diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -18970,6 +18970,9 @@
 
     MVT IdxSVT = MVT::getIntegerVT(EltSizeInBits);
     MVT IdxVT = MVT::getVectorVT(IdxSVT, NumElts);
+    if (!isTypeLegal(IdxSVT) || !isTypeLegal(IdxVT))
+      return SDValue();
+
     SDValue IdxExt = DAG.getZExtOrTrunc(N2, dl, IdxSVT);
     SDValue IdxSplat = DAG.getSplatBuildVector(IdxVT, dl, IdxExt);
     SDValue EltSplat = DAG.getSplatBuildVector(VT, dl, N1);
diff --git a/llvm/test/CodeGen/X86/insertelement-var-index.ll b/llvm/test/CodeGen/X86/insertelement-var-index.ll
--- a/llvm/test/CodeGen/X86/insertelement-var-index.ll
+++ b/llvm/test/CodeGen/X86/insertelement-var-index.ll
@@ -5,11 +5,12 @@
 ; 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
+; RUN: llc < %s -mtriple=i686-- -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,X86AVX2
 
 define <16 x i8> @undef_index(i8 %x) nounwind {
 ; ALL-LABEL: undef_index:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    retq
+; ALL-NEXT:    ret{{[l|q]}}
   %ins = insertelement <16 x i8> undef, i8 %x, i64 undef
   ret <16 x i8> %ins
 }
@@ -17,7 +18,7 @@
 define <16 x i8> @undef_scalar(<16 x i8> %x, i32 %index) nounwind {
 ; ALL-LABEL: undef_scalar:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    retq
+; ALL-NEXT:    ret{{[l|q]}}
   %ins = insertelement <16 x i8> %x, i8 undef, i32 %index
   ret <16 x i8> %ins
 }
@@ -65,6 +66,11 @@
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastb %edi, %xmm0
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i8_v16i8_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastb {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <16 x i8> undef, i8 %x, i32 %y
   ret <16 x i8> %ins
 }
@@ -100,6 +106,11 @@
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastw %edi, %xmm0
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i16_v8i16_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastw {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x i16> undef, i16 %x, i32 %y
   ret <8 x i16> %ins
 }
@@ -127,6 +138,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastd %edi, %xmm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i32_v4i32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x i32> undef, i32 %x, i32 %y
   ret <4 x i32> %ins
 }
@@ -154,6 +170,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastq %rdi, %xmm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i64_v2i64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
+; X86AVX2-NEXT:    retl
   %ins = insertelement <2 x i64> undef, i64 %x, i32 %y
   ret <2 x i64> %ins
 }
@@ -178,6 +199,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vbroadcastss %xmm0, %xmm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f32_v4f32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x float> undef, float %x, i32 %y
   ret <4 x float> %ins
 }
@@ -197,6 +223,11 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f64_v2f64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
+; X86AVX2-NEXT:    retl
   %ins = insertelement <2 x double> undef, double %x, i32 %y
   ret <2 x double> %ins
 }
@@ -236,6 +267,12 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastb (%rdi), %xmm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i8_v16i8_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastb (%eax), %xmm0
+; X86AVX2-NEXT:    retl
   %x = load i8, i8* %p
   %ins = insertelement <16 x i8> undef, i8 %x, i32 %y
   ret <16 x i8> %ins
@@ -267,6 +304,12 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastw (%rdi), %xmm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i16_v8i16_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastw (%eax), %xmm0
+; X86AVX2-NEXT:    retl
   %x = load i16, i16* %p
   %ins = insertelement <8 x i16> undef, i16 %x, i32 %y
   ret <8 x i16> %ins
@@ -283,6 +326,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastss (%rdi), %xmm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i32_v4i32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastss (%eax), %xmm0
+; X86AVX2-NEXT:    retl
   %x = load i32, i32* %p
   %ins = insertelement <4 x i32> undef, i32 %x, i32 %y
   ret <4 x i32> %ins
@@ -299,6 +348,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i64_v2i64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
+; X86AVX2-NEXT:    retl
   %x = load i64, i64* %p
   %ins = insertelement <2 x i64> undef, i64 %x, i32 %y
   ret <2 x i64> %ins
@@ -315,6 +370,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastss (%rdi), %xmm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f32_v4f32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastss (%eax), %xmm0
+; X86AVX2-NEXT:    retl
   %x = load float, float* %p
   %ins = insertelement <4 x float> undef, float %x, i32 %y
   ret <4 x float> %ins
@@ -336,6 +397,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f64_v2f64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
+; X86AVX2-NEXT:    retl
   %x = load double, double* %p
