [[TMP3]], i64 7
+; CHECK-NEXT:    [[TMP27:%.*]] = and i16 [[TMP26]], 31
+; CHECK-NEXT:    [[TMP28:%.*]] = or i16 [[TMP26]], [[TMP27]]
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <32 x i16> [[TMP3]], i64 8
+; CHECK-NEXT:    [[TMP30:%.*]] = and i16 [[TMP29]], 31
+; CHECK-NEXT:    [[TMP31:%.*]] = or i16 [[TMP29]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <32 x i16> [[TMP3]], i64 9
+; CHECK-NEXT:    [[TMP33:%.*]] = and i16 [[TMP32]], 31
+; CHECK-NEXT:    [[TMP34:%.*]] = or i16 [[TMP32]], [[TMP33]]
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <32 x i16> [[TMP3]], i64 10
+; CHECK-NEXT:    [[TMP36:%.*]] = and i16 [[TMP35]], 31
+; CHECK-NEXT:    [[TMP37:%.*]] = or i16 [[TMP35]], [[TMP36]]
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <32 x i16> [[TMP3]], i64 11
+; CHECK-NEXT:    [[TMP39:%.*]] = and i16 [[TMP38]], 31
+; CHECK-NEXT:    [[TMP40:%.*]] = or i16 [[TMP38]], [[TMP39]]
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <32 x i16> [[TMP3]], i64 12
+; CHECK-NEXT:    [[TMP42:%.*]] = and i16 [[TMP41]], 31
+; CHECK-NEXT:    [[TMP43:%.*]] = or i16 [[TMP41]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <32 x i16> [[TMP3]], i64 13
+; CHECK-NEXT:    [[TMP45:%.*]] = and i16 [[TMP44]], 31
+; CHECK-NEXT:    [[TMP46:%.*]] = or i16 [[TMP44]], [[TMP45]]
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <32 x i16> [[TMP3]], i64 14
+; CHECK-NEXT:    [[TMP48:%.*]] = and i16 [[TMP47]], 31
+; CHECK-NEXT:    [[TMP49:%.*]] = or i16 [[TMP47]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <32 x i16> [[TMP3]], i64 15
+; CHECK-NEXT:    [[TMP51:%.*]] = and i16 [[TMP50]], 31
+; CHECK-NEXT:    [[TMP52:%.*]] = or i16 [[TMP50]], [[TMP51]]
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <32 x i16> [[TMP3]], i64 16
+; CHECK-NEXT:    [[TMP54:%.*]] = and i16 [[TMP53]], 31
+; CHECK-NEXT:    [[TMP55:%.*]] = or i16 [[TMP53]], [[TMP54]]
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <32 x i16> [[TMP3]], i64 17
+; CHECK-NEXT:    [[TMP57:%.*]] = and i16 [[TMP56]], 31
+; CHECK-NEXT:    [[TMP58:%.*]] = or i16 [[TMP56]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <32 x i16> [[TMP3]], i64 18
+; CHECK-NEXT:    [[TMP60:%.*]] = and i16 [[TMP59]], 31
+; CHECK-NEXT:    [[TMP61:%.*]] = or i16 [[TMP59]], [[TMP60]]
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <32 x i16> [[TMP3]], i64 19
+; CHECK-NEXT:    [[TMP63:%.*]] = and i16 [[TMP62]], 31
+; CHECK-NEXT:    [[TMP64:%.*]] = or i16 [[TMP62]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <32 x i16> [[TMP3]], i64 20
+; CHECK-NEXT:    [[TMP66:%.*]] = and i16 [[TMP65]], 31
+; CHECK-NEXT:    [[TMP67:%.*]] = or i16 [[TMP65]], [[TMP66]]
+; CHECK-NEXT:    [[TMP68:%.*]] = extractelement <32 x i16> [[TMP3]], i64 21
+; CHECK-NEXT:    [[TMP69:%.*]] = and i16 [[TMP68]], 31
+; CHECK-NEXT:    [[TMP70:%.*]] = or i16 [[TMP68]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <32 x i16> [[TMP3]], i64 22
+; CHECK-NEXT:    [[TMP72:%.*]] = and i16 [[TMP71]], 31
+; CHECK-NEXT:    [[TMP73:%.*]] = or i16 [[TMP71]], [[TMP72]]
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <32 x i16> [[TMP3]], i64 23
+; CHECK-NEXT:    [[TMP75:%.*]] = and i16 [[TMP74]], 31
+; CHECK-NEXT:    [[TMP76:%.*]] = or i16 [[TMP74]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <32 x i16> [[TMP3]], i64 24
+; CHECK-NEXT:    [[TMP78:%.*]] = and i16 [[TMP77]], 31
+; CHECK-NEXT:    [[TMP79:%.*]] = or i16 [[TMP77]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = extractelement <32 x i16> [[TMP3]], i64 25
+; CHECK-NEXT:    [[TMP81:%.*]] = and i16 [[TMP80]], 31
+; CHECK-NEXT:    [[TMP82:%.*]] = or i16 [[TMP80]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <32 x i16> [[TMP3]], i64 26
+; CHECK-NEXT:    [[TMP84:%.*]] = and i16 [[TMP83]], 31
+; CHECK-NEXT:    [[TMP85:%.*]] = or i16 [[TMP83]], [[TMP84]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <32 x i16> [[TMP3]], i64 27
+; CHECK-NEXT:    [[TMP87:%.*]] = and i16 [[TMP86]], 31
+; CHECK-NEXT:    [[TMP88:%.*]] = or i16 [[TMP86]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <32 x i16> [[TMP3]], i64 28
+; CHECK-NEXT:    [[TMP90:%.*]] = and i16 [[TMP89]], 31
+; CHECK-NEXT:    [[TMP91:%.*]] = or i16 [[TMP89]], [[TMP90]]
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <32 x i16> [[TMP3]], i64 29
+; CHECK-NEXT:    [[TMP93:%.*]] = and i16 [[TMP92]], 31
+; CHECK-NEXT:    [[TMP94:%.*]] = or i16 [[TMP92]], [[TMP93]]
+; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <32 x i16> [[TMP3]], i64 30
+; CHECK-NEXT:    [[TMP96:%.*]] = and i16 [[TMP95]], 31
+; CHECK-NEXT:    [[TMP97:%.*]] = or i16 [[TMP95]], [[TMP96]]
+; CHECK-NEXT:    [[TMP98:%.*]] = extractelement <32 x i16> [[TMP3]], i64 31
+; CHECK-NEXT:    [[TMP99:%.*]] = and i16 [[TMP98]], 31
+; CHECK-NEXT:    [[TMP100:%.*]] = or i16 [[TMP98]], [[TMP99]]
+; CHECK-NEXT:    [[TMP101:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[TMP1]], <32 x i16> [[X1:%.*]], <32 x i16> [[TMP2]])
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i16 [[TMP100]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP102:%.*]], label [[TMP103:%.*]], !prof [[PROF1]]
+; CHECK:       102:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       103:
+; CHECK-NEXT:    [[TMP104:%.*]] = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1]], <32 x i16> [[X2:%.*]])
+; CHECK-NEXT:    [[TMP105:%.*]] = bitcast i32 [[TMP4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP106:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP107:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP101]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP108:%.*]] = xor <32 x i16> [[TMP104]], [[X1]]
+; CHECK-NEXT:    [[TMP109:%.*]] = or <32 x i16> [[TMP108]], [[TMP101]]
+; CHECK-NEXT:    [[TMP110:%.*]] = or <32 x i16> [[TMP109]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP105]], <32 x i16> [[TMP110]], <32 x i16> [[TMP107]]
