; RUN: llc -O3 -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse4.2,+aes,+pclmul < %s | FileCheck %s

target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-unknown"

; Stack reload folding tests.
;
; By including a nop call with sideeffects we can force a partial register spill of the
; relevant registers and check that the reload is correctly folded into the instruction.

define <2 x i64> @stack_fold_aesdec(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_aesdec
  ;CHECK:       aesdec {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aesdec(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aesdec(<2 x i64>, <2 x i64>) nounwind readnone

define <2 x i64> @stack_fold_aesdeclast(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_aesdeclast
  ;CHECK:       aesdeclast {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aesdeclast(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aesdeclast(<2 x i64>, <2 x i64>) nounwind readnone

define <2 x i64> @stack_fold_aesenc(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_aesenc
  ;CHECK:       aesenc {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aesenc(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aesenc(<2 x i64>, <2 x i64>) nounwind readnone

define <2 x i64> @stack_fold_aesenclast(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_aesenclast
  ;CHECK:       aesenclast {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aesenclast(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aesenclast(<2 x i64>, <2 x i64>) nounwind readnone

define <2 x i64> @stack_fold_aesimc(<2 x i64> %a0) {
  ;CHECK-LABEL: stack_fold_aesimc
  ;CHECK:       aesimc {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aesimc(<2 x i64> %a0)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aesimc(<2 x i64>) nounwind readnone

define <2 x i64> @stack_fold_aeskeygenassist(<2 x i64> %a0) {
  ;CHECK-LABEL: stack_fold_aeskeygenassist
  ;CHECK:       aeskeygenassist $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.aesni.aeskeygenassist(<2 x i64> %a0, i8 7)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.aesni.aeskeygenassist(<2 x i64>, i8) nounwind readnone

define i32 @stack_fold_crc32_32_8(i32 %a0, i8 %a1) {
  ;CHECK-LABEL: stack_fold_crc32_32_8
  ;CHECK:       crc32b {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 1-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = call i32 @llvm.x86.sse42.crc32.32.8(i32 %a0, i8 %a1)
  ret i32 %2
}
declare i32 @llvm.x86.sse42.crc32.32.8(i32, i8) nounwind

define i32 @stack_fold_crc32_32_16(i32 %a0, i16 %a1) {
  ;CHECK-LABEL: stack_fold_crc32_32_16
  ;CHECK:       crc32w {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 2-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = call i32 @llvm.x86.sse42.crc32.32.16(i32 %a0, i16 %a1)
  ret i32 %2
}
declare i32 @llvm.x86.sse42.crc32.32.16(i32, i16) nounwind

define i32 @stack_fold_crc32_32_32(i32 %a0, i32 %a1) {
  ;CHECK-LABEL: stack_fold_crc32_32_32
  ;CHECK:       crc32l {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 4-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = call i32 @llvm.x86.sse42.crc32.32.32(i32 %a0, i32 %a1)
  ret i32 %2
}
declare i32 @llvm.x86.sse42.crc32.32.32(i32, i32) nounwind

define i64 @stack_fold_crc32_64_64(i64 %a0, i64 %a1) {
  ;CHECK-LABEL: stack_fold_crc32_64_64
  ;CHECK:       crc32q {{-?[0-9]*}}(%rsp), %rax {{.*#+}} 8-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = call i64 @llvm.x86.sse42.crc32.64.64(i64 %a0, i64 %a1)
  ret i64 %2
}
declare i64 @llvm.x86.sse42.crc32.64.64(i64, i64) nounwind

define <4 x i32> @stack_fold_movd_load(i32 %a0) {
  ;CHECK-LABEL: stack_fold_movd_load
  ;CHECK:       movd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 4-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = insertelement <4 x i32> zeroinitializer, i32 %a0, i32 0
  ; add forces execution domain
  %3 = add <4 x i32> %2, <i32 1, i32 1, i32 1, i32 1>
  ret <4 x i32> %3
}

define i32 @stack_fold_movd_store(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_movd_store
  ;CHECK:       movd {{%xmm[0-9][0-9]*}}, {{-?[0-9]*}}(%rsp) {{.*#+}} 4-byte Folded Spill
  ; add forces execution domain
  %1 = add <4 x i32> %a0, <i32 1, i32 1, i32 1, i32 1>
  %2 = extractelement <4 x i32> %1, i32 0
  %3 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i32 %2
}