   %ins = insertelement <2 x double> undef, double %x, i32 %y
   ret <2 x double> %ins
@@ -375,6 +442,11 @@
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastb %edi, %ymm0
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i8_v32i8_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastb {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <32 x i8> undef, i8 %x, i32 %y
   ret <32 x i8> %ins
 }
@@ -413,6 +485,11 @@
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastw %edi, %ymm0
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i16_v16i16_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastw {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <16 x i16> undef, i16 %x, i32 %y
   ret <16 x i16> %ins
 }
@@ -444,6 +521,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastd %edi, %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i32_v8i32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x i32> undef, i32 %x, i32 %y
   ret <8 x i32> %ins
 }
@@ -475,6 +557,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastq %rdi, %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i64_v4i64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x i64> undef, i64 %x, i32 %y
   ret <4 x i64> %ins
 }
@@ -504,6 +591,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vbroadcastss %xmm0, %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f32_v8f32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x float> undef, float %x, i32 %y
   ret <8 x float> %ins
 }
@@ -533,6 +625,11 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f64_v4f64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x double> undef, double %x, i32 %y
   ret <4 x double> %ins
 }
@@ -566,6 +663,12 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastb (%rdi), %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i8_v32i8_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastb (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load i8, i8* %p
   %ins = insertelement <32 x i8> undef, i8 %x, i32 %y
   ret <32 x i8> %ins
@@ -600,6 +703,12 @@
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastw (%rdi), %ymm0
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i16_v16i16_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastw (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load i16, i16* %p
   %ins = insertelement <16 x i16> undef, i16 %x, i32 %y
   ret <16 x i16> %ins
@@ -620,6 +729,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastss (%rdi), %ymm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i32_v8i32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastss (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load i32, i32* %p
   %ins = insertelement <8 x i32> undef, i32 %x, i32 %y
   ret <8 x i32> %ins
@@ -640,6 +755,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastsd (%rdi), %ymm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i64_v4i64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastsd (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load i64, i64* %p
   %ins = insertelement <4 x i64> undef, i64 %x, i32 %y
   ret <4 x i64> %ins
@@ -660,6 +781,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastss (%rdi), %ymm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f32_v8f32_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastss (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load float, float* %p
   %ins = insertelement <8 x float> undef, float %x, i32 %y
   ret <8 x float> %ins
@@ -680,6 +807,12 @@
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vbroadcastsd (%rdi), %ymm0
 ; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f64_v4f64_undef:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vbroadcastsd (%eax), %ymm0
+; X86AVX2-NEXT:    retl
   %x = load double, double* %p
   %ins = insertelement <4 x double> undef, double %x, i32 %y
   ret <4 x double> %ins
@@ -723,6 +856,22 @@
 ; AVX512BW-NEXT:    vpcmpeqb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastb %edi, %xmm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i8_v16i8:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $15, %eax
+; X86AVX2-NEXT:    movb 8(%ebp), %cl
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movb %cl, (%esp,%eax)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <16 x i8> %v, i8 %x, i32 %y
   ret <16 x i8> %ins
 }
@@ -761,6 +910,22 @@
 ; AVX512BW-NEXT:    vpcmpeqw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastw %edi, %xmm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i16_v8i16:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movzwl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movw %cx, (%esp,%eax,2)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x i16> %v, i16 %x, i32 %y
   ret <8 x i16> %ins
 }
@@ -790,6 +955,22 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vpbroadcastd %edi, %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i32_v4i32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movl %ecx, (%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x i32> %v, i32 %x, i32 %y