+; CHECK-NEXT:    [[TMP111:%.*]] = select <32 x i1> [[TMP106]], <32 x i16> [[TMP104]], <32 x i16> [[X1]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP111]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x1
+  ret <32 x i16> %3
+}
+
+declare <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8>, <64 x i8>)
+
+define <64 x i8> @test_int_x86_avx512_pavg_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pavg_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP3]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @test_int_x86_avx512_mask_pavg_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pavg_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i64 [[X3:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <64 x i8> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <64 x i8> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <64 x i8> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP6]], <64 x i8> [[TMP11]], <64 x i8> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <64 x i1> [[TMP7]], <64 x i8> [[TMP5]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[TMP12]]
+;
+  %1 = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1)
+  %2 = bitcast i64 %x3 to <64 x i1>
+  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %x2
+  ret <64 x i8> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_int_x86_avx512_pavg_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pavg_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pavg_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pavg_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+declare <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8>, <64 x i8>)
+
+define <64 x i8>@test_int_x86_avx512_pshuf_b_512(<64 x i8> %x0, <64 x i8> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshuf_b_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[RES:%.*]] = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[RES]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
+  ret <64 x i8> %res
+}
+
+define <64 x i8>@test_int_x86_avx512_pshuf_b_512_mask(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %mask) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshuf_b_512_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[RES:%.*]] = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <64 x i1> [[MASK_CAST]], <64 x i8> [[_MSPROP]], <64 x i8> [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <64 x i8> [[RES]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <64 x i8> [[TMP8]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP5]], <64 x i8> [[TMP9]], <64 x i8> [[TMP6]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <64 x i1> [[MASK_CAST]], <64 x i8> [[RES]], <64 x i8> [[X2]]
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[RES2]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
+  %mask.cast = bitcast i64 %mask to <64 x i1>
+  %res2 = select <64 x i1> %mask.cast, <64 x i8> %res, <64 x i8> %x2
+  ret <64 x i8> %res2
+}
+
+define <64 x i8>@test_int_x86_avx512_pshuf_b_512_maskz(<64 x i8> %x0, <64 x i8> %x1, i64 %mask) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pshuf_b_512_maskz(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <64 x i8> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[RES:%.*]] = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i64 [[TMP3]] to <64 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i64 [[MASK:%.*]] to <64 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = select <64 x i1> [[MASK_CAST]], <64 x i8> [[_MSPROP]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <64 x i8> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = or <64 x i8> [[TMP6]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <64 x i8> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <64 x i1> [[TMP4]], <64 x i8> [[TMP8]], <64 x i8> [[TMP5]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <64 x i1> [[MASK_CAST]], <64 x i8> [[RES]], <64 x i8> zeroinitializer
+; CHECK-NEXT:    store <64 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <64 x i8> [[RES2]]
+;
+  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
+  %mask.cast = bitcast i64 %mask to <64 x i1>
+  %res2 = select <64 x i1> %mask.cast, <64 x i8> %res, <64 x i8> zeroinitializer
+  ret <64 x i8> %res2
+}
+
+declare <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_int_x86_avx512_pmulhu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulhu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulhu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulhu_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_int_x86_avx512_pmulh_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulh_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulh_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_int_x86_avx512_pmulhr_sw_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmulhr_sw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmulhr_sw_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmulhr_sw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+define <32 x i8>@test_int_x86_avx512_pmov_wb_512(<32 x i16> %x0) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP2]]
+;
+  %1 = trunc <32 x i16> %x0 to <32 x i8>
+  ret <32 x i8> %1
+}
+
+define <32 x i8>@test_int_x86_avx512_mask_pmov_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i8> [[_MSPROP]], <32 x i8> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i8> [[TMP4]], [[X1:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i8> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i8> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[TMP10]], <32 x i8> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i8> [[TMP4]], <32 x i8> [[X1]]
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP11]]
+;
