; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+mmx,+ssse3 | FileCheck %s

define x86_mmx @stack_fold_cvtpd2pi(<2 x double> %a0) {
; CHECK-LABEL: stack_fold_cvtpd2pi:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvtpd2pi {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 16-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call x86_mmx @llvm.x86.sse.cvtpd2pi(<2 x double> %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.sse.cvtpd2pi(<2 x double>) nounwind readnone

define <2 x double> @stack_fold_cvtpi2pd(x86_mmx %a0) {
; CHECK-LABEL: stack_fold_cvtpi2pd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvtpi2pd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call <2 x double> @llvm.x86.sse.cvtpi2pd(x86_mmx %a0) nounwind readnone
  ret <2 x double> %2
}
declare <2 x double> @llvm.x86.sse.cvtpi2pd(x86_mmx) nounwind readnone

define <4 x float> @stack_fold_cvtpi2ps(<4 x float> %a0, x86_mmx %a1) {
; CHECK-LABEL: stack_fold_cvtpi2ps:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvtpi2ps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call <4 x float> @llvm.x86.sse.cvtpi2ps(<4 x float> %a0, x86_mmx %a1) nounwind readnone
  ret <4 x float> %2
}
declare <4 x float> @llvm.x86.sse.cvtpi2ps(<4 x float>, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_cvtps2pi(<4 x float> %a0) {
; CHECK-LABEL: stack_fold_cvtps2pi:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvtps2pi {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 16-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call x86_mmx @llvm.x86.sse.cvtps2pi(<4 x float> %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.sse.cvtps2pi(<4 x float>) nounwind readnone

define x86_mmx @stack_fold_cvttpd2pi(<2 x double> %a0) {
; CHECK-LABEL: stack_fold_cvttpd2pi:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvttpd2pi {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 16-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call x86_mmx @llvm.x86.sse.cvttpd2pi(<2 x double> %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.sse.cvttpd2pi(<2 x double>) nounwind readnone

define x86_mmx @stack_fold_cvttps2pi(<4 x float> %a0) {
; CHECK-LABEL: stack_fold_cvttps2pi:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    cvttps2pi {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 16-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2 = call x86_mmx @llvm.x86.sse.cvttps2pi(<4 x float> %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.sse.cvttps2pi(<4 x float>) nounwind readnone

; TODO stack_fold_movd_load

; padd forces execution on mmx
define i32 @stack_fold_movd_store(x86_mmx %a0) nounwind {
; CHECK-LABEL: stack_fold_movd_store:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pushq %rbp
; CHECK-NEXT:    pushq %r15
; CHECK-NEXT:    pushq %r14
; CHECK-NEXT:    pushq %r13
; CHECK-NEXT:    pushq %r12
; CHECK-NEXT:    pushq %rbx
; CHECK-NEXT:    paddb %mm0, %mm0
; CHECK-NEXT:    movd %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    movl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 4-byte Reload
; CHECK-NEXT:    popq %rbx
; CHECK-NEXT:    popq %r12
; CHECK-NEXT:    popq %r13
; CHECK-NEXT:    popq %r14
; CHECK-NEXT:    popq %r15
; CHECK-NEXT:    popq %rbp
; CHECK-NEXT:    retq
  %1 = call x86_mmx @llvm.x86.mmx.padd.b(x86_mmx %a0, x86_mmx %a0)
  %2 = bitcast x86_mmx %1 to <2 x i32>
  %3 = extractelement <2 x i32> %2, i32 0
  %4 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i32 %3
}

; TODO stack_fold_movq_load

; padd forces execution on mmx
define i64 @stack_fold_movq_store(x86_mmx %a0) nounwind {
; CHECK-LABEL: stack_fold_movq_store:
; CHECK:       # %bb.0:
; CHECK-NEXT:    pushq %rbp
; CHECK-NEXT:    pushq %r15
; CHECK-NEXT:    pushq %r14
; CHECK-NEXT:    pushq %r13
; CHECK-NEXT:    pushq %r12
; CHECK-NEXT:    pushq %rbx
; CHECK-NEXT:    paddb %mm0, %mm0
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; CHECK-NEXT:    popq %rbx
; CHECK-NEXT:    popq %r12
; CHECK-NEXT:    popq %r13
; CHECK-NEXT:    popq %r14
; CHECK-NEXT:    popq %r15
; CHECK-NEXT:    popq %rbp
; CHECK-NEXT:    retq
  %1 = call x86_mmx @llvm.x86.mmx.padd.b(x86_mmx %a0, x86_mmx %a0)
  %2 = bitcast x86_mmx %1 to i64
  %3 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"()
  ret i64 %2
}

define x86_mmx @stack_fold_pabsb(x86_mmx %a0) {
; CHECK-LABEL: stack_fold_pabsb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pabsb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pabs.b(x86_mmx %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pabs.b(x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pabsd(x86_mmx %a0) {
; CHECK-LABEL: stack_fold_pabsd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pabsd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pabs.d(x86_mmx %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pabs.d(x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pabsw(x86_mmx %a0) {
; CHECK-LABEL: stack_fold_pabsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pabsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pabs.w(x86_mmx %a0) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pabs.w(x86_mmx) nounwind readnone

define x86_mmx @stack_fold_packssdw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_packssdw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    packssdw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.packssdw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.packssdw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_packsswb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_packsswb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    packsswb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.packsswb(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.packsswb(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_packuswb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_packuswb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    packuswb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.packuswb(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.packuswb(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padd.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padd.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padd.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padd.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padd.q(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padd.q(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddsb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddsb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddsb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padds.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padds.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padds.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padds.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddusb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddusb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddusb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.paddus.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.paddus.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddusw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddusw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddusw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.paddus.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.paddus.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_paddw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_paddw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    paddw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.padd.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.padd.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_palignr(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_palignr:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    palignr $1, {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.palignr.b(x86_mmx %a, x86_mmx %b, i8 1) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.palignr.b(x86_mmx, x86_mmx, i8) nounwind readnone

define x86_mmx @stack_fold_pand(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pand:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pand {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pand(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pand(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pandn(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pandn:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pandn {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pandn(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pandn(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pavgb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pavgb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pavgb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pavg.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pavg.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pavgw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pavgw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pavgw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pavg.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pavg.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpeqb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpeqb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpeqb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpeq.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpeq.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpeqd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpeqd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpeqd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpeq.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpeq.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpeqw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpeqw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpeqw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpeq.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpeq.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpgtb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpgtb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpgtb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpgt.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpgt.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpgtd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpgtd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpgtd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpgt.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpgt.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pcmpgtw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pcmpgtw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pcmpgtw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pcmpgt.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pcmpgt.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phaddd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phaddd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phaddd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phadd.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phadd.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phaddsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phaddsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phaddsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phadd.sw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phadd.sw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phaddw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phaddw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phaddw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phadd.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phadd.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phsubd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phsubd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phsubd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phsub.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phsub.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phsubsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phsubsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phsubsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phsub.sw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phsub.sw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_phsubw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_phsubw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    phsubw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.phsub.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.phsub.w(x86_mmx, x86_mmx) nounwind readnone

; TODO stack_fold_pinsrw

define x86_mmx @stack_fold_pmaddubsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmaddubsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmaddubsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pmadd.ub.sw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pmadd.ub.sw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmaddwd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmaddwd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmaddwd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmadd.wd(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmadd.wd(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmaxsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmaxsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmaxsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmaxs.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmaxs.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmaxub(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmaxub:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmaxub {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmaxu.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmaxu.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pminsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pminsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pminsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmins.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmins.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pminub(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pminub:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pminub {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pminu.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pminu.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmulhrsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmulhrsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmulhrsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pmul.hr.sw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pmul.hr.sw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmulhuw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmulhuw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmulhuw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmulhu.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmulhu.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmulhw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmulhw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmulhw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmulh.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmulh.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmullw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmullw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmullw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmull.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmull.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pmuludq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pmuludq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pmuludq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pmulu.dq(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pmulu.dq(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_por(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_por:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    por {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.por(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.por(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psadbw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psadbw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psadbw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psad.bw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psad.bw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pshufb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pshufb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Reload
; CHECK-NEXT:    pshufb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.pshuf.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.pshuf.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pshufw(x86_mmx %a) {
; CHECK-LABEL: stack_fold_pshufw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pshufw $1, {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mem[1,0,0,0]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm1},~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %a, i8 1) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8) nounwind readnone

define x86_mmx @stack_fold_psignb(x86_mmx %a0, x86_mmx %a1) {
; CHECK-LABEL: stack_fold_psignb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psignb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.psign.b(x86_mmx %a0, x86_mmx %a1) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.psign.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psignd(x86_mmx %a0, x86_mmx %a1) {
; CHECK-LABEL: stack_fold_psignd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psignd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.psign.d(x86_mmx %a0, x86_mmx %a1) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.psign.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psignw(x86_mmx %a0, x86_mmx %a1) {
; CHECK-LABEL: stack_fold_psignw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psignw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.ssse3.psign.w(x86_mmx %a0, x86_mmx %a1) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.ssse3.psign.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pslld(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pslld:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pslld {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psll.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psll.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psllq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psllq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psllq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psll.q(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psll.q(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psllw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psllw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psllw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psll.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psll.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psrad(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psrad:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psrad {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psra.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psra.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psraw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psraw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psraw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psra.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psra.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psrld(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psrld:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psrld {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psrl.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psrl.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psrlq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psrlq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psrlq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psrl.q(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psrl.q(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psrlw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psrlw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psrlw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psrl.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psrl.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psub.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psub.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psub.d(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psub.d(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psub.q(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psub.q(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubsb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubsb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubsb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psubs.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psubs.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubsw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubsw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubsw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psubs.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psubs.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubusb(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubusb:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubusb {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psubus.b(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psubus.b(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubusw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubusw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubusw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psubus.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psubus.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_psubw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_psubw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    psubw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.psub.w(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.psub.w(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpckhbw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpckhbw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[4],mem[4],mm0[5],mem[5],mm0[6],mem[6],mm0[7],mem[7]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpckhbw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpckhbw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpckhdq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpckhdq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpckhdq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[1],mem[1]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpckhdq(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpckhdq(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpckhwd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpckhwd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[2],mem[2],mm0[3],mem[3]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpckhwd(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpckhwd(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpcklbw(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpcklbw:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[0],mem[0],mm0[1],mem[1],mm0[2],mem[2],mm0[3],mem[3]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpcklbw(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpcklbw(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpckldq(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpckldq:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[0],mem[0]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpckldq(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpckldq(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_punpcklwd(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_punpcklwd:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    # mm0 = mm0[0],mem[0],mm0[1],mem[1]
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.punpcklwd(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.punpcklwd(x86_mmx, x86_mmx) nounwind readnone

define x86_mmx @stack_fold_pxor(x86_mmx %a, x86_mmx %b) {
; CHECK-LABEL: stack_fold_pxor:
; CHECK:       # %bb.0:
; CHECK-NEXT:    movq %mm1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT:    #APP
; CHECK-NEXT:    nop
; CHECK-NEXT:    #NO_APP
; CHECK-NEXT:    pxor {{[-0-9]+}}(%r{{[sb]}}p), %mm0 # 8-byte Folded Reload
; CHECK-NEXT:    movq2dq %mm0, %xmm0
; CHECK-NEXT:    retq
  %1 = tail call x86_mmx asm sideeffect "nop", "=y,~{mm2},~{mm3},~{mm4},~{mm5},~{mm6},~{mm7}"()
  %2 = call x86_mmx @llvm.x86.mmx.pxor(x86_mmx %a, x86_mmx %b) nounwind readnone
  ret x86_mmx %2
}
declare x86_mmx @llvm.x86.mmx.pxor(x86_mmx, x86_mmx) nounwind readnone