   ret <4 x i32> %ins
 }
@@ -820,6 +1001,31 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vpbroadcastq %rdi, %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i64_v2i64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    pushl %esi
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $48, %esp
+; X86AVX2-NEXT:    movl 8(%ebp), %eax
+; X86AVX2-NEXT:    movl 12(%ebp), %ecx
+; X86AVX2-NEXT:    movl 16(%ebp), %edx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    leal (%edx,%edx), %esi
+; X86AVX2-NEXT:    andl $3, %esi
+; X86AVX2-NEXT:    movl %eax, (%esp,%esi,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%esp)
+; X86AVX2-NEXT:    leal 1(%edx,%edx), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    movl %ecx, 16(%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    leal -4(%ebp), %esp
+; X86AVX2-NEXT:    popl %esi
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <2 x i64> %v, i64 %x, i32 %y
   ret <2 x i64> %ins
 }
@@ -869,6 +1075,14 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %k1
 ; AVX512-NEXT:    vbroadcastss %xmm1, %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f32_v4f32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastd {{[0-9]+}}(%esp), %xmm1
+; X86AVX2-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %xmm2
+; X86AVX2-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x float> %v, float %x, i32 %y
   ret <4 x float> %ins
 }
@@ -922,6 +1136,22 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %k1
 ; AVX512-NEXT:    vmovddup {{.*#+}} xmm0 {%k1} = xmm1[0,0]
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f64_v2f64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 16(%ebp), %eax
+; X86AVX2-NEXT:    andl $1, %eax
+; X86AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    vmovsd %xmm1, (%esp,%eax,8)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <2 x double> %v, double %x, i32 %y
   ret <2 x double> %ins
 }
@@ -963,6 +1193,23 @@
 ; AVX512BW-NEXT:    vpcmpeqb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastb (%rdi), %xmm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i8_v16i8:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $15, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movb (%ecx), %cl
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movb %cl, (%esp,%eax)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i8, i8* %p
   %ins = insertelement <16 x i8> %v, i8 %x, i32 %y
   ret <16 x i8> %ins
@@ -1005,6 +1252,23 @@
 ; AVX512BW-NEXT:    vpcmpeqw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastw (%rdi), %xmm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i16_v8i16:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movzwl (%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movw %cx, (%esp,%eax,2)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i16, i16* %p
   %ins = insertelement <8 x i16> %v, i16 %x, i32 %y
   ret <8 x i16> %ins
@@ -1037,6 +1301,23 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vpbroadcastd (%rdi), %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i32_v4i32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movl (%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    movl %ecx, (%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i32, i32* %p
   %ins = insertelement <4 x i32> %v, i32 %x, i32 %y
   ret <4 x i32> %ins
@@ -1070,6 +1351,32 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vpbroadcastq (%rdi), %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i64_v2i64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    pushl %esi
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $48, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movl (%ecx), %edx
+; X86AVX2-NEXT:    movl 4(%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    leal (%eax,%eax), %esi
+; X86AVX2-NEXT:    andl $3, %esi
+; X86AVX2-NEXT:    movl %edx, (%esp,%esi,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%esp)
+; X86AVX2-NEXT:    leal 1(%eax,%eax), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    movl %ecx, 16(%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps {{[0-9]+}}(%esp), %xmm0
+; X86AVX2-NEXT:    leal -4(%ebp), %esp
+; X86AVX2-NEXT:    popl %esi
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i64, i64* %p
   %ins = insertelement <2 x i64> %v, i64 %x, i32 %y
   ret <2 x i64> %ins
@@ -1122,6 +1429,15 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f32_v4f32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastd {{[0-9]+}}(%esp), %xmm1
+; X86AVX2-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; X86AVX2-NEXT:    vbroadcastss (%eax), %xmm2
+; X86AVX2-NEXT:    vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; X86AVX2-NEXT:    retl
   %x = load float, float* %p
   %ins = insertelement <4 x float> %v, float %x, i32 %y
   ret <4 x float> %ins