+  %1 = trunc <32 x i16> %x0 to <32 x i8>
+  %2 = bitcast i32 %x2 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %x1
+  ret <32 x i8> %3
+}
+
+define <32 x i8>@test_int_x86_avx512_maskz_pmov_wb_512(<32 x i16> %x0, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmov_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = trunc <32 x i16> [[TMP1]] to <32 x i8>
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc <32 x i16> [[X0:%.*]] to <32 x i8>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[X2:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[_MSPROP]], <32 x i8> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i8> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i8> [[TMP7]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i8> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP4]], <32 x i8> [[TMP9]], <32 x i8> [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = select <32 x i1> [[TMP5]], <32 x i8> [[TMP3]], <32 x i8> zeroinitializer
+; CHECK-NEXT:    store <32 x i8> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[TMP10]]
+;
+  %1 = trunc <32 x i16> %x0 to <32 x i8>
+  %2 = bitcast i32 %x2 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
+declare void @llvm.x86.avx512.mask.pmov.wb.mem.512(ptr %ptr, <32 x i16>, i32)
+
+define void @test_int_x86_avx512_mask_pmov_wb_mem_512(ptr %ptr, <32 x i16> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmov_wb_mem_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmov.wb.mem.512(ptr [[PTR:%.*]], <32 x i16> [[X1:%.*]], i32 -1)
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP7]], 0
+; CHECK-NEXT:    [[_MSOR4:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    [[_MSCMP5:%.*]] = icmp ne i32 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSOR6:%.*]] = or i1 [[_MSOR4]], [[_MSCMP5]]
+; CHECK-NEXT:    br i1 [[_MSOR6]], label [[TMP8:%.*]], label [[TMP9:%.*]], !prof [[PROF1]]
+; CHECK:       8:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       9:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmov.wb.mem.512(ptr [[PTR]], <32 x i16> [[X1]], i32 [[X2:%.*]])
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.x86.avx512.mask.pmov.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 -1)
+  call void @llvm.x86.avx512.mask.pmov.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 %x2)
+  ret void
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16>, <32 x i8>, i32)
+
+define <32 x i8>@test_int_x86_avx512_pmovs_wb_512(<32 x i16> %x0, <32 x i8> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmovs_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i8> [[TMP2]] to i256
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i256 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> [[X1:%.*]], i32 -1)
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 -1)
+  ret <32 x i8> %res
+}
+
+define <32 x i8>@test_int_x86_avx512_mask_pmovs_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovs_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i8> [[TMP2]] to i256
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i256 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i32 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSOR3:%.*]] = or i1 [[_MSOR]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR3]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> [[X1:%.*]], i32 [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2)
+  ret <32 x i8> %res
+}
+
+define <32 x i8>@test_int_x86_avx512_maskz_pmovs_wb_512(<32 x i16> %x0, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmovs_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i32 [[TMP2]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> zeroinitializer, i32 [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> zeroinitializer, i32 %x2)
+  ret <32 x i8> %res
+}
+
+declare void @llvm.x86.avx512.mask.pmovs.wb.mem.512(ptr %ptr, <32 x i16>, i32)
+
+define void @test_int_x86_avx512_mask_pmovs_wb_mem_512(ptr %ptr, <32 x i16> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovs_wb_mem_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(ptr [[PTR:%.*]], <32 x i16> [[X1:%.*]], i32 -1)
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP7]], 0
+; CHECK-NEXT:    [[_MSOR4:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    [[_MSCMP5:%.*]] = icmp ne i32 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSOR6:%.*]] = or i1 [[_MSOR4]], [[_MSCMP5]]
+; CHECK-NEXT:    br i1 [[_MSOR6]], label [[TMP8:%.*]], label [[TMP9:%.*]], !prof [[PROF1]]
+; CHECK:       8:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       9:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(ptr [[PTR]], <32 x i16> [[X1]], i32 [[X2:%.*]])
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 -1)
+  call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 %x2)
+  ret void
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16>, <32 x i8>, i32)
+
+define <32 x i8>@test_int_x86_avx512_pmovus_wb_512(<32 x i16> %x0, <32 x i8> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmovus_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i8> [[TMP2]] to i256
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i256 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> [[X1:%.*]], i32 -1)
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 -1)
+  ret <32 x i8> %res
+}
+
+define <32 x i8>@test_int_x86_avx512_mask_pmovus_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovus_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 96) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i8> [[TMP2]] to i256
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i256 [[TMP5]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i32 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSOR3:%.*]] = or i1 [[_MSOR]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR3]], label [[TMP6:%.*]], label [[TMP7:%.*]], !prof [[PROF1]]