define <2 x i64> @stack_fold_movq_load(<2 x i64> %a0) {
  ;CHECK-LABEL: stack_fold_movq_load
  ;CHECK:       movq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <2 x i64> %a0, <2 x i64> zeroinitializer, <2 x i32> <i32 0, i32 2>
  ; add forces execution domain
  %3 = add <2 x i64> %2, <i64 1, i64 1>
  ret <2 x i64> %3
}

define i64 @stack_fold_movq_store(<2 x i64> %a0) {
  ;CHECK-LABEL: stack_fold_movq_store
  ;CHECK:       movq {{%xmm[0-9][0-9]*}}, {{-?[0-9]*}}(%rsp) {{.*#+}} 8-byte Folded Spill
  ; add forces execution domain
  %1 = add <2 x i64> %a0, <i64 1, i64 1>
  %2 = extractelement <2 x i64> %1, i32 0
  %3 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i64 %2
}

define <8 x i16> @stack_fold_mpsadbw(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_mpsadbw
  ;CHECK:       mpsadbw $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse41.mpsadbw(<16 x i8> %a0, <16 x i8> %a1, i8 7)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse41.mpsadbw(<16 x i8>, <16 x i8>, i8) nounwind readnone

define <16 x i8> @stack_fold_pabsb(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pabsb
  ;CHECK:       pabsb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <16 x i8> %a0, zeroinitializer
  %3 = sub <16 x i8> zeroinitializer, %a0
  %4 = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %3
  ret <16 x i8> %4
}

define <4 x i32> @stack_fold_pabsd(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_pabsd
  ;CHECK:       pabsd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <4 x i32> %a0, zeroinitializer
  %3 = sub <4 x i32> zeroinitializer, %a0
  %4 = select <4 x i1> %2, <4 x i32> %a0, <4 x i32> %3
  ret <4 x i32> %4
}

define <8 x i16> @stack_fold_pabsw(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pabsw
  ;CHECK:       pabsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <8 x i16> %a0, zeroinitializer
  %3 = sub <8 x i16> zeroinitializer, %a0
  %4 = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %3
  ret <8 x i16> %4
}

define <8 x i16> @stack_fold_packssdw(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_packssdw
  ;CHECK:       packssdw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.packssdw.128(<4 x i32> %a0, <4 x i32> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.packssdw.128(<4 x i32>, <4 x i32>) nounwind readnone

define <16 x i8> @stack_fold_packsswb(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_packsswb
  ;CHECK:       packsswb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_packusdw(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_packusdw
  ;CHECK:       packusdw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse41.packusdw(<4 x i32> %a0, <4 x i32> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse41.packusdw(<4 x i32>, <4 x i32>) nounwind readnone

define <16 x i8> @stack_fold_packuswb(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_packuswb
  ;CHECK:       packuswb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.packuswb.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.packuswb.128(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_paddb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_paddb
  ;CHECK:       paddb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = add <16 x i8> %a0, %a1
  ret <16 x i8> %2
}

define <4 x i32> @stack_fold_paddd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_paddd
  ;CHECK:       paddd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = add <4 x i32> %a0, %a1
  ret <4 x i32> %2
}

define <2 x i64> @stack_fold_paddq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_paddq
  ;CHECK:       paddq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = add <2 x i64> %a0, %a1
  ret <2 x i64> %2
}

define <16 x i8> @stack_fold_paddsb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_paddsb
  ;CHECK:       paddsb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone

define <8 x i16> @stack_fold_paddsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_paddsw
  ;CHECK:       paddsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_paddusb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_paddusb
  ;CHECK:       paddusb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8>, <16 x i8>) nounwind readnone

define <8 x i16> @stack_fold_paddusw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_paddusw
  ;CHECK:       paddusw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_paddw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_paddw
  ;CHECK:       paddw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = add <8 x i16> %a0, %a1
  ret <8 x i16> %2
}