@@ -1177,6 +1493,23 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k1
 ; AVX512-NEXT:    vmovddup {{.*#+}} xmm0 {%k1} = mem[0,0]
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f64_v2f64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-16, %esp
+; X86AVX2-NEXT:    subl $32, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $1, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; X86AVX2-NEXT:    vmovaps %xmm0, (%esp)
+; X86AVX2-NEXT:    vmovsd %xmm1, (%esp,%eax,8)
+; X86AVX2-NEXT:    vmovaps (%esp), %xmm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load double, double* %p
   %ins = insertelement <2 x double> %v, double %x, i32 %y
   ret <2 x double> %ins
@@ -1230,6 +1563,22 @@
 ; AVX512BW-NEXT:    vpcmpeqb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastb %edi, %ymm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i8_v32i8:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $31, %eax
+; X86AVX2-NEXT:    movb 8(%ebp), %cl
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movb %cl, (%esp,%eax)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <32 x i8> %v, i8 %x, i32 %y
   ret <32 x i8> %ins
 }
@@ -1282,6 +1631,22 @@
 ; AVX512BW-NEXT:    vpcmpeqw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastw %edi, %ymm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i16_v16i16:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $15, %eax
+; X86AVX2-NEXT:    movzwl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movw %cx, (%esp,%eax,2)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <16 x i16> %v, i16 %x, i32 %y
   ret <16 x i16> %ins
 }
@@ -1319,6 +1684,22 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vpbroadcastd %edi, %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i32_v8i32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movl %ecx, (%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x i32> %v, i32 %x, i32 %y
   ret <8 x i32> %ins
 }
@@ -1357,6 +1738,31 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vpbroadcastq %rdi, %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_i64_v4i64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    pushl %esi
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $96, %esp
+; X86AVX2-NEXT:    movl 8(%ebp), %eax
+; X86AVX2-NEXT:    movl 12(%ebp), %ecx
+; X86AVX2-NEXT:    movl 16(%ebp), %edx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    leal (%edx,%edx), %esi
+; X86AVX2-NEXT:    andl $7, %esi
+; X86AVX2-NEXT:    movl %eax, (%esp,%esi,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%esp)
+; X86AVX2-NEXT:    leal 1(%edx,%edx), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movl %ecx, 32(%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    leal -4(%ebp), %esp
+; X86AVX2-NEXT:    popl %esi
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x i64> %v, i64 %x, i32 %y
   ret <4 x i64> %ins
 }
@@ -1400,6 +1806,14 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %k1
 ; AVX512-NEXT:    vbroadcastss %xmm1, %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f32_v8f32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    vpbroadcastd {{[0-9]+}}(%esp), %ymm1
+; X86AVX2-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}, %ymm1, %ymm1
+; X86AVX2-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %ymm2
+; X86AVX2-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; X86AVX2-NEXT:    retl
   %ins = insertelement <8 x float> %v, float %x, i32 %y
   ret <8 x float> %ins
 }
@@ -1446,6 +1860,22 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %k1
 ; AVX512-NEXT:    vbroadcastsd %xmm1, %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: arg_f64_v4f64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 16(%ebp), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    vmovsd %xmm1, (%esp,%eax,8)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %ins = insertelement <4 x double> %v, double %x, i32 %y
   ret <4 x double> %ins
 }
@@ -1501,6 +1931,23 @@
 ; AVX512BW-NEXT:    vpcmpeqb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastb (%rdi), %ymm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i8_v32i8:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $31, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movb (%ecx), %cl
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movb %cl, (%esp,%eax)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i8, i8* %p
   %ins = insertelement <32 x i8> %v, i8 %x, i32 %y
   ret <32 x i8> %ins
@@ -1557,6 +2004,23 @@
 ; AVX512BW-NEXT:    vpcmpeqw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512BW-NEXT:    vpbroadcastw (%rdi), %ymm0 {%k1}
 ; AVX512BW-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i16_v16i16:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $15, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movzwl (%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movw %cx, (%esp,%eax,2)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i16, i16* %p
   %ins = insertelement <16 x i16> %v, i16 %x, i32 %y
   ret <16 x i16> %ins
@@ -1597,6 +2061,23 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vpbroadcastd (%rdi), %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i32_v8i32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movl (%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    movl %ecx, (%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i32, i32* %p
   %ins = insertelement <8 x i32> %v, i32 %x, i32 %y
   ret <8 x i32> %ins
@@ -1638,6 +2119,32 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vpbroadcastq (%rdi), %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_i64_v4i64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    pushl %esi
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $96, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    movl (%ecx), %edx
+; X86AVX2-NEXT:    movl 4(%ecx), %ecx
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    leal (%eax,%eax), %esi
+; X86AVX2-NEXT:    andl $7, %esi
+; X86AVX2-NEXT:    movl %edx, (%esp,%esi,4)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    vmovaps %ymm0, {{[0-9]+}}(%esp)
+; X86AVX2-NEXT:    leal 1(%eax,%eax), %eax
+; X86AVX2-NEXT:    andl $7, %eax
+; X86AVX2-NEXT:    movl %ecx, 32(%esp,%eax,4)
+; X86AVX2-NEXT:    vmovaps {{[0-9]+}}(%esp), %ymm0
+; X86AVX2-NEXT:    leal -4(%ebp), %esp
+; X86AVX2-NEXT:    popl %esi
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load i64, i64* %p
   %ins = insertelement <4 x i64> %v, i64 %x, i32 %y
   ret <4 x i64> %ins
@@ -1682,6 +2189,15 @@
 ; AVX512-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f32_v8f32:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    vpbroadcastd {{[0-9]+}}(%esp), %ymm1
+; X86AVX2-NEXT:    vpcmpeqd {{\.?LCPI[0-9]+_[0-9]+}}, %ymm1, %ymm1
+; X86AVX2-NEXT:    vbroadcastss (%eax), %ymm2
+; X86AVX2-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
+; X86AVX2-NEXT:    retl
   %x = load float, float* %p
   %ins = insertelement <8 x float> %v, float %x, i32 %y
   ret <8 x float> %ins
@@ -1729,6 +2245,23 @@
 ; AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %k1
 ; AVX512-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
 ; AVX512-NEXT:    retq
+;
+; X86AVX2-LABEL: load_f64_v4f64:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    pushl %ebp
+; X86AVX2-NEXT:    movl %esp, %ebp
+; X86AVX2-NEXT:    andl $-32, %esp
+; X86AVX2-NEXT:    subl $64, %esp
+; X86AVX2-NEXT:    movl 12(%ebp), %eax
+; X86AVX2-NEXT:    andl $3, %eax
+; X86AVX2-NEXT:    movl 8(%ebp), %ecx
+; X86AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; X86AVX2-NEXT:    vmovaps %ymm0, (%esp)
+; X86AVX2-NEXT:    vmovsd %xmm1, (%esp,%eax,8)
+; X86AVX2-NEXT:    vmovaps (%esp), %ymm0
+; X86AVX2-NEXT:    movl %ebp, %esp
+; X86AVX2-NEXT:    popl %ebp
+; X86AVX2-NEXT:    retl
   %x = load double, double* %p
   %ins = insertelement <4 x double> %v, double %x, i32 %y
   ret <4 x double> %ins
@@ -1737,18 +2270,44 @@
 ; Don't die trying to insert to an invalid index.
 
 define i32 @PR44139(<16 x i64>* %p) {
-; ALL-LABEL: PR44139:
-; ALL:       # %bb.0:
-; ALL-NEXT:    movl (%rdi), %eax
-; ALL-NEXT:    leal 2147483647(%rax), %ecx
-; ALL-NEXT:    testl %eax, %eax
-; ALL-NEXT:    cmovnsl %eax, %ecx
-; ALL-NEXT:    andl $-2147483648, %ecx # imm = 0x80000000
-; ALL-NEXT:    addl %eax, %ecx
-; ALL-NEXT:    # kill: def $eax killed $eax killed $rax
-; ALL-NEXT:    xorl %edx, %edx
-; ALL-NEXT:    divl %ecx
-; ALL-NEXT:    retq
+; SSE-LABEL: PR44139:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movl (%rdi), %eax
+; SSE-NEXT:    leal 2147483647(%rax), %ecx
+; SSE-NEXT:    testl %eax, %eax
+; SSE-NEXT:    cmovnsl %eax, %ecx
+; SSE-NEXT:    andl $-2147483648, %ecx # imm = 0x80000000
+; SSE-NEXT:    addl %eax, %ecx
+; SSE-NEXT:    # kill: def $eax killed $eax killed $rax
+; SSE-NEXT:    xorl %edx, %edx
+; SSE-NEXT:    divl %ecx
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: PR44139:
+; AVX:       # %bb.0:
+; AVX-NEXT:    movl (%rdi), %eax
+; AVX-NEXT:    leal 2147483647(%rax), %ecx
+; AVX-NEXT:    testl %eax, %eax
+; AVX-NEXT:    cmovnsl %eax, %ecx
+; AVX-NEXT:    andl $-2147483648, %ecx # imm = 0x80000000
+; AVX-NEXT:    addl %eax, %ecx
+; AVX-NEXT:    # kill: def $eax killed $eax killed $rax
+; AVX-NEXT:    xorl %edx, %edx
+; AVX-NEXT:    divl %ecx
+; AVX-NEXT:    retq
+;
+; X86AVX2-LABEL: PR44139:
+; X86AVX2:       # %bb.0:
+; X86AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86AVX2-NEXT:    movl (%eax), %eax
+; X86AVX2-NEXT:    leal 2147483647(%eax), %ecx
+; X86AVX2-NEXT:    testl %eax, %eax
+; X86AVX2-NEXT:    cmovnsl %eax, %ecx
+; X86AVX2-NEXT:    andl $-2147483648, %ecx # imm = 0x80000000
+; X86AVX2-NEXT:    addl %eax, %ecx
+; X86AVX2-NEXT:    xorl %edx, %edx
+; X86AVX2-NEXT:    divl %ecx
+; X86AVX2-NEXT:    retl
   %L = load <16 x i64>, <16 x i64>* %p
   %E1 = extractelement <16 x i64> %L, i64 0
   %tempvector = insertelement <16 x i64> undef, i64 %E1, i32 0