+; CHECK:       6:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       7:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> [[X1:%.*]], i32 [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2)
+  ret <32 x i8> %res
+}
+
+define <32 x i8>@test_int_x86_avx512_maskz_pmovus_wb_512(<32 x i16> %x0, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_pmovus_wb_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i32 [[TMP2]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP4:%.*]], label [[TMP5:%.*]], !prof [[PROF1]]
+; CHECK:       4:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       5:
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> [[X0:%.*]], <32 x i8> zeroinitializer, i32 [[X2:%.*]])
+; CHECK-NEXT:    store <32 x i8> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i8> [[RES]]
+;
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> zeroinitializer, i32 %x2)
+  ret <32 x i8> %res
+}
+
+declare void @llvm.x86.avx512.mask.pmovus.wb.mem.512(ptr %ptr, <32 x i16>, i32)
+
+define void @test_int_x86_avx512_mask_pmovus_wb_mem_512(ptr %ptr, <32 x i16> %x1, i32 %x2) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmovus_wb_mem_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(ptr [[PTR:%.*]], <32 x i16> [[X1:%.*]], i32 -1)
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP7]], 0
+; CHECK-NEXT:    [[_MSOR4:%.*]] = or i1 [[_MSCMP2]], [[_MSCMP3]]
+; CHECK-NEXT:    [[_MSCMP5:%.*]] = icmp ne i32 [[TMP3]], 0
+; CHECK-NEXT:    [[_MSOR6:%.*]] = or i1 [[_MSOR4]], [[_MSCMP5]]
+; CHECK-NEXT:    br i1 [[_MSOR6]], label [[TMP8:%.*]], label [[TMP9:%.*]], !prof [[PROF1]]
+; CHECK:       8:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       9:
+; CHECK-NEXT:    call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(ptr [[PTR]], <32 x i16> [[X1]], i32 [[X2:%.*]])
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 -1)
+  call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(ptr %ptr, <32 x i16> %x1, i32 %x2)
+  ret void
+}
+
+declare <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8>, <64 x i8>)
+
+define <32 x i16> @test_int_x86_avx512_pmaddubs_w_512(<64 x i8> %x0, <64 x i8> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaddubs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP7]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_int_x86_avx512_mask_pmaddubs_w_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaddubs_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP16]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+declare <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16>, <32 x i16>)
+
+define <16 x i32> @test_int_x86_avx512_pmaddw_d_512(<32 x i16> %x0, <32 x i16> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_pmaddw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP3]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP4]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP5:%.*]], label [[TMP6:%.*]], !prof [[PROF1]]
+; CHECK:       5:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       6:
+; CHECK-NEXT:    [[TMP7:%.*]] = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <16 x i32> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <16 x i32> [[TMP7]]
+;
+  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <16 x i32> %1
+}
+
+define <16 x i32> @test_int_x86_avx512_mask_pmaddw_d_512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2, i16 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaddw_d_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i16, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <32 x i16> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <32 x i16> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP1:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP1]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i16 [[TMP3]] to <16 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i16 [[X3:%.*]] to <16 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> zeroinitializer, <16 x i32> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <16 x i32> [[TMP9]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <16 x i32> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <16 x i32> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP10]], <16 x i32> [[TMP15]], <16 x i32> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <16 x i1> [[TMP11]], <16 x i32> [[TMP9]], <16 x i32> [[X2]]
+; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <16 x i32> [[TMP16]]
+;
+  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i16 %x3 to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x2
+  ret <16 x i32> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8>, <64 x i8>, i32)
+
+define { <32 x i16>, <32 x i16>, <32 x i16> } @test_int_x86_avx512_mask_dbpsadbw_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x3, i32 %x4) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_dbpsadbw_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i512 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP6]], 0
+; CHECK-NEXT:    [[_MSOR:%.*]] = or i1 [[_MSCMP]], [[_MSCMP2]]
+; CHECK-NEXT:    br i1 [[_MSOR]], label [[TMP7:%.*]], label [[TMP8:%.*]], !prof [[PROF1]]
+; CHECK:       7:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       8:
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]], i32 2)
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[X4:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> zeroinitializer, <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[TMP9]], [[X3:%.*]]
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP10]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP9]], <32 x i16> [[X3]]
+; CHECK-NEXT:    [[TMP17:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP3:%.*]] = icmp ne i512 [[TMP17]], 0
