diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 53ef3a81177e9..afe5984b09763 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -18818,9 +18818,33 @@ SDValue X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
   SDValue N0 = Op.getOperand(0);
   SDValue N1 = Op.getOperand(1);
   SDValue N2 = Op.getOperand(2);
-
   auto *N2C = dyn_cast<ConstantSDNode>(N2);
-  if (!N2C || N2C->getAPIntValue().uge(NumElts))
+
+  if (!N2C) {
+    // Variable insertion indices, usually we're better off spilling to stack,
+    // but AVX512 can use a variable compare+select by comparing against all
+    // possible vector indices.
+    if (!(Subtarget.hasBWI() ||
+          (Subtarget.hasAVX512() && EltVT.getScalarSizeInBits() >= 32)))
+      return SDValue();
+
+    MVT IdxSVT = MVT::getIntegerVT(EltVT.getScalarSizeInBits());
+    MVT IdxVT = MVT::getVectorVT(IdxSVT, NumElts);
+    SDValue IdxExt = DAG.getZExtOrTrunc(N2, dl, IdxSVT);
+    SDValue IdxSplat = DAG.getSplatBuildVector(IdxVT, dl, IdxExt);
+    SDValue EltSplat = DAG.getSplatBuildVector(VT, dl, N1);
+
+    SmallVector<SDValue, 16> RawIndices;
+    for (unsigned I = 0; I != NumElts; ++I)
+      RawIndices.push_back(DAG.getConstant(I, dl, IdxSVT));
+    SDValue Indices = DAG.getBuildVector(IdxVT, dl, RawIndices);
+
+    // inselt N0, N1, N2 --> select (SplatN2 == {0,1,2...}) ? SplatN1 : N0.
+    return DAG.getSelectCC(dl, IdxSplat, Indices, EltSplat, N0,
+                           ISD::CondCode::SETEQ);
+  }
+
+  if (N2C->getAPIntValue().uge(NumElts))
     return SDValue();
   uint64_t IdxVal = N2C->getZExtValue();
 
diff --git a/llvm/test/CodeGen/X86/avx512-insert-extract.ll b/llvm/test/CodeGen/X86/avx512-insert-extract.ll
index 50026ec543e2c..ea7f928d09be3 100644
--- a/llvm/test/CodeGen/X86/avx512-insert-extract.ll
+++ b/llvm/test/CodeGen/X86/avx512-insert-extract.ll
@@ -1653,25 +1653,16 @@ define i32 @test_insertelement_variable_v32i1(<32 x i8> %a, i8 %b, i32 %index) {
 ;
 ; SKX-LABEL: test_insertelement_variable_v32i1:
 ; SKX:       ## %bb.0:
-; SKX-NEXT:    pushq %rbp
-; SKX-NEXT:    .cfi_def_cfa_offset 16
-; SKX-NEXT:    .cfi_offset %rbp, -16
-; SKX-NEXT:    movq %rsp, %rbp
-; SKX-NEXT:    .cfi_def_cfa_register %rbp
-; SKX-NEXT:    andq $-32, %rsp
-; SKX-NEXT:    subq $64, %rsp
-; SKX-NEXT:    ## kill: def $esi killed $esi def $rsi
 ; SKX-NEXT:    vptestmb %ymm0, %ymm0, %k0
-; SKX-NEXT:    andl $31, %esi
 ; SKX-NEXT:    testb %dil, %dil
+; SKX-NEXT:    setne %al
+; SKX-NEXT:    vpbroadcastb %esi, %ymm0
+; SKX-NEXT:    vpcmpeqb {{.*}}(%rip), %ymm0, %k1
 ; SKX-NEXT:    vpmovm2b %k0, %ymm0
-; SKX-NEXT:    vmovdqa %ymm0, (%rsp)
-; SKX-NEXT:    setne (%rsp,%rsi)
-; SKX-NEXT:    vpsllw $7, (%rsp), %ymm0
+; SKX-NEXT:    vpbroadcastb %eax, %ymm0 {%k1}
+; SKX-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; SKX-NEXT:    vpmovb2m %ymm0, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
-; SKX-NEXT:    movq %rbp, %rsp
-; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
   %t1 = icmp ugt <32 x i8> %a, zeroinitializer
@@ -1731,25 +1722,16 @@ define i64 @test_insertelement_variable_v64i1(<64 x i8> %a, i8 %b, i32 %index) {
 ;
 ; SKX-LABEL: test_insertelement_variable_v64i1:
 ; SKX:       ## %bb.0:
-; SKX-NEXT:    pushq %rbp
-; SKX-NEXT:    .cfi_def_cfa_offset 16
-; SKX-NEXT:    .cfi_offset %rbp, -16
-; SKX-NEXT:    movq %rsp, %rbp
-; SKX-NEXT:    .cfi_def_cfa_register %rbp
-; SKX-NEXT:    andq $-64, %rsp
-; SKX-NEXT:    subq $128, %rsp
-; SKX-NEXT:    ## kill: def $esi killed $esi def $rsi
 ; SKX-NEXT:    vptestmb %zmm0, %zmm0, %k0
-; SKX-NEXT:    andl $63, %esi
 ; SKX-NEXT:    testb %dil, %dil
+; SKX-NEXT:    setne %al
+; SKX-NEXT:    vpbroadcastb %esi, %zmm0
+; SKX-NEXT:    vpcmpeqb {{.*}}(%rip), %zmm0, %k1
 ; SKX-NEXT:    vpmovm2b %k0, %zmm0
-; SKX-NEXT:    vmovdqa64 %zmm0, (%rsp)
-; SKX-NEXT:    setne (%rsp,%rsi)
-; SKX-NEXT:    vpsllw $7, (%rsp), %zmm0
+; SKX-NEXT:    vpbroadcastb %eax, %zmm0 {%k1}
+; SKX-NEXT:    vpsllw $7, %zmm0, %zmm0
 ; SKX-NEXT:    vpmovb2m %zmm0, %k0
 ; SKX-NEXT:    kmovq %k0, %rax
-; SKX-NEXT:    movq %rbp, %rsp
-; SKX-NEXT:    popq %rbp
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
   %t1 = icmp ugt <64 x i8> %a, zeroinitializer
diff --git a/llvm/test/CodeGen/X86/insertelement-var-index.ll b/llvm/test/CodeGen/X86/insertelement-var-index.ll
index 9acc259dc3254..58a5d59834534 100644
--- a/llvm/test/CodeGen/X86/insertelement-var-index.ll
+++ b/llvm/test/CodeGen/X86/insertelement-var-index.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2   | FileCheck %s --check-prefixes=ALL,SSE,SSE2
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=ALL,SSE,SSE41
