From 7e927698621f476e698eea14e6954057c81ccf4b Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Thu, 20 Feb 2020 13:46:56 -0800
Subject: [PATCH] [X86] Expand vselect of v1i1 under avx512.

We already do this for v2i1, v4i1, etc.
---
 llvm/lib/Target/X86/X86ISelLowering.cpp |  2 +-
 llvm/test/CodeGen/X86/avx512-select.ll  | 86 +++++++++++++++++++++++++
 2 files changed, 87 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index bf304fc0060be..c56dd33c98aef 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1449,6 +1449,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       setOperationAction(ISD::SADDSAT,          VT, Custom);
       setOperationAction(ISD::USUBSAT,          VT, Custom);
       setOperationAction(ISD::SSUBSAT,          VT, Custom);
+      setOperationAction(ISD::VSELECT,          VT,  Expand);
     }
 
     for (auto VT : { MVT::v2i1, MVT::v4i1, MVT::v8i1, MVT::v16i1 }) {
@@ -1464,7 +1465,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       setOperationAction(ISD::INSERT_SUBVECTOR, VT, Custom);
       setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
       setOperationAction(ISD::VECTOR_SHUFFLE,   VT,  Custom);
-      setOperationAction(ISD::VSELECT,          VT,  Expand);
     }
 
     for (auto VT : { MVT::v1i1, MVT::v2i1, MVT::v4i1, MVT::v8i1 })
diff --git a/llvm/test/CodeGen/X86/avx512-select.ll b/llvm/test/CodeGen/X86/avx512-select.ll
index cd4a70c4f73c4..5cbca00501776 100644
--- a/llvm/test/CodeGen/X86/avx512-select.ll
+++ b/llvm/test/CodeGen/X86/avx512-select.ll
@@ -591,3 +591,89 @@ define <16 x i64> @narrowExtractedVectorSelect_crash(<16 x i64> %arg, <16 x i16>
   %tmp3 = zext <16 x i16> %tmp2 to <16 x i64>
   ret <16 x i64> %tmp3
 }
+
+define void @vselect_v1i1(<1 x i1>* %w, <1 x i1>* %x, <1 x i1>* %y) nounwind {
+; X86-AVX512F-LABEL: vselect_v1i1:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    pushl %esi
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512F-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-AVX512F-NEXT:    movzbl (%edx), %esi
+; X86-AVX512F-NEXT:    kmovw %esi, %k0
+; X86-AVX512F-NEXT:    movzbl (%ecx), %ecx
+; X86-AVX512F-NEXT:    kmovw %ecx, %k1
+; X86-AVX512F-NEXT:    movzbl (%eax), %eax
+; X86-AVX512F-NEXT:    kmovw %eax, %k2
+; X86-AVX512F-NEXT:    kandw %k2, %k0, %k0
+; X86-AVX512F-NEXT:    kxnorw %k0, %k0, %k3
+; X86-AVX512F-NEXT:    kxorw %k3, %k2, %k2
+; X86-AVX512F-NEXT:    kandw %k2, %k1, %k1
+; X86-AVX512F-NEXT:    korw %k1, %k0, %k0
+; X86-AVX512F-NEXT:    kmovw %k0, %eax
+; X86-AVX512F-NEXT:    movb %al, (%edx)
+; X86-AVX512F-NEXT:    popl %esi
+; X86-AVX512F-NEXT:    retl
+;
+; X64-AVX512F-LABEL: vselect_v1i1:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    movzbl (%rsi), %eax
+; X64-AVX512F-NEXT:    kmovw %eax, %k0
+; X64-AVX512F-NEXT:    movzbl (%rdx), %eax
+; X64-AVX512F-NEXT:    kmovw %eax, %k1
+; X64-AVX512F-NEXT:    movzbl (%rdi), %eax
+; X64-AVX512F-NEXT:    kmovw %eax, %k2
+; X64-AVX512F-NEXT:    kandw %k2, %k0, %k0
+; X64-AVX512F-NEXT:    kxnorw %k0, %k0, %k3
+; X64-AVX512F-NEXT:    kxorw %k3, %k2, %k2
+; X64-AVX512F-NEXT:    kandw %k2, %k1, %k1
+; X64-AVX512F-NEXT:    korw %k1, %k0, %k0
+; X64-AVX512F-NEXT:    kmovw %k0, %eax
+; X64-AVX512F-NEXT:    movb %al, (%rsi)
+; X64-AVX512F-NEXT:    retq
+;
+; X86-AVX512BW-LABEL: vselect_v1i1:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    pushl %esi
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-AVX512BW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-AVX512BW-NEXT:    movzbl (%edx), %esi
+; X86-AVX512BW-NEXT:    kmovd %esi, %k0
+; X86-AVX512BW-NEXT:    movzbl (%ecx), %ecx
+; X86-AVX512BW-NEXT:    kmovd %ecx, %k1
+; X86-AVX512BW-NEXT:    movzbl (%eax), %eax
+; X86-AVX512BW-NEXT:    kmovd %eax, %k2
+; X86-AVX512BW-NEXT:    kandw %k2, %k0, %k0
+; X86-AVX512BW-NEXT:    kxnorw %k0, %k0, %k3
+; X86-AVX512BW-NEXT:    kxorw %k3, %k2, %k2
+; X86-AVX512BW-NEXT:    kandw %k2, %k1, %k1
+; X86-AVX512BW-NEXT:    korw %k1, %k0, %k0
+; X86-AVX512BW-NEXT:    kmovd %k0, %eax
+; X86-AVX512BW-NEXT:    movb %al, (%edx)
+; X86-AVX512BW-NEXT:    popl %esi
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512BW-LABEL: vselect_v1i1:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    movzbl (%rsi), %eax
+; X64-AVX512BW-NEXT:    kmovd %eax, %k0
+; X64-AVX512BW-NEXT:    movzbl (%rdx), %eax
+; X64-AVX512BW-NEXT:    kmovd %eax, %k1
+; X64-AVX512BW-NEXT:    movzbl (%rdi), %eax
+; X64-AVX512BW-NEXT:    kmovd %eax, %k2
+; X64-AVX512BW-NEXT:    kandw %k2, %k0, %k0
+; X64-AVX512BW-NEXT:    kxnorw %k0, %k0, %k3
+; X64-AVX512BW-NEXT:    kxorw %k3, %k2, %k2
+; X64-AVX512BW-NEXT:    kandw %k2, %k1, %k1
+; X64-AVX512BW-NEXT:    korw %k1, %k0, %k0
+; X64-AVX512BW-NEXT:    kmovd %k0, %eax
+; X64-AVX512BW-NEXT:    movb %al, (%rsi)
+; X64-AVX512BW-NEXT:    retq
+  %a = load <1 x i1>, <1 x i1>* %x
+  %b = load <1 x i1>, <1 x i1>* %y
+  %b2 = load <1 x i1>, <1 x i1>* %w
+  %c = select <1 x i1> %b2, <1 x i1> %a, <1 x i1> %b
+  store <1 x i1> %c, <1 x i1>* %x
+  ret void
+}