+; CHECK-NEXT:    [[TMP18:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP4:%.*]] = icmp ne i512 [[TMP18]], 0
+; CHECK-NEXT:    [[_MSOR5:%.*]] = or i1 [[_MSCMP3]], [[_MSCMP4]]
+; CHECK-NEXT:    br i1 [[_MSOR5]], label [[TMP19:%.*]], label [[TMP20:%.*]], !prof [[PROF1]]
+; CHECK:       19:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       20:
+; CHECK-NEXT:    [[TMP21:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0]], <64 x i8> [[X1]], i32 3)
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP23:%.*]] = bitcast i32 [[X4]] to <32 x i1>
+; CHECK-NEXT:    [[TMP24:%.*]] = select <32 x i1> [[TMP23]], <32 x i16> zeroinitializer, <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP25:%.*]] = xor <32 x i16> [[TMP21]], zeroinitializer
+; CHECK-NEXT:    [[TMP26:%.*]] = or <32 x i16> [[TMP25]], zeroinitializer
+; CHECK-NEXT:    [[TMP27:%.*]] = or <32 x i16> [[TMP26]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT1:%.*]] = select <32 x i1> [[TMP22]], <32 x i16> [[TMP27]], <32 x i16> [[TMP24]]
+; CHECK-NEXT:    [[TMP28:%.*]] = select <32 x i1> [[TMP23]], <32 x i16> [[TMP21]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <64 x i8> [[TMP1]] to i512
+; CHECK-NEXT:    [[_MSCMP6:%.*]] = icmp ne i512 [[TMP29]], 0
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <64 x i8> [[TMP2]] to i512
+; CHECK-NEXT:    [[_MSCMP7:%.*]] = icmp ne i512 [[TMP30]], 0
+; CHECK-NEXT:    [[_MSOR8:%.*]] = or i1 [[_MSCMP6]], [[_MSCMP7]]
+; CHECK-NEXT:    br i1 [[_MSOR8]], label [[TMP31:%.*]], label [[TMP32:%.*]], !prof [[PROF1]]
+; CHECK:       31:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       32:
+; CHECK-NEXT:    [[TMP33:%.*]] = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> [[X0]], <64 x i8> [[X1]], i32 4)
+; CHECK-NEXT:    [[TMP34:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } { <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1), <32 x i16> splat (i16 -1) }, <32 x i16> [[_MSPROP_SELECT]], 0
+; CHECK-NEXT:    [[RES1:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> [[TMP16]], 0
+; CHECK-NEXT:    [[TMP35:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP34]], <32 x i16> [[_MSPROP_SELECT1]], 1
+; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES1]], <32 x i16> [[TMP28]], 1
+; CHECK-NEXT:    [[TMP36:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP35]], <32 x i16> zeroinitializer, 2
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } [[RES2]], <32 x i16> [[TMP33]], 2
+; CHECK-NEXT:    store { <32 x i16>, <32 x i16>, <32 x i16> } [[TMP36]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <32 x i16>, <32 x i16>, <32 x i16> } [[RES3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2)
+  %2 = bitcast i32 %x4 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x3
+  %4 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 3)
+  %5 = bitcast i32 %x4 to <32 x i1>
+  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
+  %7 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 4)
+  %res1 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> } poison, <32 x i16> %3, 0
+  %res2 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res1, <32 x i16> %6, 1
+  %res3 = insertvalue { <32 x i16>, <32 x i16>, <32 x i16> }  %res2, <32 x i16> %7, 2
+  ret { <32 x i16>, <32 x i16>, <32 x i16> } %res3
+}
+
+declare  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8>, <64 x i8>)
+
+define { <8 x i64>, <8 x i64> } @test_int_x86_avx512_mask_psadb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2){
+; CHECK-LABEL: @test_int_x86_avx512_mask_psadb_w_512(
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[RES0:%.*]] = call <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> [[X0:%.*]], <64 x i8> [[X1:%.*]])
+; CHECK-NEXT:    [[RES1:%.*]] = call <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> [[X0]], <64 x i8> [[X2:%.*]])
+; CHECK-NEXT:    [[RES2:%.*]] = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64> [[RES0]], 0
+; CHECK-NEXT:    [[RES3:%.*]] = insertvalue { <8 x i64>, <8 x i64> } [[RES2]], <8 x i64> [[RES1]], 1
+; CHECK-NEXT:    store { <8 x i64>, <8 x i64> } zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret { <8 x i64>, <8 x i64> } [[RES3]]
+;
+  %res0 = call  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> %x0, <64 x i8> %x1)
+  %res1 = call  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> %x0, <64 x i8> %x2)
+  %res2 = insertvalue { <8 x i64>, <8 x i64> } poison, <8 x i64> %res0, 0
+  %res3 = insertvalue { <8 x i64>, <8 x i64> }  %res2, <8 x i64> %res1, 1
+  ret { <8 x i64>, <8 x i64> } %res3
+}
+
+declare <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16>, <32 x i16>) nounwind readnone
+
+define <32 x i16> @test_x86_avx512_psrlv_w_512_const() optsize #0 {
+; CHECK-LABEL: @test_x86_avx512_psrlv_w_512_const(
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[RES1:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 -1>)
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES1]]
+;
+  %res1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,  i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 -1>)
+  ret <32 x i16> %res1
+}
+
+define <32 x i16>@test_int_x86_avx512_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_maskz_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psrlv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16>@test_int_x86_avx512_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_maskz_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psrav32_hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_psrav32_hi_const(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psrav32_hi_const(
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> <i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51>, <32 x i16> <i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49>)