define <16 x i8> @stack_fold_palignr(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_palignr
  ;CHECK:       palignr $1, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a1, <16 x i8> %a0, <16 x i32> <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>
  ret <16 x i8> %2
}

define <16 x i8> @stack_fold_pand(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pand
  ;CHECK:       pand {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = and <16 x i8> %a0, %a1
  ; add forces execution domain
  %3 = add <16 x i8> %2, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
  ret <16 x i8> %3
}

define <16 x i8> @stack_fold_pandn(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pandn
  ;CHECK:       pandn {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = xor <16 x i8> %a0, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  %3 = and <16 x i8> %2, %a1
  ; add forces execution domain
  %4 = add <16 x i8> %3, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
  ret <16 x i8> %4
}

define <16 x i8> @stack_fold_pavgb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pavgb
  ;CHECK:       pavgb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = zext <16 x i8> %a0 to <16 x i16>
  %3 = zext <16 x i8> %a1 to <16 x i16>
  %4 = add <16 x i16> %2, %3
  %5 = add <16 x i16> %4, <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>
  %6 = lshr <16 x i16> %5, <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>
  %7 = trunc <16 x i16> %6 to <16 x i8>
  ret <16 x i8> %7
}

define <8 x i16> @stack_fold_pavgw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pavgw
  ;CHECK:       pavgw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = zext <8 x i16> %a0 to <8 x i32>
  %3 = zext <8 x i16> %a1 to <8 x i32>
  %4 = add <8 x i32> %2, %3
  %5 = add <8 x i32> %4, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
  %6 = lshr <8 x i32> %5, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
  %7 = trunc <8 x i32> %6 to <8 x i16>
  ret <8 x i16> %7
}

define <16 x i8> @stack_fold_pblendvb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %c) {
  ;CHECK-LABEL: stack_fold_pblendvb
  ;CHECK:       pblendvb %xmm0, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a1, <16 x i8> %c, <16 x i8> %a0)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone

define <8 x i16> @stack_fold_pblendw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pblendw
  ;CHECK:       pblendw $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> %a1, <8 x i32> <i32 8, i32 9, i32 10, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %2
}

define <2 x i64> @stack_fold_pclmulqdq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_pclmulqdq
  ;CHECK:       pclmulqdq $0, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %a0, <2 x i64> %a1, i8 0)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.pclmulqdq(<2 x i64>, <2 x i64>, i8) nounwind readnone

define <16 x i8> @stack_fold_pcmpeqb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpeqb
  ;CHECK:       pcmpeqb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp eq <16 x i8> %a0, %a1
  %3 = sext <16 x i1> %2 to <16 x i8>
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pcmpeqd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpeqd
  ;CHECK:       pcmpeqd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp eq <4 x i32> %a0, %a1
  %3 = sext <4 x i1> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pcmpeqq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpeqq
  ;CHECK:       pcmpeqq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp eq <2 x i64> %a0, %a1
  %3 = sext <2 x i1> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_pcmpeqw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpeqw
  ;CHECK:       pcmpeqw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp eq <8 x i16> %a0, %a1
  %3 = sext <8 x i1> %2 to <8 x i16>
  ret <8 x i16> %3
}

define i32 @stack_fold_pcmpestri(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpestri
  ;CHECK:       pcmpestri $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{rax},~{flags}"()
  %2 = call i32 @llvm.x86.sse42.pcmpestri128(<16 x i8> %a0, i32 7, <16 x i8> %a1, i32 7, i8 7)
  ret i32 %2
}
declare i32 @llvm.x86.sse42.pcmpestri128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone

define <16 x i8> @stack_fold_pcmpestrm(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpestrm
  ;CHECK:       pcmpestrm $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{rax},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8> %a0, i32 7, <16 x i8> %a1, i32 7, i8 7)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse42.pcmpestrm128(<16 x i8>, i32, <16 x i8>, i32, i8) nounwind readnone

define <16 x i8> @stack_fold_pcmpgtb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpgtb
  ;CHECK:       pcmpgtb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <16 x i8> %a0, %a1
  %3 = sext <16 x i1> %2 to <16 x i8>
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pcmpgtd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpgtd
  ;CHECK:       pcmpgtd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <4 x i32> %a0, %a1
  %3 = sext <4 x i1> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pcmpgtq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpgtq
  ;CHECK:       pcmpgtq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <2 x i64> %a0, %a1
  %3 = sext <2 x i1> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_pcmpgtw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpgtw
  ;CHECK:       pcmpgtw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <8 x i16> %a0, %a1
  %3 = sext <8 x i1> %2 to <8 x i16>
  ret <8 x i16> %3
}

define i32 @stack_fold_pcmpistri(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpistri
  ;CHECK:       pcmpistri $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call i32 @llvm.x86.sse42.pcmpistri128(<16 x i8> %a0, <16 x i8> %a1, i8 7)
  ret i32 %2
}
declare i32 @llvm.x86.sse42.pcmpistri128(<16 x i8>, <16 x i8>, i8) nounwind readnone

define <16 x i8> @stack_fold_pcmpistrm(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pcmpistrm
  ;CHECK:       pcmpistrm $7, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %a1, i8 7)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8>, <16 x i8>, i8) nounwind readnone

; TODO stack_fold_pextrb

; We can't naively fold pextrw as it only writes to a 16-bit memory location
; even though it can store to a 32-bit register.
define i16 @stack_fold_pextrw(<8 x i16> %a0) {
; CHECK-LABEL: stack_fold_pextrw
; CHECK:       pextrw $1, {{%xmm[0-9][0-9]*}}, %[[GPR32:(e[a-z]+|r[0-9]+d)]]
; CHECK:       movl %[[GPR32]], {{-?[0-9]*}}(%rsp) {{.*#+}} 4-byte Spill
; CHECK:       movl {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 4-byte Reload
entry:
; add forces execution domain
  %add = add <8 x i16> %a0, <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>
  %extract = extractelement <8 x i16> %add, i32 1
  %asm = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i16 %extract
}

define i32 @stack_fold_pextrd(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_pextrd
  ;CHECK:       pextrd $1, {{%xmm[0-9][0-9]*}}, {{-?[0-9]*}}(%rsp) {{.*#+}} 4-byte Folded Spill
  ;CHECK:       movl    {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 4-byte Reload
  ; add forces execution domain
  %1 = add <4 x i32> %a0, <i32 1, i32 2, i32 3, i32 4>
  %2 = extractelement <4 x i32> %1, i32 1
  %3 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i32 %2
}

define i64 @stack_fold_pextrq(<2 x i64> %a0) {
  ;CHECK-LABEL: stack_fold_pextrq
  ;CHECK:       pextrq $1, {{%xmm[0-9][0-9]*}}, {{-?[0-9]*}}(%rsp) {{.*#+}} 8-byte Folded Spill
  ;CHECK:       movq    {{-?[0-9]*}}(%rsp), %rax {{.*#+}} 8-byte Reload
  %1 = extractelement <2 x i64> %a0, i32 1
  %2 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i64 %1
}

define <4 x i32> @stack_fold_phaddd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_phaddd
  ;CHECK:       phaddd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.ssse3.phadd.d.128(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.ssse3.phadd.d.128(<4 x i32>, <4 x i32>) nounwind readnone

define <8 x i16> @stack_fold_phaddsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_phaddsw
  ;CHECK:       phaddsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.phadd.sw.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.phadd.sw.128(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_phaddw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_phaddw
  ;CHECK:       phaddw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.phadd.w.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.phadd.w.128(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_phminposuw(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_phminposuw
  ;CHECK:       phminposuw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16> %a0)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16>) nounwind readnone

define <4 x i32> @stack_fold_phsubd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_phsubd
  ;CHECK:       phsubd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.ssse3.phsub.d.128(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.ssse3.phsub.d.128(<4 x i32>, <4 x i32>) nounwind readnone

define <8 x i16> @stack_fold_phsubsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_phsubsw
  ;CHECK:       phsubsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.phsub.sw.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.phsub.sw.128(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_phsubw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_phsubw
  ;CHECK:       phsubw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.phsub.w.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.phsub.w.128(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_pinsrb(<16 x i8> %a0, i8 %a1) {
  ;CHECK-LABEL: stack_fold_pinsrb
  ;CHECK:       pinsrb $1, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 4-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = insertelement <16 x i8> %a0, i8 %a1, i32 1
  ret <16 x i8> %2
}

define <4 x i32> @stack_fold_pinsrd(<4 x i32> %a0, i32 %a1) {
  ;CHECK-LABEL: stack_fold_pinsrd
  ;CHECK:       pinsrd $1, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 4-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = insertelement <4 x i32> %a0, i32 %a1, i32 1
  ret <4 x i32> %2
}

define <2 x i64> @stack_fold_pinsrq(<2 x i64> %a0, i64 %a1) {
  ;CHECK-LABEL: stack_fold_pinsrq
  ;CHECK:       pinsrq $1, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 8-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = insertelement <2 x i64> %a0, i64 %a1, i32 1
  ret <2 x i64> %2
}

define <8 x i16> @stack_fold_pinsrw(<8 x i16> %a0, i16 %a1) {
  ;CHECK-LABEL: stack_fold_pinsrw
  ;CHECK:       pinsrw $1, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 4-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  %2 = insertelement <8 x i16> %a0, i16 %a1, i32 1
  ret <8 x i16> %2
}

define <8 x i16> @stack_fold_pmaddubsw(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pmaddubsw
  ;CHECK:       pmaddubsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a0, <16 x i8> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8>, <16 x i8>) nounwind readnone

define <4 x i32> @stack_fold_pmaddwd(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmaddwd
  ;CHECK:       pmaddwd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_pmaxsb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxsb
  ;CHECK:       pmaxsb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <16 x i8> %a0, %a1
  %3 = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %a1
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pmaxsd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxsd
  ;CHECK:       pmaxsd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <4 x i32> %a0, %a1
  %3 = select <4 x i1> %2, <4 x i32> %a0, <4 x i32> %a1
  ret <4 x i32> %3
}

define <8 x i16> @stack_fold_pmaxsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxsw
  ;CHECK:       pmaxsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp sgt <8 x i16> %a0, %a1
  %3 = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %a1
  ret <8 x i16> %3
}

define <16 x i8> @stack_fold_pmaxub(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxub
  ;CHECK:       pmaxub {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ugt <16 x i8> %a0, %a1
  %3 = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %a1
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pmaxud(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxud
  ;CHECK:       pmaxud {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ugt <4 x i32> %a0, %a1
  %3 = select <4 x i1> %2, <4 x i32> %a0, <4 x i32> %a1
  ret <4 x i32> %3
}

define <8 x i16> @stack_fold_pmaxuw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmaxuw
  ;CHECK:       pmaxuw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ugt <8 x i16> %a0, %a1
  %3 = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %a1
  ret <8 x i16> %3
}

define <16 x i8> @stack_fold_pminsb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pminsb
  ;CHECK:       pminsb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp slt <16 x i8> %a0, %a1
  %3 = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %a1
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pminsd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pminsd
  ;CHECK:       pminsd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp slt <4 x i32> %a0, %a1
  %3 = select <4 x i1> %2, <4 x i32> %a0, <4 x i32> %a1
  ret <4 x i32> %3
}

define <8 x i16> @stack_fold_pminsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pminsw
  ;CHECK:       pminsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp slt <8 x i16> %a0, %a1
  %3 = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %a1
  ret <8 x i16> %3
}

define <16 x i8> @stack_fold_pminub(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pminub
  ;CHECK:       pminub {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ult <16 x i8> %a0, %a1
  %3 = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %a1
  ret <16 x i8> %3
}

define <4 x i32> @stack_fold_pminud(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pminud
  ;CHECK:       pminud {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ult <4 x i32> %a0, %a1
  %3 = select <4 x i1> %2, <4 x i32> %a0, <4 x i32> %a1
  ret <4 x i32> %3
}

define <8 x i16> @stack_fold_pminuw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pminuw
  ;CHECK:       pminuw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = icmp ult <8 x i16> %a0, %a1
  %3 = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %a1
  ret <8 x i16> %3
}

define <4 x i32> @stack_fold_pmovsxbd(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxbd
  ;CHECK:       pmovsxbd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %3 = sext <4 x i8> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pmovsxbq(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxbq
  ;CHECK:       pmovsxbq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> undef, <2 x i32> <i32 0, i32 1>
  %3 = sext <2 x i8> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_pmovsxbw(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxbw
  ;CHECK:       pmovsxbw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %3 = sext <8 x i8> %2 to <8 x i16>
  ret <8 x i16> %3
}

define <2 x i64> @stack_fold_pmovsxdq(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxdq
  ;CHECK:       pmovsxdq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <4 x i32> %a0, <4 x i32> undef, <2 x i32> <i32 0, i32 1>
  %3 = sext <2 x i32> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <4 x i32> @stack_fold_pmovsxwd(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxwd
  ;CHECK:       pmovsxwd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %3 = sext <4 x i16> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pmovsxwq(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pmovsxwq
  ;CHECK:       pmovsxwq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
  %3 = sext <2 x i16> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <4 x i32> @stack_fold_pmovzxbd(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxbd
  ;CHECK:       pmovzxbd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 17, i32 18, i32 1, i32 19, i32 20, i32 21, i32 2, i32 22, i32 23, i32 24, i32 3, i32 25, i32 26, i32 27>
  %3 = bitcast <16 x i8> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pmovzxbq(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxbq
  ;CHECK:       pmovzxbq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 1, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28>
  %3 = bitcast <16 x i8> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_pmovzxbw(<16 x i8> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxbw
  ;CHECK:       pmovzxbw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
  %3 = bitcast <16 x i8> %2 to <8 x i16>
  ret <8 x i16> %3
}

define <2 x i64> @stack_fold_pmovzxdq(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxdq
  ;CHECK:       pmovzxdq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <4 x i32> %a0, <4 x i32> zeroinitializer, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %3 = bitcast <4 x i32> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <4 x i32> @stack_fold_pmovzxwd(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxwd
  ;CHECK:       pmovzxwd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> zeroinitializer, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
  %3 = bitcast <8 x i16> %2 to <4 x i32>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_pmovzxwq(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pmovzxwq
  ;CHECK:       pmovzxwq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> zeroinitializer, <8 x i32> <i32 0, i32 8, i32 9, i32 10, i32 1, i32 11, i32 12, i32 13>
  %3 = bitcast <8 x i16> %2 to <2 x i64>
  ret <2 x i64> %3
}

define <2 x i64> @stack_fold_pmuldq(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pmuldq
  ;CHECK:       pmuldq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = bitcast <4 x i32> %a0 to <2 x i64>
  %3 = bitcast <4 x i32> %a1 to <2 x i64>
  %4 = shl <2 x i64> %2, <i64 32, i64 32>
  %5 = ashr <2 x i64> %4, <i64 32, i64 32>
  %6 = shl <2 x i64> %3, <i64 32, i64 32>
  %7 = ashr <2 x i64> %6, <i64 32, i64 32>
  %8 = mul <2 x i64> %5, %7
  ret <2 x i64> %8
}

define <8 x i16> @stack_fold_pmulhrsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmulhrsw
  ;CHECK:       pmulhrsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.pmul.hr.sw.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.pmul.hr.sw.128(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_pmulhuw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmulhuw
  ;CHECK:       pmulhuw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_pmulhw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmulhw
  ;CHECK:       pmulhw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16>, <8 x i16>) nounwind readnone

define <4 x i32> @stack_fold_pmulld(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pmulld
  ;CHECK:       pmulld {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = mul <4 x i32> %a0, %a1
  ret <4 x i32> %2
}

define <8 x i16> @stack_fold_pmullw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_pmullw
  ;CHECK:       pmullw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = mul <8 x i16> %a0, %a1
  ret <8 x i16> %2
}

define <2 x i64> @stack_fold_pmuludq(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pmuludq
  ;CHECK:       pmuludq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = bitcast <4 x i32> %a0 to <2 x i64>
  %3 = bitcast <4 x i32> %a1 to <2 x i64>
  %4 = and <2 x i64> %2, <i64 4294967295, i64 4294967295>
  %5 = and <2 x i64> %3, <i64 4294967295, i64 4294967295>
  %6 = mul <2 x i64> %4, %5
  ret <2 x i64> %6
}

define <16 x i8> @stack_fold_por(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_por
  ;CHECK:       por {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = or <16 x i8> %a0, %a1
  ; add forces execution domain
  %3 = add <16 x i8> %2, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
  ret <16 x i8> %3
}

define <2 x i64> @stack_fold_psadbw(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_psadbw
  ;CHECK:       psadbw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8> %a0, <16 x i8> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8>, <16 x i8>) nounwind readnone

define <16 x i8> @stack_fold_pshufb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pshufb
  ;CHECK:       pshufb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8>, <16 x i8>) nounwind readnone

define <4 x i32> @stack_fold_pshufd(<4 x i32> %a0) {
  ;CHECK-LABEL: stack_fold_pshufd
  ;CHECK:       pshufd $27, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <4 x i32> %a0, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
  ret <4 x i32> %2
}

define <8 x i16> @stack_fold_pshufhw(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pshufhw
  ;CHECK:       pshufhw $11, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 6, i32 4, i32 4>
  ret <8 x i16> %2
}

define <8 x i16> @stack_fold_pshuflw(<8 x i16> %a0) {
  ;CHECK-LABEL: stack_fold_pshuflw
  ;CHECK:       pshuflw $27, {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %2
}

define <16 x i8> @stack_fold_psignb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_psignb
  ;CHECK:       psignb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.ssse3.psign.b.128(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.ssse3.psign.b.128(<16 x i8>, <16 x i8>) nounwind readnone

define <4 x i32> @stack_fold_psignd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_psignd
  ;CHECK:       psignd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.ssse3.psign.d.128(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.ssse3.psign.d.128(<4 x i32>, <4 x i32>) nounwind readnone

define <8 x i16> @stack_fold_psignw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psignw
  ;CHECK:       psignw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.ssse3.psign.w.128(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.ssse3.psign.w.128(<8 x i16>, <8 x i16>) nounwind readnone

define <4 x i32> @stack_fold_pslld(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_pslld
  ;CHECK:       pslld {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.sse2.psll.d(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.sse2.psll.d(<4 x i32>, <4 x i32>) nounwind readnone

define <2 x i64> @stack_fold_psllq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_psllq
  ;CHECK:       psllq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.sse2.psll.q(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.sse2.psll.q(<2 x i64>, <2 x i64>) nounwind readnone

define <8 x i16> @stack_fold_psllw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psllw
  ;CHECK:       psllw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.psll.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.psll.w(<8 x i16>, <8 x i16>) nounwind readnone

define <4 x i32> @stack_fold_psrad(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_psrad
  ;CHECK:       psrad {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.sse2.psra.d(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.sse2.psra.d(<4 x i32>, <4 x i32>) nounwind readnone

define <8 x i16> @stack_fold_psraw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psraw
  ;CHECK:       psraw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.psra.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.psra.w(<8 x i16>, <8 x i16>) nounwind readnone

define <4 x i32> @stack_fold_psrld(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_psrld
  ;CHECK:       psrld {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <4 x i32> @llvm.x86.sse2.psrl.d(<4 x i32> %a0, <4 x i32> %a1)
  ret <4 x i32> %2
}
declare <4 x i32> @llvm.x86.sse2.psrl.d(<4 x i32>, <4 x i32>) nounwind readnone

define <2 x i64> @stack_fold_psrlq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_psrlq
  ;CHECK:       psrlq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <2 x i64> @llvm.x86.sse2.psrl.q(<2 x i64> %a0, <2 x i64> %a1)
  ret <2 x i64> %2
}
declare <2 x i64> @llvm.x86.sse2.psrl.q(<2 x i64>, <2 x i64>) nounwind readnone

define <8 x i16> @stack_fold_psrlw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psrlw
  ;CHECK:       psrlw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_psubb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_psubb
  ;CHECK:       psubb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = sub <16 x i8> %a0, %a1
  ret <16 x i8> %2
}

define <4 x i32> @stack_fold_psubd(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_psubd
  ;CHECK:       psubd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = sub <4 x i32> %a0, %a1
  ret <4 x i32> %2
}

define <2 x i64> @stack_fold_psubq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_psubq
  ;CHECK:       psubq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = sub <2 x i64> %a0, %a1
  ret <2 x i64> %2
}

define <16 x i8> @stack_fold_psubsb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_psubsb
  ;CHECK:       psubsb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone

define <8 x i16> @stack_fold_psubsw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psubsw
  ;CHECK:       psubsw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone

define <16 x i8> @stack_fold_psubusb(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_psubusb
  ;CHECK:       psubusb {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8> %a0, <16 x i8> %a1)
  ret <16 x i8> %2
}
declare <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8>, <16 x i8>) nounwind readnone

define <8 x i16> @stack_fold_psubusw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psubusw
  ;CHECK:       psubusw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16>, <8 x i16>) nounwind readnone

define <8 x i16> @stack_fold_psubw(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_psubw
  ;CHECK:       psubw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = sub <8 x i16> %a0, %a1
  ret <8 x i16> %2
}

define i32 @stack_fold_ptest(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_ptest
  ;CHECK:       ptest {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call i32 @llvm.x86.sse41.ptestc(<2 x i64> %a0, <2 x i64> %a1)
  ret i32 %2
}
declare i32 @llvm.x86.sse41.ptestc(<2 x i64>, <2 x i64>) nounwind readnone

define <16 x i8> @stack_fold_punpckhbw(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_punpckhbw
  ;CHECK:       punpckhbw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> %a1, <16 x i32> <i32 8, i32 24, i32 9, i32 25, i32 10, i32 26, i32 11, i32 27, i32 12, i32 28, i32 13, i32 29, i32 14, i32 30, i32 15, i32 31>
  ret <16 x i8> %2
}

define <4 x i32> @stack_fold_punpckhdq(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_punpckhdq
  ;CHECK:       punpckhdq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <4 x i32> %a0, <4 x i32> %a1, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  ; add forces execution domain
  %3 = add <4 x i32> %2, <i32 1, i32 1, i32 1, i32 1>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_punpckhqdq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_punpckhqdq
  ;CHECK:       punpckhqdq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <2 x i64> %a0, <2 x i64> %a1, <2 x i32> <i32 1, i32 3>
  ; add forces execution domain
  %3 = add <2 x i64> %2, <i64 1, i64 1>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_punpckhwd(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_punpckhwd
  ;CHECK:       punpckhwd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> %a1, <8 x i32> <i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
  ret <8 x i16> %2
}

define <16 x i8> @stack_fold_punpcklbw(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_punpcklbw
  ;CHECK:       punpcklbw {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <16 x i8> %a0, <16 x i8> %a1, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
  ret <16 x i8> %2
}

define <4 x i32> @stack_fold_punpckldq(<4 x i32> %a0, <4 x i32> %a1) {
  ;CHECK-LABEL: stack_fold_punpckldq
  ;CHECK:       punpckldq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <4 x i32> %a0, <4 x i32> %a1, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  ; add forces execution domain
  %3 = add <4 x i32> %2, <i32 1, i32 1, i32 1, i32 1>
  ret <4 x i32> %3
}

define <2 x i64> @stack_fold_punpcklqdq(<2 x i64> %a0, <2 x i64> %a1) {
  ;CHECK-LABEL: stack_fold_punpcklqdq
  ;CHECK:       punpcklqdq {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <2 x i64> %a0, <2 x i64> %a1, <2 x i32> <i32 0, i32 2>
  ; add forces execution domain
  %3 = add <2 x i64> %2, <i64 1, i64 1>
  ret <2 x i64> %3
}

define <8 x i16> @stack_fold_punpcklwd(<8 x i16> %a0, <8 x i16> %a1) {
  ;CHECK-LABEL: stack_fold_punpcklwd
  ;CHECK:       punpcklwd {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = shufflevector <8 x i16> %a0, <8 x i16> %a1, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
  ret <8 x i16> %2
}

define <16 x i8> @stack_fold_pxor(<16 x i8> %a0, <16 x i8> %a1) {
  ;CHECK-LABEL: stack_fold_pxor
  ;CHECK:       pxor {{-?[0-9]*}}(%rsp), {{%xmm[0-9][0-9]*}} {{.*#+}} 16-byte Folded Reload
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = xor <16 x i8> %a0, %a1
  ; add forces execution domain
  %3 = add <16 x i8> %2, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
  ret <16 x i8> %3
}