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx    | FileCheck %s --check-prefixes=ALL,AVX,AVX1
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2   | FileCheck %s --check-prefixes=ALL,AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx    | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX1
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2   | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX2
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512vl  | FileCheck %s --check-prefixes=ALL,AVX,AVX512,AVX512F
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefixes=ALL,AVX,AVX512,AVX512BW
 
@@ -699,14 +699,30 @@ define <16 x i8> @arg_i8_v16i8(<16 x i8> %v, i8 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i8_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $15, %esi
-; AVX-NEXT:    movb %dil, -24(%rsp,%rsi)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i8_v16i8:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $15, %esi
+; AVX1OR2-NEXT:    movb %dil, -24(%rsp,%rsi)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: arg_i8_v16i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    andl $15, %esi
+; AVX512F-NEXT:    movb %dil, -24(%rsp,%rsi)
+; AVX512F-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: arg_i8_v16i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %esi, %xmm1
+; AVX512BW-NEXT:    vpcmpeqb {{.*}}(%rip), %xmm1, %k1
+; AVX512BW-NEXT:    vpbroadcastb %edi, %xmm0 {%k1}
+; AVX512BW-NEXT:    retq
   %ins = insertelement <16 x i8> %v, i8 %x, i32 %y
   ret <16 x i8> %ins
 }
@@ -721,14 +737,30 @@ define <8 x i16> @arg_i16_v8i16(<8 x i16> %v, i16 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i16_v8i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    movw %di, -24(%rsp,%rsi,2)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i16_v8i16:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $7, %esi
+; AVX1OR2-NEXT:    movw %di, -24(%rsp,%rsi,2)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: arg_i16_v8i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    andl $7, %esi
+; AVX512F-NEXT:    movw %di, -24(%rsp,%rsi,2)
+; AVX512F-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: arg_i16_v8i16:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %esi, %xmm1
+; AVX512BW-NEXT:    vpcmpeqw {{.*}}(%rip), %xmm1, %k1
+; AVX512BW-NEXT:    vpbroadcastw %edi, %xmm0 {%k1}
+; AVX512BW-NEXT:    retq
   %ins = insertelement <8 x i16> %v, i16 %x, i32 %y
   ret <8 x i16> %ins
 }
@@ -743,14 +775,21 @@ define <4 x i32> @arg_i32_v4i32(<4 x i32> %v, i32 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i32_v4i32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    movl %edi, -24(%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i32_v4i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    movl %edi, -24(%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_i32_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %xmm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vpbroadcastd %edi, %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <4 x i32> %v, i32 %x, i32 %y
   ret <4 x i32> %ins
 }
@@ -765,14 +804,22 @@ define <2 x i64> @arg_i64_v2i64(<2 x i64> %v, i64 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i64_v2i64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $1, %esi
-; AVX-NEXT:    movq %rdi, -24(%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i64_v2i64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $1, %esi
+; AVX1OR2-NEXT:    movq %rdi, -24(%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_i64_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %xmm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vpbroadcastq %rdi, %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <2 x i64> %v, i64 %x, i32 %y
   ret <2 x i64> %ins
 }
@@ -787,14 +834,21 @@ define <4 x float> @arg_f32_v4f32(<4 x float> %v, float %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_f32_v4f32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $edi killed $edi def $rdi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $3, %edi
-; AVX-NEXT:    vmovss %xmm1, -24(%rsp,%rdi,4)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_f32_v4f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $edi killed $edi def $rdi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $3, %edi
+; AVX1OR2-NEXT:    vmovss %xmm1, -24(%rsp,%rdi,4)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_f32_v4f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %edi, %xmm2
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm2, %k1
+; AVX512-NEXT:    vbroadcastss %xmm1, %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <4 x float> %v, float %x, i32 %y
   ret <4 x float> %ins
 }
@@ -809,14 +863,22 @@ define <2 x double> @arg_f64_v2f64(<2 x double> %v, double %x, i32 %y) nounwind
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_f64_v2f64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $edi killed $edi def $rdi
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $1, %edi
-; AVX-NEXT:    vmovsd %xmm1, -24(%rsp,%rdi,8)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_f64_v2f64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $edi killed $edi def $rdi
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $1, %edi
+; AVX1OR2-NEXT:    vmovsd %xmm1, -24(%rsp,%rdi,8)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_f64_v2f64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %edi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %xmm2
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %xmm2, %k1
+; AVX512-NEXT:    vmovddup {{.*#+}} xmm0 {%k1} = xmm1[0,0]
+; AVX512-NEXT:    retq
   %ins = insertelement <2 x double> %v, double %x, i32 %y
   ret <2 x double> %ins
 }
@@ -832,15 +894,32 @@ define <16 x i8> @load_i8_v16i8(<16 x i8> %v, i8* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i8_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movb (%rdi), %al
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $15, %esi
-; AVX-NEXT:    movb %al, -24(%rsp,%rsi)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i8_v16i8:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movb (%rdi), %al
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $15, %esi
+; AVX1OR2-NEXT:    movb %al, -24(%rsp,%rsi)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: load_i8_v16i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    movb (%rdi), %al
+; AVX512F-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    andl $15, %esi
+; AVX512F-NEXT:    movb %al, -24(%rsp,%rsi)
+; AVX512F-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: load_i8_v16i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %esi, %xmm1
+; AVX512BW-NEXT:    vpcmpeqb {{.*}}(%rip), %xmm1, %k1
+; AVX512BW-NEXT:    vpbroadcastb (%rdi), %xmm0 {%k1}
+; AVX512BW-NEXT:    retq
   %x = load i8, i8* %p
   %ins = insertelement <16 x i8> %v, i8 %x, i32 %y
   ret <16 x i8> %ins
@@ -857,15 +936,32 @@ define <8 x i16> @load_i16_v8i16(<8 x i16> %v, i16* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i16_v8i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movzwl (%rdi), %eax
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    movw %ax, -24(%rsp,%rsi,2)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i16_v8i16:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movzwl (%rdi), %eax
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $7, %esi
+; AVX1OR2-NEXT:    movw %ax, -24(%rsp,%rsi,2)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: load_i16_v8i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    movzwl (%rdi), %eax
+; AVX512F-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX512F-NEXT:    andl $7, %esi
+; AVX512F-NEXT:    movw %ax, -24(%rsp,%rsi,2)
+; AVX512F-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: load_i16_v8i16:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %esi, %xmm1
+; AVX512BW-NEXT:    vpcmpeqw {{.*}}(%rip), %xmm1, %k1
+; AVX512BW-NEXT:    vpbroadcastw (%rdi), %xmm0 {%k1}
+; AVX512BW-NEXT:    retq
   %x = load i16, i16* %p
   %ins = insertelement <8 x i16> %v, i16 %x, i32 %y
   ret <8 x i16> %ins
@@ -882,15 +978,22 @@ define <4 x i32> @load_i32_v4i32(<4 x i32> %v, i32* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i32_v4i32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movl (%rdi), %eax
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    movl %eax, -24(%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i32_v4i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movl (%rdi), %eax
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    movl %eax, -24(%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_i32_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %xmm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vpbroadcastd (%rdi), %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load i32, i32* %p
   %ins = insertelement <4 x i32> %v, i32 %x, i32 %y
   ret <4 x i32> %ins
@@ -907,15 +1010,23 @@ define <2 x i64> @load_i64_v2i64(<2 x i64> %v, i64* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i64_v2i64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movq (%rdi), %rax
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $1, %esi
-; AVX-NEXT:    movq %rax, -24(%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i64_v2i64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movq (%rdi), %rax
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $1, %esi
+; AVX1OR2-NEXT:    movq %rax, -24(%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_i64_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %xmm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vpbroadcastq (%rdi), %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load i64, i64* %p
   %ins = insertelement <2 x i64> %v, i64 %x, i32 %y
   ret <2 x i64> %ins
@@ -932,15 +1043,22 @@ define <4 x float> @load_f32_v4f32(<4 x float> %v, float* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_f32_v4f32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    vmovss %xmm1, -24(%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_f32_v4f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    vmovss %xmm1, -24(%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_f32_v4f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %xmm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load float, float* %p
   %ins = insertelement <4 x float> %v, float %x, i32 %y
   ret <4 x float> %ins
@@ -957,15 +1075,23 @@ define <2 x double> @load_f64_v2f64(<2 x double> %v, double* %p, i32 %y) nounwin
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_f64_v2f64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $1, %esi
-; AVX-NEXT:    vmovsd %xmm1, -24(%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_f64_v2f64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX1OR2-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVX1OR2-NEXT:    andl $1, %esi
+; AVX1OR2-NEXT:    vmovsd %xmm1, -24(%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_f64_v2f64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %xmm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %xmm1, %k1
+; AVX512-NEXT:    vmovddup {{.*#+}} xmm0 {%k1} = mem[0,0]
+; AVX512-NEXT:    retq
   %x = load double, double* %p
   %ins = insertelement <2 x double> %v, double %x, i32 %y
   ret <2 x double> %ins
@@ -983,20 +1109,42 @@ define <32 x i8> @arg_i8_v32i8(<32 x i8> %v, i8 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i8_v32i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $31, %esi
-; AVX-NEXT:    movb %dil, (%rsp,%rsi)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i8_v32i8:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $31, %esi
+; AVX1OR2-NEXT:    movb %dil, (%rsp,%rsi)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: arg_i8_v32i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $31, %esi
+; AVX512F-NEXT:    movb %dil, (%rsp,%rsi)
+; AVX512F-NEXT:    vmovaps (%rsp), %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: arg_i8_v32i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %esi, %ymm1
+; AVX512BW-NEXT:    vpcmpeqb {{.*}}(%rip), %ymm1, %k1
+; AVX512BW-NEXT:    vpbroadcastb %edi, %ymm0 {%k1}
+; AVX512BW-NEXT:    retq
   %ins = insertelement <32 x i8> %v, i8 %x, i32 %y
   ret <32 x i8> %ins
 }
@@ -1013,20 +1161,42 @@ define <16 x i16> @arg_i16_v16i16(<16 x i16> %v, i16 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i16_v16i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $15, %esi
-; AVX-NEXT:    movw %di, (%rsp,%rsi,2)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i16_v16i16:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $15, %esi
+; AVX1OR2-NEXT:    movw %di, (%rsp,%rsi,2)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: arg_i16_v16i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $15, %esi
+; AVX512F-NEXT:    movw %di, (%rsp,%rsi,2)
+; AVX512F-NEXT:    vmovaps (%rsp), %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: arg_i16_v16i16:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %esi, %ymm1
+; AVX512BW-NEXT:    vpcmpeqw {{.*}}(%rip), %ymm1, %k1
+; AVX512BW-NEXT:    vpbroadcastw %edi, %ymm0 {%k1}
+; AVX512BW-NEXT:    retq
   %ins = insertelement <16 x i16> %v, i16 %x, i32 %y
   ret <16 x i16> %ins
 }
@@ -1043,20 +1213,27 @@ define <8 x i32> @arg_i32_v8i32(<8 x i32> %v, i32 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i32_v8i32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    movl %edi, (%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i32_v8i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $7, %esi
+; AVX1OR2-NEXT:    movl %edi, (%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_i32_v8i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %ymm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vpbroadcastd %edi, %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <8 x i32> %v, i32 %x, i32 %y
   ret <8 x i32> %ins
 }
@@ -1073,20 +1250,28 @@ define <4 x i64> @arg_i64_v4i64(<4 x i64> %v, i64 %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_i64_v4i64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    movq %rdi, (%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_i64_v4i64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    movq %rdi, (%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_i64_v4i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %ymm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vpbroadcastq %rdi, %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <4 x i64> %v, i64 %x, i32 %y
   ret <4 x i64> %ins
 }
@@ -1103,20 +1288,27 @@ define <8 x float> @arg_f32_v8f32(<8 x float> %v, float %x, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_f32_v8f32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $edi killed $edi def $rdi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $7, %edi
-; AVX-NEXT:    vmovss %xmm1, (%rsp,%rdi,4)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_f32_v8f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $edi killed $edi def $rdi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $7, %edi
+; AVX1OR2-NEXT:    vmovss %xmm1, (%rsp,%rdi,4)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_f32_v8f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %edi, %ymm2
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %ymm2, %k1
+; AVX512-NEXT:    vbroadcastss %xmm1, %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <8 x float> %v, float %x, i32 %y
   ret <8 x float> %ins
 }
@@ -1133,20 +1325,28 @@ define <4 x double> @arg_f64_v4f64(<4 x double> %v, double %x, i32 %y) nounwind
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: arg_f64_v4f64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $edi killed $edi def $rdi
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $3, %edi
-; AVX-NEXT:    vmovsd %xmm1, (%rsp,%rdi,8)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: arg_f64_v4f64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $edi killed $edi def $rdi
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $3, %edi
+; AVX1OR2-NEXT:    vmovsd %xmm1, (%rsp,%rdi,8)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: arg_f64_v4f64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %edi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %ymm2
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %ymm2, %k1
+; AVX512-NEXT:    vbroadcastsd %xmm1, %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %ins = insertelement <4 x double> %v, double %x, i32 %y
   ret <4 x double> %ins
 }
@@ -1164,21 +1364,44 @@ define <32 x i8> @load_i8_v32i8(<32 x i8> %v, i8* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i8_v32i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movb (%rdi), %al
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $31, %esi
-; AVX-NEXT:    movb %al, (%rsp,%rsi)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i8_v32i8:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movb (%rdi), %al
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $31, %esi
+; AVX1OR2-NEXT:    movb %al, (%rsp,%rsi)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: load_i8_v32i8:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    movb (%rdi), %al
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $31, %esi
+; AVX512F-NEXT:    movb %al, (%rsp,%rsi)
+; AVX512F-NEXT:    vmovaps (%rsp), %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: load_i8_v32i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %esi, %ymm1
+; AVX512BW-NEXT:    vpcmpeqb {{.*}}(%rip), %ymm1, %k1
+; AVX512BW-NEXT:    vpbroadcastb (%rdi), %ymm0 {%k1}
+; AVX512BW-NEXT:    retq
   %x = load i8, i8* %p
   %ins = insertelement <32 x i8> %v, i8 %x, i32 %y
   ret <32 x i8> %ins
@@ -1197,21 +1420,44 @@ define <16 x i16> @load_i16_v16i16(<16 x i16> %v, i16* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i16_v16i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movzwl (%rdi), %eax
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $15, %esi
-; AVX-NEXT:    movw %ax, (%rsp,%rsi,2)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i16_v16i16:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movzwl (%rdi), %eax
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $15, %esi
+; AVX1OR2-NEXT:    movw %ax, (%rsp,%rsi,2)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512F-LABEL: load_i16_v16i16:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX512F-NEXT:    movzwl (%rdi), %eax
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $15, %esi
+; AVX512F-NEXT:    movw %ax, (%rsp,%rsi,2)
+; AVX512F-NEXT:    vmovaps (%rsp), %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: load_i16_v16i16:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %esi, %ymm1
+; AVX512BW-NEXT:    vpcmpeqw {{.*}}(%rip), %ymm1, %k1
+; AVX512BW-NEXT:    vpbroadcastw (%rdi), %ymm0 {%k1}
+; AVX512BW-NEXT:    retq
   %x = load i16, i16* %p
   %ins = insertelement <16 x i16> %v, i16 %x, i32 %y
   ret <16 x i16> %ins
@@ -1230,21 +1476,28 @@ define <8 x i32> @load_i32_v8i32(<8 x i32> %v, i32* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i32_v8i32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movl (%rdi), %eax
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    movl %eax, (%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i32_v8i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movl (%rdi), %eax
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $7, %esi
+; AVX1OR2-NEXT:    movl %eax, (%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_i32_v8i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %ymm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vpbroadcastd (%rdi), %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load i32, i32* %p
   %ins = insertelement <8 x i32> %v, i32 %x, i32 %y
   ret <8 x i32> %ins
@@ -1263,21 +1516,29 @@ define <4 x i64> @load_i64_v4i64(<4 x i64> %v, i64* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_i64_v4i64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    movq (%rdi), %rax
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    movq %rax, (%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_i64_v4i64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    movq (%rdi), %rax
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    movq %rax, (%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_i64_v4i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %ymm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vpbroadcastq (%rdi), %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load i64, i64* %p
   %ins = insertelement <4 x i64> %v, i64 %x, i32 %y
   ret <4 x i64> %ins
@@ -1296,21 +1557,28 @@ define <8 x float> @load_f32_v8f32(<8 x float> %v, float* %p, i32 %y) nounwind {
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_f32_v8f32:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    vmovss %xmm1, (%rsp,%rsi,4)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_f32_v8f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $7, %esi
+; AVX1OR2-NEXT:    vmovss %xmm1, (%rsp,%rsi,4)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_f32_v8f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd %esi, %ymm1
+; AVX512-NEXT:    vpcmpeqd {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load float, float* %p
   %ins = insertelement <8 x float> %v, float %x, i32 %y
   ret <8 x float> %ins
@@ -1329,21 +1597,29 @@ define <4 x double> @load_f64_v4f64(<4 x double> %v, double* %p, i32 %y) nounwin
 ; SSE-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm1
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: load_f64_v4f64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    pushq %rbp
-; AVX-NEXT:    movq %rsp, %rbp
-; AVX-NEXT:    andq $-32, %rsp
-; AVX-NEXT:    subq $64, %rsp
-; AVX-NEXT:    # kill: def $esi killed $esi def $rsi
-; AVX-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX-NEXT:    vmovaps %ymm0, (%rsp)
-; AVX-NEXT:    andl $3, %esi
-; AVX-NEXT:    vmovsd %xmm1, (%rsp,%rsi,8)
-; AVX-NEXT:    vmovaps (%rsp), %ymm0
-; AVX-NEXT:    movq %rbp, %rsp
-; AVX-NEXT:    popq %rbp
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: load_f64_v4f64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    pushq %rbp
+; AVX1OR2-NEXT:    movq %rsp, %rbp
+; AVX1OR2-NEXT:    andq $-32, %rsp
+; AVX1OR2-NEXT:    subq $64, %rsp
+; AVX1OR2-NEXT:    # kill: def $esi killed $esi def $rsi
+; AVX1OR2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX1OR2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX1OR2-NEXT:    andl $3, %esi
+; AVX1OR2-NEXT:    vmovsd %xmm1, (%rsp,%rsi,8)
+; AVX1OR2-NEXT:    vmovaps (%rsp), %ymm0
+; AVX1OR2-NEXT:    movq %rbp, %rsp
+; AVX1OR2-NEXT:    popq %rbp
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: load_f64_v4f64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movslq %esi, %rax
+; AVX512-NEXT:    vpbroadcastq %rax, %ymm1
+; AVX512-NEXT:    vpcmpeqq {{.*}}(%rip), %ymm1, %k1
+; AVX512-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
+; AVX512-NEXT:    retq
   %x = load double, double* %p
   %ins = insertelement <4 x double> %v, double %x, i32 %y
   ret <4 x double> %ins