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> <i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51>, <32 x i16> <i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49>)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_psllv32hi(<32 x i16> %x0, <32 x i16> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_maskz_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_psllv32hi(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16>, <32 x i16>)
+
+define <32 x i16>@test_int_x86_avx512_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP3]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
+  ret <32 x i16> %1
+}
+
+define <32 x i16>@test_int_x86_avx512_mask_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_mask_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 192) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP8:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[_MSPROP]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <32 x i16> [[TMP5]], [[X2:%.*]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP11]], <32 x i16> [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[TMP7]], <32 x i16> [[TMP5]], <32 x i16> [[X2]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
+  ret <32 x i16> %3
+}
+
+define <32 x i16>@test_int_x86_avx512_maskz_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x3) #0 {
+; CHECK-LABEL: @test_int_x86_avx512_maskz_permvar_hi_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSPROP:%.*]] = or <32 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> [[X0:%.*]], <32 x i16> [[X1:%.*]])
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[X3:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[_MSPROP]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[_MSPROP]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[TMP11]]
+;
+  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
+  %2 = bitcast i32 %x3 to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_x86_avx512_psll_w_512(<32 x i16> %a0, <8 x i16> %a1) #0 {
+; CHECK-LABEL: @test_x86_avx512_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_psll_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP11]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_psll_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_psll_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16>, <8 x i16>) nounwind readnone
+
+
+define <32 x i16> @test_x86_avx512_psllv_w_512_const() optsize #0 {
+; CHECK-LABEL: @test_x86_avx512_psllv_w_512_const(
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[RES1:%.*]] = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 -1>)
+; CHECK-NEXT:    store <32 x i16> zeroinitializer, ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES1]]
+;
+  %res1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4,  i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,  i16 1, i16 1, i16 -1>)
+  ret <32 x i16> %res1
+}
+declare <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16>, <32 x i16>) nounwind readnone
+
+define <32 x i16> @test_x86_avx512_pslli_w_512(<32 x i16> %a0) #0 {
+; CHECK-LABEL: @test_x86_avx512_pslli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP3:%.*]] = or <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    store <32 x i16> [[TMP3]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_pslli_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_pslli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_pslli_w_512(<32 x i16> %a0, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_pslli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16>, i32) nounwind readnone
+
+
+define <32 x i16> @test_x86_avx512_psra_w_512(<32 x i16> %a0, <8 x i16> %a1) #0 {
+; CHECK-LABEL: @test_x86_avx512_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_psra_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP11]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_psra_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_psra_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16>, <8 x i16>) nounwind readnone
+
+
+define <32 x i16> @test_x86_avx512_psrai_w_512(<32 x i16> %a0) #0 {
+; CHECK-LABEL: @test_x86_avx512_psrai_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP3:%.*]] = or <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    store <32 x i16> [[TMP3]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_psrai_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_psrai_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_psrai_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_psrai_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16>, i32) nounwind readnone
+
+
+define <32 x i16> @test_x86_avx512_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1) #0 {
+; CHECK-LABEL: @test_x86_avx512_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i128 [[TMP3]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne i64 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = sext i1 [[TMP5]] to i512
+; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i512 [[TMP6]] to <32 x i16>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP7]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP9]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 144) to ptr), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i128 [[TMP5]] to i64
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne i64 [[TMP6]], 0
+; CHECK-NEXT:    [[TMP8:%.*]] = sext i1 [[TMP7]] to i512
+; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i512 [[TMP8]] to <32 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP11:%.*]] = or <32 x i16> [[TMP10]], [[TMP9]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP13:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP11]], <32 x i16> [[TMP4]]
+; CHECK-NEXT:    [[TMP14:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], [[TMP11]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i16> [[TMP15]], [[TMP4]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP12]], <32 x i16> [[TMP16]], <32 x i16> [[TMP13]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_psrl_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 80) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i16> [[TMP2]] to i128
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP4]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne i64 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i1 [[TMP6]] to i512
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i512 [[TMP7]] to <32 x i16>
+; CHECK-NEXT:    [[TMP9:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP1]], <8 x i16> [[A1:%.*]])
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP8]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP3]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP12:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP10]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = or <32 x i16> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP11]], <32 x i16> [[TMP15]], <32 x i16> [[TMP12]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16>, <8 x i16>) nounwind readnone
+
+define <32 x i16> @test_x86_avx512_psrl_w_512_load(<32 x i16> %a0, ptr %p) #0 {
+; CHECK-LABEL: @test_x86_avx512_psrl_w_512_load(
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[_MSCMP]], label [[TMP3:%.*]], label [[TMP4:%.*]], !prof [[PROF1]]
+; CHECK:       3:
+; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR8]]
+; CHECK-NEXT:    unreachable
+; CHECK:       4:
+; CHECK-NEXT:    [[A1:%.*]] = load <8 x i16>, ptr [[P:%.*]], align 16
+; CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[P]] to i64
+; CHECK-NEXT:    [[TMP6:%.*]] = xor i64 [[TMP5]], 87960930222080
+; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i16>, ptr [[TMP7]], align 16
+; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <8 x i16> [[_MSLD]] to i128
+; CHECK-NEXT:    [[TMP9:%.*]] = trunc i128 [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne i64 [[TMP9]], 0
+; CHECK-NEXT:    [[TMP11:%.*]] = sext i1 [[TMP10]] to i512
+; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i512 [[TMP11]] to <32 x i16>
+; CHECK-NEXT:    [[TMP13:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[TMP2]], <8 x i16> [[A1]])
+; CHECK-NEXT:    [[TMP14:%.*]] = or <32 x i16> [[TMP13]], [[TMP12]]
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> [[A0:%.*]], <8 x i16> [[A1]])
+; CHECK-NEXT:    store <32 x i16> [[TMP14]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %a1 = load <8 x i16>, ptr %p
+  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_x86_avx512_psrli_w_512(<32 x i16> %a0) #0 {
+; CHECK-LABEL: @test_x86_avx512_psrli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP2:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP3:%.*]] = or <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    store <32 x i16> [[TMP3]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  ret <32 x i16> %res
+}
+define <32 x i16> @test_x86_avx512_mask_psrli_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_mask_psrli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = or <32 x i16> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP7:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP5]], <32 x i16> [[TMP3]]
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <32 x i16> [[RES]], [[PASSTHRU:%.*]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = or <32 x i16> [[TMP9]], [[TMP3]]
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP6]], <32 x i16> [[TMP10]], <32 x i16> [[TMP7]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> [[PASSTHRU]]
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
+  ret <32 x i16> %res2
+}
+define <32 x i16> @test_x86_avx512_maskz_psrli_w_512(<32 x i16> %a0, i32 %mask) #0 {
+; CHECK-LABEL: @test_x86_avx512_maskz_psrli_w_512(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @__msan_param_tls, align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8
+; CHECK-NEXT:    call void @llvm.donothing()
+; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[TMP1]], i32 7)
+; CHECK-NEXT:    [[TMP4:%.*]] = or <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[RES:%.*]] = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> [[A0:%.*]], i32 7)
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i32 [[TMP2]] to <32 x i1>
+; CHECK-NEXT:    [[MASK_CAST:%.*]] = bitcast i32 [[MASK:%.*]] to <32 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[TMP4]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <32 x i16> [[RES]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = or <32 x i16> [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP9:%.*]] = or <32 x i16> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <32 x i1> [[TMP5]], <32 x i16> [[TMP9]], <32 x i16> [[TMP6]]
+; CHECK-NEXT:    [[RES2:%.*]] = select <32 x i1> [[MASK_CAST]], <32 x i16> [[RES]], <32 x i16> zeroinitializer
+; CHECK-NEXT:    store <32 x i16> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    ret <32 x i16> [[RES2]]
+;
+  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
+  %mask.cast = bitcast i32 %mask to <32 x i1>
+  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
+  ret <32 x i16> %res2
+}
+declare <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16>, i32) nounwind readnone
+
+attributes #0 = { sanitize_memory }

From acb30465f1b267d3cf15043bb3aaff9ea4b79902 Mon Sep 17 00:00:00 2001
From: Thurston Dang <thurston@google.com>
Date: Tue, 8 Jul 2025 17:04:35 +0000
Subject: [PATCH 2/2] undef -> poison

---
 .../X86/avx512bw-intrinsics-upgrade.ll        | 48 +++++++++----------
 .../X86/avx512bw-intrinsics.ll                | 48 +++++++++----------
 2 files changed, 48 insertions(+), 48 deletions(-)

diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
index c8e7db81ec753..db4ad6b8fc28b 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics-upgrade.ll
@@ -2264,9 +2264,9 @@ define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) noun
 ; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2282,8 +2282,8 @@ define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) noun
 ; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %res
 }
@@ -2307,9 +2307,9 @@ define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32
 ; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2333,8 +2333,8 @@ define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32
 ; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
   ret <32 x i16> %res
 }
@@ -2357,9 +2357,9 @@ define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i3
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2383,8 +2383,8 @@ define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i3
 ; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packssdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
   ret <32 x i16> %res
 }
@@ -2835,9 +2835,9 @@ define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) nou
 ; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2853,8 +2853,8 @@ define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) nou
 ; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %res
 }
@@ -2878,9 +2878,9 @@ define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <3
 ; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2904,8 +2904,8 @@ define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <3
 ; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask)
   ret <32 x i16> %res
 }
@@ -2928,9 +2928,9 @@ define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -2954,8 +2954,8 @@ define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i
 ; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.x86.avx512.mask.packusdw.512(<16 x i32> %a, <16 x i32> %b, <32 x i16> zeroinitializer, i32 %mask)
   ret <32 x i16> %res
 }
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll
index 6f2a6ccbfa702..3f09e54a0224d 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512bw-intrinsics.ll
@@ -531,9 +531,9 @@ define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0 {
 ; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -549,8 +549,8 @@ define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0 {
 ; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
   ret <32 x i16> %1
 }
@@ -574,9 +574,9 @@ define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32
 ; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -600,8 +600,8 @@ define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <32
 ; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
   %2 = bitcast i32 %mask to <32 x i1>
   %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
@@ -626,9 +626,9 @@ define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i3
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -652,8 +652,8 @@ define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i3
 ; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
   %2 = bitcast i32 %mask to <32 x i1>
   %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
@@ -1122,9 +1122,9 @@ define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0
 ; CHECK-NEXT:    [[TMP7:%.*]] = inttoptr i64 [[TMP6]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP7]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP8]], 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -1140,8 +1140,8 @@ define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, ptr %ptr_b) #0
 ; CHECK-NEXT:    ret <32 x i16> [[TMP12]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
   ret <32 x i16> %1
 }
@@ -1165,9 +1165,9 @@ define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <3
 ; CHECK-NEXT:    [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP10]], 0
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -1191,8 +1191,8 @@ define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, ptr %ptr_b, <3
 ; CHECK-NEXT:    ret <32 x i16> [[TMP21]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
   %2 = bitcast i32 %mask to <32 x i1>
   %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
@@ -1217,9 +1217,9 @@ define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load i32, ptr [[TMP8]], align 4
 ; CHECK-NEXT:    [[_MSPROP:%.*]] = insertelement <16 x i32> splat (i32 -1), i32 [[_MSLD]], i32 0
-; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> undef, i32 [[Q]], i32 0
+; CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <16 x i32> poison, i32 [[Q]], i32 0
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = shufflevector <16 x i32> [[_MSPROP]], <16 x i32> splat (i32 -1), <16 x i32> zeroinitializer
-; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> undef, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[B:%.*]] = shufflevector <16 x i32> [[VECINIT_I]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast <16 x i32> [[TMP2]] to i512
 ; CHECK-NEXT:    [[_MSCMP2:%.*]] = icmp ne i512 [[TMP9]], 0
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <16 x i32> [[_MSPROP1]] to i512
@@ -1243,8 +1243,8 @@ define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, ptr %ptr_b, i
 ; CHECK-NEXT:    ret <32 x i16> [[TMP20]]
 ;
   %q = load i32, ptr %ptr_b
-  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
-  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %vecinit.i = insertelement <16 x i32> poison, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> poison, <16 x i32> zeroinitializer
   %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
   %2 = bitcast i32 %mask to <32 x i1>
   %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer