nuudlman/llvm: llvm/test/CodeGen/X86/midpoint-int-vec-128.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX1-FALLBACK
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX2-FALLBACK
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop | FileCheck %s --check-prefixes=XOP,XOP-FALLBACK
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOP,XOPAVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOP,XOPAVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512VL,AVX512VL-FALLBACK
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=AVX512BW-FALLBACK
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512VL,AVX512VLBW

; These test cases are inspired by C++2a std::midpoint().
; See https://bugs.llvm.org/show_bug.cgi?id=40965

; Using 128-bit vector regs.

; ---------------------------------------------------------------------------- ;
; 32-bit width. 128 / 32 = 4 elts.
; ---------------------------------------------------------------------------- ;

; Values come from regs

define <4 x i32> @vec128_i32_signed_reg_reg(<4 x i32> %a1, <4 x i32> %a2) nounwind {
; SSE2-LABEL: vec128_i32_signed_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa %xmm0, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pcmpgtd %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubd %xmm4, %xmm2
; SSE2-NEXT:    psrld $1, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm3, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm1, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    paddd %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i32_signed_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm1, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pminsd %xmm1, %xmm3
; SSE41-NEXT:    pmaxsd %xmm0, %xmm1
; SSE41-NEXT:    psubd %xmm3, %xmm1
; SSE41-NEXT:    psrld $1, %xmm1
; SSE41-NEXT:    pmulld %xmm1, %xmm2
; SSE41-NEXT:    paddd %xmm2, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i32_signed_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i32_signed_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX2-FALLBACK-NEXT:    vpor %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i32_signed_reg_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i32_signed_reg_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i32_signed_reg_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; XOPAVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i32_signed_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512F-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i32_signed_reg_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpcmpgtd %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubd %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i32_signed_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %t3 = icmp sgt <4 x i32> %a1, %a2 ; signed
  %t4 = select <4 x i1> %t3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  %t5 = select <4 x i1> %t3, <4 x i32> %a2, <4 x i32> %a1
  %t6 = select <4 x i1> %t3, <4 x i32> %a1, <4 x i32> %a2
  %t7 = sub <4 x i32> %t6, %t5
  %t8 = lshr <4 x i32> %t7, <i32 1, i32 1, i32 1, i32 1>
  %t9 = mul nsw <4 x i32> %t8, %t4 ; signed
  %a10 = add nsw <4 x i32> %t9, %a1 ; signed
  ret <4 x i32> %a10
}

define <4 x i32> @vec128_i32_unsigned_reg_reg(<4 x i32> %a1, <4 x i32> %a2) nounwind {
; SSE2-LABEL: vec128_i32_unsigned_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; SSE2-NEXT:    movdqa %xmm1, %xmm3
; SSE2-NEXT:    pxor %xmm2, %xmm3
; SSE2-NEXT:    pxor %xmm0, %xmm2
; SSE2-NEXT:    movdqa %xmm2, %xmm4
; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [1,1,1,1]
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    pcmpgtd %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm2
; SSE2-NEXT:    pand %xmm3, %xmm2
; SSE2-NEXT:    pandn %xmm1, %xmm3
; SSE2-NEXT:    por %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm2
; SSE2-NEXT:    pand %xmm4, %xmm2
; SSE2-NEXT:    pandn %xmm1, %xmm4
; SSE2-NEXT:    por %xmm2, %xmm4
; SSE2-NEXT:    psubd %xmm3, %xmm4
; SSE2-NEXT:    psrld $1, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm5, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[0,2,2,3]
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm1, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    paddd %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i32_unsigned_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pminud %xmm1, %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pcmpeqd %xmm2, %xmm3
; SSE41-NEXT:    pcmpeqd %xmm4, %xmm4
; SSE41-NEXT:    pxor %xmm3, %xmm4
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
; SSE41-NEXT:    pmaxud %xmm0, %xmm1
; SSE41-NEXT:    psubd %xmm2, %xmm1
; SSE41-NEXT:    psrld $1, %xmm1
; SSE41-NEXT:    pmulld %xmm1, %xmm4
; SSE41-NEXT:    paddd %xmm4, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i32_unsigned_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpminud %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX1-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i32_unsigned_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpminud %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX2-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [1,1,1,1]
; AVX2-FALLBACK-NEXT:    vpor %xmm4, %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i32_unsigned_reg_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vpcomgtud %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminud %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i32_unsigned_reg_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vpcomgtud %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i32_unsigned_reg_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vpcomgtud %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; XOPAVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminud %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i32_unsigned_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpcmpnleud %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512F-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminud %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i32_unsigned_reg_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpcmpnleud %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminud %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubd %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i32_unsigned_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpnleud %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminud %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %t3 = icmp ugt <4 x i32> %a1, %a2
  %t4 = select <4 x i1> %t3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  %t5 = select <4 x i1> %t3, <4 x i32> %a2, <4 x i32> %a1
  %t6 = select <4 x i1> %t3, <4 x i32> %a1, <4 x i32> %a2
  %t7 = sub <4 x i32> %t6, %t5
  %t8 = lshr <4 x i32> %t7, <i32 1, i32 1, i32 1, i32 1>
  %t9 = mul <4 x i32> %t8, %t4
  %a10 = add <4 x i32> %t9, %a1
  ret <4 x i32> %a10
}

; Values are loaded. Only check signed case.

define <4 x i32> @vec128_i32_signed_mem_reg(ptr %a1_addr, <4 x i32> %a2) nounwind {
; SSE2-LABEL: vec128_i32_signed_mem_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa %xmm1, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pcmpgtd %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm0, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm0, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubd %xmm4, %xmm2
; SSE2-NEXT:    psrld $1, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm3, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm4, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-NEXT:    paddd %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i32_signed_mem_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm1
; SSE41-NEXT:    movdqa %xmm1, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm1, %xmm3
; SSE41-NEXT:    pminsd %xmm0, %xmm3
; SSE41-NEXT:    pmaxsd %xmm1, %xmm0
; SSE41-NEXT:    psubd %xmm3, %xmm0
; SSE41-NEXT:    psrld $1, %xmm0
; SSE41-NEXT:    pmulld %xmm2, %xmm0
; SSE41-NEXT:    paddd %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i32_signed_mem_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsd %xmm0, %xmm1, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpsubd %xmm3, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsrld $1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i32_signed_mem_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX2-FALLBACK-NEXT:    vpor %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsd %xmm0, %xmm1, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpsubd %xmm3, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsrld $1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i32_signed_mem_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtd %xmm0, %xmm1, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsd %xmm0, %xmm1, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    vpsubd %xmm3, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    vpsrld $1, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    vpmacsdd %xmm1, %xmm2, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i32_signed_mem_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX1-NEXT:    vpcomgtd %xmm0, %xmm1, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsd %xmm0, %xmm1, %xmm3
; XOPAVX1-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm0, %xmm0
; XOPAVX1-NEXT:    vpsrld $1, %xmm0, %xmm0
; XOPAVX1-NEXT:    vpmacsdd %xmm1, %xmm2, %xmm0, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i32_signed_mem_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX2-NEXT:    vpcomgtd %xmm0, %xmm1, %xmm2
; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; XOPAVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsd %xmm0, %xmm1, %xmm3
; XOPAVX2-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vpsubd %xmm3, %xmm0, %xmm0
; XOPAVX2-NEXT:    vpsrld $1, %xmm0, %xmm0
; XOPAVX2-NEXT:    vpmacsdd %xmm1, %xmm2, %xmm0, %xmm0
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i32_signed_mem_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtd %zmm0, %zmm1, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512F-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsd %xmm0, %xmm1, %xmm2
; AVX512F-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpsubd %xmm2, %xmm0, %xmm0
; AVX512F-NEXT:    vpsrld $1, %xmm0, %xmm0
; AVX512F-NEXT:    vpmulld %xmm3, %xmm0, %xmm0
; AVX512F-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i32_signed_mem_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-NEXT:    vpcmpgtd %xmm0, %xmm1, %k1
; AVX512VL-NEXT:    vpminsd %xmm0, %xmm1, %xmm2
; AVX512VL-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    vpsubd %xmm2, %xmm0, %xmm0
; AVX512VL-NEXT:    vpsrld $1, %xmm0, %xmm0
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubd %xmm0, %xmm2, %xmm0 {%k1}
; AVX512VL-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i32_signed_mem_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtd %zmm0, %zmm1, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsd %xmm0, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsd %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsubd %xmm2, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsrld $1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpmulld %xmm3, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a1 = load <4 x i32>, ptr %a1_addr
  %t3 = icmp sgt <4 x i32> %a1, %a2 ; signed
  %t4 = select <4 x i1> %t3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  %t5 = select <4 x i1> %t3, <4 x i32> %a2, <4 x i32> %a1
  %t6 = select <4 x i1> %t3, <4 x i32> %a1, <4 x i32> %a2
  %t7 = sub <4 x i32> %t6, %t5
  %t8 = lshr <4 x i32> %t7, <i32 1, i32 1, i32 1, i32 1>
  %t9 = mul nsw <4 x i32> %t8, %t4 ; signed
  %a10 = add nsw <4 x i32> %t9, %a1 ; signed
  ret <4 x i32> %a10
}

define <4 x i32> @vec128_i32_signed_reg_mem(<4 x i32> %a1, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i32_signed_reg_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa %xmm0, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pcmpgtd %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubd %xmm4, %xmm2
; SSE2-NEXT:    psrld $1, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm3, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm1, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE2-NEXT:    paddd %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i32_signed_reg_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm1
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm1, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pminsd %xmm1, %xmm3
; SSE41-NEXT:    pmaxsd %xmm0, %xmm1
; SSE41-NEXT:    psubd %xmm3, %xmm1
; SSE41-NEXT:    psrld $1, %xmm1
; SSE41-NEXT:    pmulld %xmm2, %xmm1
; SSE41-NEXT:    paddd %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i32_signed_reg_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i32_signed_reg_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX2-FALLBACK-NEXT:    vpor %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i32_signed_reg_mem:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i32_signed_reg_mem:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX1-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i32_signed_reg_mem:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX2-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; XOPAVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i32_signed_reg_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512F-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i32_signed_reg_mem:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-NEXT:    vpcmpgtd %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubd %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i32_signed_reg_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a2 = load <4 x i32>, ptr %a2_addr
  %t3 = icmp sgt <4 x i32> %a1, %a2 ; signed
  %t4 = select <4 x i1> %t3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  %t5 = select <4 x i1> %t3, <4 x i32> %a2, <4 x i32> %a1
  %t6 = select <4 x i1> %t3, <4 x i32> %a1, <4 x i32> %a2
  %t7 = sub <4 x i32> %t6, %t5
  %t8 = lshr <4 x i32> %t7, <i32 1, i32 1, i32 1, i32 1>
  %t9 = mul nsw <4 x i32> %t8, %t4 ; signed
  %a10 = add nsw <4 x i32> %t9, %a1 ; signed
  ret <4 x i32> %a10
}

define <4 x i32> @vec128_i32_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i32_signed_mem_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa (%rsi), %xmm0
; SSE2-NEXT:    movdqa %xmm1, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pcmpgtd %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm0, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm0, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubd %xmm4, %xmm2
; SSE2-NEXT:    psrld $1, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm3, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,1,3,3]
; SSE2-NEXT:    pmuludq %xmm4, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
; SSE2-NEXT:    paddd %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i32_signed_mem_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm1
; SSE41-NEXT:    movdqa (%rsi), %xmm0
; SSE41-NEXT:    movdqa %xmm1, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm1, %xmm3
; SSE41-NEXT:    pminsd %xmm0, %xmm3
; SSE41-NEXT:    pmaxsd %xmm1, %xmm0
; SSE41-NEXT:    psubd %xmm3, %xmm0
; SSE41-NEXT:    psrld $1, %xmm0
; SSE41-NEXT:    pmulld %xmm2, %xmm0
; SSE41-NEXT:    paddd %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i32_signed_mem_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX1-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i32_signed_mem_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX2-FALLBACK-NEXT:    vpor %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i32_signed_mem_mem:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; XOP-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i32_signed_mem_mem:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm0
; XOPAVX1-NEXT:    vmovdqa (%rsi), %xmm1
; XOPAVX1-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i32_signed_mem_mem:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm0
; XOPAVX2-NEXT:    vmovdqa (%rsi), %xmm1
; XOPAVX2-NEXT:    vpcomgtd %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; XOPAVX2-NEXT:    vpor %xmm3, %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsd %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmacsdd %xmm0, %xmm2, %xmm1, %xmm0
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i32_signed_mem_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512F-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512F-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512F-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i32_signed_mem_mem:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VL-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VL-NEXT:    vpcmpgtd %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubd %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i32_signed_mem_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa32 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsd %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubd %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrld $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a1 = load <4 x i32>, ptr %a1_addr
  %a2 = load <4 x i32>, ptr %a2_addr
  %t3 = icmp sgt <4 x i32> %a1, %a2 ; signed
  %t4 = select <4 x i1> %t3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
  %t5 = select <4 x i1> %t3, <4 x i32> %a2, <4 x i32> %a1
  %t6 = select <4 x i1> %t3, <4 x i32> %a1, <4 x i32> %a2
  %t7 = sub <4 x i32> %t6, %t5
  %t8 = lshr <4 x i32> %t7, <i32 1, i32 1, i32 1, i32 1>
  %t9 = mul nsw <4 x i32> %t8, %t4 ; signed
  %a10 = add nsw <4 x i32> %t9, %a1 ; signed
  ret <4 x i32> %a10
}

; ---------------------------------------------------------------------------- ;
; 64-bit width. 128 / 64 = 2 elts.
; ---------------------------------------------------------------------------- ;

; Values come from regs

define <2 x i64> @vec128_i64_signed_reg_reg(<2 x i64> %a1, <2 x i64> %a2) nounwind {
; SSE2-LABEL: vec128_i64_signed_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648]
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pxor %xmm4, %xmm5
; SSE2-NEXT:    pxor %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm4, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[0,0,2,2]
; SSE2-NEXT:    movdqa %xmm4, %xmm3
; SSE2-NEXT:    pcmpeqd %xmm5, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
; SSE2-NEXT:    pand %xmm7, %xmm6
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; SSE2-NEXT:    por %xmm6, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1]
; SSE2-NEXT:    por %xmm3, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm4, %xmm5
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[0,0,2,2]
; SSE2-NEXT:    pand %xmm7, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pand %xmm5, %xmm4
; SSE2-NEXT:    pandn %xmm1, %xmm5
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    pandn %xmm1, %xmm3
; SSE2-NEXT:    por %xmm4, %xmm3
; SSE2-NEXT:    psubq %xmm5, %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm1
; SSE2-NEXT:    psrlq $1, %xmm1
; SSE2-NEXT:    psrlq $33, %xmm3
; SSE2-NEXT:    pmuludq %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm2, %xmm4
; SSE2-NEXT:    psrlq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm1, %xmm4
; SSE2-NEXT:    paddq %xmm3, %xmm4
; SSE2-NEXT:    psllq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm2, %xmm1
; SSE2-NEXT:    paddq %xmm4, %xmm0
; SSE2-NEXT:    paddq %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i64_signed_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    pxor %xmm0, %xmm5
; SSE41-NEXT:    pxor %xmm2, %xmm0
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pcmpgtd %xmm5, %xmm3
; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
; SSE41-NEXT:    movdqa %xmm0, %xmm6
; SSE41-NEXT:    pcmpeqd %xmm5, %xmm6
; SSE41-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
; SSE41-NEXT:    pand %xmm4, %xmm7
; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
; SSE41-NEXT:    por %xmm7, %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1,1]
; SSE41-NEXT:    por %xmm3, %xmm4
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm5
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,2,2]
; SSE41-NEXT:    pand %xmm6, %xmm0
; SSE41-NEXT:    por %xmm5, %xmm0
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm5
; SSE41-NEXT:    movdqa %xmm3, %xmm0
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm1
; SSE41-NEXT:    psubq %xmm5, %xmm1
; SSE41-NEXT:    movdqa %xmm1, %xmm3
; SSE41-NEXT:    psrlq $1, %xmm3
; SSE41-NEXT:    psrlq $33, %xmm1
; SSE41-NEXT:    pmuludq %xmm4, %xmm1
; SSE41-NEXT:    movdqa %xmm4, %xmm0
; SSE41-NEXT:    psrlq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm3, %xmm0
; SSE41-NEXT:    paddq %xmm1, %xmm0
; SSE41-NEXT:    psllq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm4, %xmm3
; SSE41-NEXT:    paddq %xmm2, %xmm0
; SSE41-NEXT:    paddq %xmm3, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i64_signed_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX1-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX1-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i64_signed_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX2-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX2-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i64_signed_reg_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vpcomgtq %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; XOP-NEXT:    vpcomltq %xmm1, %xmm0, %xmm4
; XOP-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; XOP-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; XOP-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $1, %xmm1, %xmm2
; XOP-NEXT:    vpsrlq $33, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $32, %xmm3, %xmm4
; XOP-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; XOP-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; XOP-NEXT:    vpsllq $32, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; XOP-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; XOP-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i64_signed_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512F-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512F-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512F-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512F-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512F-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512F-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i64_signed_reg_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpcmpgtq %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsq %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsq %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubq %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i64_signed_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512BW-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %t3 = icmp sgt <2 x i64> %a1, %a2 ; signed
  %t4 = select <2 x i1> %t3, <2 x i64> <i64 -1, i64 -1>, <2 x i64> <i64 1, i64 1>
  %t5 = select <2 x i1> %t3, <2 x i64> %a2, <2 x i64> %a1
  %t6 = select <2 x i1> %t3, <2 x i64> %a1, <2 x i64> %a2
  %t7 = sub <2 x i64> %t6, %t5
  %t8 = lshr <2 x i64> %t7, <i64 1, i64 1>
  %t9 = mul nsw <2 x i64> %t8, %t4 ; signed
  %a10 = add nsw <2 x i64> %t9, %a1 ; signed
  ret <2 x i64> %a10
}

define <2 x i64> @vec128_i64_unsigned_reg_reg(<2 x i64> %a1, <2 x i64> %a2) nounwind {
; SSE2-LABEL: vec128_i64_unsigned_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [9223372039002259456,9223372039002259456]
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pxor %xmm4, %xmm5
; SSE2-NEXT:    pxor %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm4, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[0,0,2,2]
; SSE2-NEXT:    movdqa %xmm4, %xmm3
; SSE2-NEXT:    pcmpeqd %xmm5, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
; SSE2-NEXT:    pand %xmm7, %xmm6
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; SSE2-NEXT:    por %xmm6, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1]
; SSE2-NEXT:    por %xmm3, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm4, %xmm5
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[0,0,2,2]
; SSE2-NEXT:    pand %xmm7, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pand %xmm5, %xmm4
; SSE2-NEXT:    pandn %xmm1, %xmm5
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    pandn %xmm1, %xmm3
; SSE2-NEXT:    por %xmm4, %xmm3
; SSE2-NEXT:    psubq %xmm5, %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm1
; SSE2-NEXT:    psrlq $1, %xmm1
; SSE2-NEXT:    psrlq $33, %xmm3
; SSE2-NEXT:    pmuludq %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm2, %xmm4
; SSE2-NEXT:    psrlq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm1, %xmm4
; SSE2-NEXT:    paddq %xmm3, %xmm4
; SSE2-NEXT:    psllq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm2, %xmm1
; SSE2-NEXT:    paddq %xmm4, %xmm0
; SSE2-NEXT:    paddq %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i64_unsigned_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [9223372039002259456,9223372039002259456]
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    pxor %xmm0, %xmm5
; SSE41-NEXT:    pxor %xmm2, %xmm0
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pcmpgtd %xmm5, %xmm3
; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
; SSE41-NEXT:    movdqa %xmm0, %xmm6
; SSE41-NEXT:    pcmpeqd %xmm5, %xmm6
; SSE41-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
; SSE41-NEXT:    pand %xmm4, %xmm7
; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
; SSE41-NEXT:    por %xmm7, %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1,1]
; SSE41-NEXT:    por %xmm3, %xmm4
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm5
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,2,2]
; SSE41-NEXT:    pand %xmm6, %xmm0
; SSE41-NEXT:    por %xmm5, %xmm0
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm5
; SSE41-NEXT:    movdqa %xmm3, %xmm0
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm1
; SSE41-NEXT:    psubq %xmm5, %xmm1
; SSE41-NEXT:    movdqa %xmm1, %xmm3
; SSE41-NEXT:    psrlq $1, %xmm3
; SSE41-NEXT:    psrlq $33, %xmm1
; SSE41-NEXT:    pmuludq %xmm4, %xmm1
; SSE41-NEXT:    movdqa %xmm4, %xmm0
; SSE41-NEXT:    psrlq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm3, %xmm0
; SSE41-NEXT:    paddq %xmm1, %xmm0
; SSE41-NEXT:    psllq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm4, %xmm3
; SSE41-NEXT:    paddq %xmm2, %xmm0
; SSE41-NEXT:    paddq %xmm3, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i64_unsigned_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX1-FALLBACK-NEXT:    vpxor %xmm2, %xmm1, %xmm3
; AVX1-FALLBACK-NEXT:    vpxor %xmm2, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm5
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm5, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $32, %xmm5, %xmm3
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm3
; AVX1-FALLBACK-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
; AVX1-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm5, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i64_unsigned_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX2-FALLBACK-NEXT:    vpxor %xmm2, %xmm1, %xmm3
; AVX2-FALLBACK-NEXT:    vpxor %xmm2, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm3, %xmm2, %xmm4
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm5
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm5, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $32, %xmm5, %xmm3
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm3
; AVX2-FALLBACK-NEXT:    vpaddq %xmm1, %xmm3, %xmm1
; AVX2-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm5, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i64_unsigned_reg_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vpcomgtuq %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; XOP-NEXT:    vpcomltuq %xmm1, %xmm0, %xmm4
; XOP-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; XOP-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; XOP-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $1, %xmm1, %xmm2
; XOP-NEXT:    vpsrlq $33, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $32, %xmm3, %xmm4
; XOP-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; XOP-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; XOP-NEXT:    vpsllq $32, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; XOP-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; XOP-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i64_unsigned_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vpcmpnleuq %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminuq %zmm1, %zmm0, %zmm2
; AVX512F-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512F-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512F-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512F-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512F-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512F-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i64_unsigned_reg_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminuq %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxuq %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubq %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i64_unsigned_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpnleuq %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminuq %zmm1, %zmm0, %zmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm1
; AVX512BW-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %t3 = icmp ugt <2 x i64> %a1, %a2
  %t4 = select <2 x i1> %t3, <2 x i64> <i64 -1, i64 -1>, <2 x i64> <i64 1, i64 1>
  %t5 = select <2 x i1> %t3, <2 x i64> %a2, <2 x i64> %a1
  %t6 = select <2 x i1> %t3, <2 x i64> %a1, <2 x i64> %a2
  %t7 = sub <2 x i64> %t6, %t5
  %t8 = lshr <2 x i64> %t7, <i64 1, i64 1>
  %t9 = mul <2 x i64> %t8, %t4
  %a10 = add <2 x i64> %t9, %a1
  ret <2 x i64> %a10
}

; Values are loaded. Only check signed case.

define <2 x i64> @vec128_i64_signed_mem_reg(ptr %a1_addr, <2 x i64> %a2) nounwind {
; SSE2-LABEL: vec128_i64_signed_mem_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648]
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pxor %xmm4, %xmm5
; SSE2-NEXT:    pxor %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm4, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[0,0,2,2]
; SSE2-NEXT:    movdqa %xmm4, %xmm3
; SSE2-NEXT:    pcmpeqd %xmm5, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
; SSE2-NEXT:    pand %xmm7, %xmm6
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; SSE2-NEXT:    por %xmm6, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1]
; SSE2-NEXT:    por %xmm3, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm4, %xmm5
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[0,0,2,2]
; SSE2-NEXT:    pand %xmm7, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pand %xmm5, %xmm4
; SSE2-NEXT:    pandn %xmm0, %xmm5
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    pandn %xmm0, %xmm3
; SSE2-NEXT:    por %xmm4, %xmm3
; SSE2-NEXT:    psubq %xmm5, %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm4
; SSE2-NEXT:    psrlq $1, %xmm4
; SSE2-NEXT:    psrlq $33, %xmm3
; SSE2-NEXT:    pmuludq %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm2, %xmm0
; SSE2-NEXT:    psrlq $32, %xmm0
; SSE2-NEXT:    pmuludq %xmm4, %xmm0
; SSE2-NEXT:    paddq %xmm3, %xmm0
; SSE2-NEXT:    psllq $32, %xmm0
; SSE2-NEXT:    pmuludq %xmm2, %xmm4
; SSE2-NEXT:    paddq %xmm1, %xmm0
; SSE2-NEXT:    paddq %xmm4, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i64_signed_mem_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm1
; SSE41-NEXT:    movdqa (%rdi), %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    pxor %xmm0, %xmm5
; SSE41-NEXT:    pxor %xmm3, %xmm0
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
; SSE41-NEXT:    movdqa %xmm0, %xmm6
; SSE41-NEXT:    pcmpeqd %xmm5, %xmm6
; SSE41-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
; SSE41-NEXT:    pand %xmm4, %xmm7
; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; SSE41-NEXT:    por %xmm7, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1,1]
; SSE41-NEXT:    por %xmm2, %xmm4
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm5
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,2,2]
; SSE41-NEXT:    pand %xmm6, %xmm0
; SSE41-NEXT:    por %xmm5, %xmm0
; SSE41-NEXT:    movdqa %xmm1, %xmm5
; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm5
; SSE41-NEXT:    movdqa %xmm2, %xmm0
; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm1
; SSE41-NEXT:    psubq %xmm5, %xmm1
; SSE41-NEXT:    movdqa %xmm1, %xmm2
; SSE41-NEXT:    psrlq $1, %xmm2
; SSE41-NEXT:    psrlq $33, %xmm1
; SSE41-NEXT:    pmuludq %xmm4, %xmm1
; SSE41-NEXT:    movdqa %xmm4, %xmm0
; SSE41-NEXT:    psrlq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm2, %xmm0
; SSE41-NEXT:    paddq %xmm1, %xmm0
; SSE41-NEXT:    psllq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm4, %xmm2
; SSE41-NEXT:    paddq %xmm3, %xmm0
; SSE41-NEXT:    paddq %xmm2, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i64_signed_mem_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm1, %xmm0, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsubq %xmm4, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsrlq $1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpsrlq $33, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX1-FALLBACK-NEXT:    vpsllq $32, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i64_signed_mem_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm1, %xmm0, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsubq %xmm4, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsrlq $1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpsrlq $33, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX2-FALLBACK-NEXT:    vpsllq $32, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i64_signed_mem_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-NEXT:    vpcomgtq %xmm0, %xmm1, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; XOP-NEXT:    vpcomltq %xmm0, %xmm1, %xmm4
; XOP-NEXT:    vblendvpd %xmm4, %xmm1, %xmm0, %xmm4
; XOP-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
; XOP-NEXT:    vpsubq %xmm4, %xmm0, %xmm0
; XOP-NEXT:    vpsrlq $1, %xmm0, %xmm2
; XOP-NEXT:    vpsrlq $33, %xmm0, %xmm0
; XOP-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; XOP-NEXT:    vpsrlq $32, %xmm3, %xmm4
; XOP-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; XOP-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; XOP-NEXT:    vpsllq $32, %xmm0, %xmm0
; XOP-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; XOP-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; XOP-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i64_signed_mem_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtq %zmm0, %zmm1, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsq %zmm0, %zmm1, %zmm2
; AVX512F-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
; AVX512F-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
; AVX512F-NEXT:    vpsrlq $1, %xmm0, %xmm2
; AVX512F-NEXT:    vpsrlq $33, %xmm0, %xmm0
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; AVX512F-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512F-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512F-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX512F-NEXT:    vpsllq $32, %xmm0, %xmm0
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512F-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i64_signed_mem_reg:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-NEXT:    vpcmpgtq %xmm0, %xmm1, %k1
; AVX512VL-NEXT:    vpminsq %xmm0, %xmm1, %xmm2
; AVX512VL-NEXT:    vpmaxsq %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
; AVX512VL-NEXT:    vpsrlq $1, %xmm0, %xmm0
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubq %xmm0, %xmm2, %xmm0 {%k1}
; AVX512VL-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i64_signed_mem_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtq %zmm0, %zmm1, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsq %zmm0, %zmm1, %zmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
; AVX512BW-FALLBACK-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsrlq $1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpsrlq $33, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm4, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsllq $32, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a1 = load <2 x i64>, ptr %a1_addr
  %t3 = icmp sgt <2 x i64> %a1, %a2 ; signed
  %t4 = select <2 x i1> %t3, <2 x i64> <i64 -1, i64 -1>, <2 x i64> <i64 1, i64 1>
  %t5 = select <2 x i1> %t3, <2 x i64> %a2, <2 x i64> %a1
  %t6 = select <2 x i1> %t3, <2 x i64> %a1, <2 x i64> %a2
  %t7 = sub <2 x i64> %t6, %t5
  %t8 = lshr <2 x i64> %t7, <i64 1, i64 1>
  %t9 = mul nsw <2 x i64> %t8, %t4 ; signed
  %a10 = add nsw <2 x i64> %t9, %a1 ; signed
  ret <2 x i64> %a10
}

define <2 x i64> @vec128_i64_signed_reg_mem(<2 x i64> %a1, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i64_signed_reg_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648]
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pxor %xmm4, %xmm5
; SSE2-NEXT:    pxor %xmm3, %xmm4
; SSE2-NEXT:    movdqa %xmm5, %xmm1
; SSE2-NEXT:    pcmpgtd %xmm4, %xmm1
; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[0,0,2,2]
; SSE2-NEXT:    movdqa %xmm5, %xmm2
; SSE2-NEXT:    pcmpeqd %xmm4, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[1,1,3,3]
; SSE2-NEXT:    pand %xmm7, %xmm6
; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; SSE2-NEXT:    por %xmm6, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1,1]
; SSE2-NEXT:    por %xmm2, %xmm1
; SSE2-NEXT:    pcmpgtd %xmm5, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
; SSE2-NEXT:    pand %xmm7, %xmm5
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubq %xmm4, %xmm2
; SSE2-NEXT:    movdqa %xmm2, %xmm3
; SSE2-NEXT:    psrlq $1, %xmm3
; SSE2-NEXT:    psrlq $33, %xmm2
; SSE2-NEXT:    pmuludq %xmm1, %xmm2
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    psrlq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm3, %xmm4
; SSE2-NEXT:    paddq %xmm2, %xmm4
; SSE2-NEXT:    psllq $32, %xmm4
; SSE2-NEXT:    pmuludq %xmm1, %xmm3
; SSE2-NEXT:    paddq %xmm4, %xmm0
; SSE2-NEXT:    paddq %xmm3, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i64_signed_reg_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm1
; SSE41-NEXT:    movdqa (%rdi), %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [2147483648,2147483648]
; SSE41-NEXT:    pxor %xmm5, %xmm0
; SSE41-NEXT:    pxor %xmm3, %xmm5
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,0,2,2]
; SSE41-NEXT:    movdqa %xmm0, %xmm6
; SSE41-NEXT:    pcmpeqd %xmm5, %xmm6
; SSE41-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
; SSE41-NEXT:    pand %xmm4, %xmm7
; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; SSE41-NEXT:    por %xmm7, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1,1]
; SSE41-NEXT:    por %xmm2, %xmm4
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm5
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,2,2]
; SSE41-NEXT:    pand %xmm6, %xmm0
; SSE41-NEXT:    por %xmm5, %xmm0
; SSE41-NEXT:    movdqa %xmm3, %xmm5
; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm5
; SSE41-NEXT:    movdqa %xmm2, %xmm0
; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm3
; SSE41-NEXT:    psubq %xmm5, %xmm3
; SSE41-NEXT:    movdqa %xmm3, %xmm2
; SSE41-NEXT:    psrlq $1, %xmm2
; SSE41-NEXT:    psrlq $33, %xmm3
; SSE41-NEXT:    pmuludq %xmm4, %xmm3
; SSE41-NEXT:    movdqa %xmm4, %xmm0
; SSE41-NEXT:    psrlq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm2, %xmm0
; SSE41-NEXT:    paddq %xmm3, %xmm0
; SSE41-NEXT:    psllq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm4, %xmm2
; SSE41-NEXT:    paddq %xmm1, %xmm0
; SSE41-NEXT:    paddq %xmm2, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i64_signed_reg_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX1-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX1-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i64_signed_reg_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX2-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX2-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i64_signed_reg_mem:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-NEXT:    vpcomgtq %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; XOP-NEXT:    vpcomltq %xmm1, %xmm0, %xmm4
; XOP-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; XOP-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; XOP-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $1, %xmm1, %xmm2
; XOP-NEXT:    vpsrlq $33, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $32, %xmm3, %xmm4
; XOP-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; XOP-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; XOP-NEXT:    vpsllq $32, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; XOP-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; XOP-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i64_signed_reg_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512F-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512F-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512F-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512F-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512F-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512F-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i64_signed_reg_mem:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-NEXT:    vpcmpgtq %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsq %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsq %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubq %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i64_signed_reg_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512BW-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a2 = load <2 x i64>, ptr %a2_addr
  %t3 = icmp sgt <2 x i64> %a1, %a2 ; signed
  %t4 = select <2 x i1> %t3, <2 x i64> <i64 -1, i64 -1>, <2 x i64> <i64 1, i64 1>
  %t5 = select <2 x i1> %t3, <2 x i64> %a2, <2 x i64> %a1
  %t6 = select <2 x i1> %t3, <2 x i64> %a1, <2 x i64> %a2
  %t7 = sub <2 x i64> %t6, %t5
  %t8 = lshr <2 x i64> %t7, <i64 1, i64 1>
  %t9 = mul nsw <2 x i64> %t8, %t4 ; signed
  %a10 = add nsw <2 x i64> %t9, %a1 ; signed
  ret <2 x i64> %a10
}

define <2 x i64> @vec128_i64_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i64_signed_mem_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa (%rsi), %xmm0
; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648]
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pxor %xmm4, %xmm5
; SSE2-NEXT:    pxor %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm4, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm5, %xmm2
; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[0,0,2,2]
; SSE2-NEXT:    movdqa %xmm4, %xmm3
; SSE2-NEXT:    pcmpeqd %xmm5, %xmm3
; SSE2-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[1,1,3,3]
; SSE2-NEXT:    pand %xmm7, %xmm6
; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; SSE2-NEXT:    por %xmm6, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1]
; SSE2-NEXT:    por %xmm3, %xmm2
; SSE2-NEXT:    pcmpgtd %xmm4, %xmm5
; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[0,0,2,2]
; SSE2-NEXT:    pand %xmm7, %xmm4
; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pand %xmm5, %xmm4
; SSE2-NEXT:    pandn %xmm0, %xmm5
; SSE2-NEXT:    por %xmm4, %xmm5
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    pandn %xmm0, %xmm3
; SSE2-NEXT:    por %xmm4, %xmm3
; SSE2-NEXT:    psubq %xmm5, %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm4
; SSE2-NEXT:    psrlq $1, %xmm4
; SSE2-NEXT:    psrlq $33, %xmm3
; SSE2-NEXT:    pmuludq %xmm2, %xmm3
; SSE2-NEXT:    movdqa %xmm2, %xmm0
; SSE2-NEXT:    psrlq $32, %xmm0
; SSE2-NEXT:    pmuludq %xmm4, %xmm0
; SSE2-NEXT:    paddq %xmm3, %xmm0
; SSE2-NEXT:    psllq $32, %xmm0
; SSE2-NEXT:    pmuludq %xmm2, %xmm4
; SSE2-NEXT:    paddq %xmm1, %xmm0
; SSE2-NEXT:    paddq %xmm4, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i64_signed_mem_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm2
; SSE41-NEXT:    movdqa (%rsi), %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
; SSE41-NEXT:    movdqa %xmm3, %xmm5
; SSE41-NEXT:    pxor %xmm0, %xmm5
; SSE41-NEXT:    pxor %xmm2, %xmm0
; SSE41-NEXT:    movdqa %xmm0, %xmm1
; SSE41-NEXT:    pcmpgtd %xmm5, %xmm1
; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[0,0,2,2]
; SSE41-NEXT:    movdqa %xmm0, %xmm6
; SSE41-NEXT:    pcmpeqd %xmm5, %xmm6
; SSE41-NEXT:    pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
; SSE41-NEXT:    pand %xmm4, %xmm7
; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; SSE41-NEXT:    por %xmm7, %xmm1
; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1,1]
; SSE41-NEXT:    por %xmm1, %xmm4
; SSE41-NEXT:    pcmpgtd %xmm0, %xmm5
; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,2,2]
; SSE41-NEXT:    pand %xmm6, %xmm0
; SSE41-NEXT:    por %xmm5, %xmm0
; SSE41-NEXT:    movdqa %xmm3, %xmm5
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm5
; SSE41-NEXT:    movdqa %xmm1, %xmm0
; SSE41-NEXT:    blendvpd %xmm0, %xmm2, %xmm3
; SSE41-NEXT:    psubq %xmm5, %xmm3
; SSE41-NEXT:    movdqa %xmm3, %xmm1
; SSE41-NEXT:    psrlq $1, %xmm1
; SSE41-NEXT:    psrlq $33, %xmm3
; SSE41-NEXT:    pmuludq %xmm4, %xmm3
; SSE41-NEXT:    movdqa %xmm4, %xmm0
; SSE41-NEXT:    psrlq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm1, %xmm0
; SSE41-NEXT:    paddq %xmm3, %xmm0
; SSE41-NEXT:    psllq $32, %xmm0
; SSE41-NEXT:    pmuludq %xmm4, %xmm1
; SSE41-NEXT:    paddq %xmm2, %xmm0
; SSE41-NEXT:    paddq %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i64_signed_mem_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX1-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX1-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX1-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX1-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i64_signed_mem_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; AVX2-FALLBACK-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX2-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX2-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i64_signed_mem_mem:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm0
; XOP-NEXT:    vmovdqa (%rsi), %xmm1
; XOP-NEXT:    vpcomgtq %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm3
; XOP-NEXT:    vpcomltq %xmm1, %xmm0, %xmm4
; XOP-NEXT:    vblendvpd %xmm4, %xmm0, %xmm1, %xmm4
; XOP-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm1
; XOP-NEXT:    vpsubq %xmm4, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $1, %xmm1, %xmm2
; XOP-NEXT:    vpsrlq $33, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlq $32, %xmm3, %xmm4
; XOP-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; XOP-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; XOP-NEXT:    vpsllq $32, %xmm1, %xmm1
; XOP-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; XOP-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; XOP-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i64_signed_mem_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512F-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512F-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512F-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512F-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512F-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512F-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512F-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512F-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512F-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512F-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512F-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512F-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-LABEL: vec128_i64_signed_mem_mem:
; AVX512VL:       # %bb.0:
; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VL-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VL-NEXT:    vpcmpgtq %xmm1, %xmm0, %k1
; AVX512VL-NEXT:    vpminsq %xmm1, %xmm0, %xmm2
; AVX512VL-NEXT:    vpmaxsq %xmm1, %xmm0, %xmm1
; AVX512VL-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VL-NEXT:    vpsubq %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VL-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512VL-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i64_signed_mem_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqa64 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsq %zmm1, %zmm0, %zmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm1
; AVX512BW-FALLBACK-NEXT:    vpsubq %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $1, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpsrlq $33, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlq $32, %xmm3, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm4, %xmm2, %xmm4
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm1, %xmm4, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsllq $32, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
  %a1 = load <2 x i64>, ptr %a1_addr
  %a2 = load <2 x i64>, ptr %a2_addr
  %t3 = icmp sgt <2 x i64> %a1, %a2 ; signed
  %t4 = select <2 x i1> %t3, <2 x i64> <i64 -1, i64 -1>, <2 x i64> <i64 1, i64 1>
  %t5 = select <2 x i1> %t3, <2 x i64> %a2, <2 x i64> %a1
  %t6 = select <2 x i1> %t3, <2 x i64> %a1, <2 x i64> %a2
  %t7 = sub <2 x i64> %t6, %t5
  %t8 = lshr <2 x i64> %t7, <i64 1, i64 1>
  %t9 = mul nsw <2 x i64> %t8, %t4 ; signed
  %a10 = add nsw <2 x i64> %t9, %a1 ; signed
  ret <2 x i64> %a10
}

; ---------------------------------------------------------------------------- ;
; 16-bit width. 128 / 16 = 8 elts.
; ---------------------------------------------------------------------------- ;

; Values come from regs

define <8 x i16> @vec128_i16_signed_reg_reg(<8 x i16> %a1, <8 x i16> %a2) nounwind {
; SSE-LABEL: vec128_i16_signed_reg_reg:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa %xmm0, %xmm2
; SSE-NEXT:    pcmpgtw %xmm1, %xmm2
; SSE-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE-NEXT:    movdqa %xmm0, %xmm3
; SSE-NEXT:    pminsw %xmm1, %xmm3
; SSE-NEXT:    pmaxsw %xmm0, %xmm1
; SSE-NEXT:    psubw %xmm3, %xmm1
; SSE-NEXT:    psrlw $1, %xmm1
; SSE-NEXT:    pmullw %xmm1, %xmm2
; SSE-NEXT:    paddw %xmm2, %xmm0
; SSE-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i16_signed_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i16_signed_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i16_signed_reg_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vpcomgtw %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; XOP-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; XOP-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
; XOP-NEXT:    vpmacsww %xmm0, %xmm2, %xmm1, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i16_signed_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i16_signed_reg_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i16_signed_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpgtw %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu16 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i16_signed_reg_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %t3 = icmp sgt <8 x i16> %a1, %a2 ; signed
  %t4 = select <8 x i1> %t3, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t5 = select <8 x i1> %t3, <8 x i16> %a2, <8 x i16> %a1
  %t6 = select <8 x i1> %t3, <8 x i16> %a1, <8 x i16> %a2
  %t7 = sub <8 x i16> %t6, %t5
  %t8 = lshr <8 x i16> %t7, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t9 = mul nsw <8 x i16> %t8, %t4 ; signed
  %a10 = add nsw <8 x i16> %t9, %a1 ; signed
  ret <8 x i16> %a10
}

define <8 x i16> @vec128_i16_unsigned_reg_reg(<8 x i16> %a1, <8 x i16> %a2) nounwind {
; SSE2-LABEL: vec128_i16_unsigned_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
; SSE2-NEXT:    movdqa %xmm1, %xmm3
; SSE2-NEXT:    pxor %xmm2, %xmm3
; SSE2-NEXT:    pxor %xmm0, %xmm2
; SSE2-NEXT:    pcmpgtw %xmm3, %xmm2
; SSE2-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE2-NEXT:    movdqa %xmm0, %xmm3
; SSE2-NEXT:    psubusw %xmm1, %xmm3
; SSE2-NEXT:    psubusw %xmm0, %xmm1
; SSE2-NEXT:    psubw %xmm0, %xmm3
; SSE2-NEXT:    paddw %xmm1, %xmm3
; SSE2-NEXT:    paddw %xmm0, %xmm3
; SSE2-NEXT:    psrlw $1, %xmm3
; SSE2-NEXT:    pmullw %xmm2, %xmm3
; SSE2-NEXT:    paddw %xmm3, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i16_unsigned_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pminuw %xmm1, %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pcmpeqw %xmm2, %xmm3
; SSE41-NEXT:    pcmpeqd %xmm4, %xmm4
; SSE41-NEXT:    pxor %xmm3, %xmm4
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
; SSE41-NEXT:    pmaxuw %xmm0, %xmm1
; SSE41-NEXT:    psubw %xmm2, %xmm1
; SSE41-NEXT:    psrlw $1, %xmm1
; SSE41-NEXT:    pmullw %xmm1, %xmm4
; SSE41-NEXT:    paddw %xmm4, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i16_unsigned_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX1-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i16_unsigned_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX2-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i16_unsigned_reg_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vpcomgtuw %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-NEXT:    vpminuw %xmm1, %xmm0, %xmm3
; XOP-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; XOP-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
; XOP-NEXT:    vpmacsww %xmm0, %xmm2, %xmm1, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i16_unsigned_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm3
; AVX512F-NEXT:    vpternlogq $15, %zmm3, %zmm3, %zmm3
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX512F-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i16_unsigned_reg_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm2, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i16_unsigned_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpnleuw %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu16 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i16_unsigned_reg_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vpcmpnleuw %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminuw %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %t3 = icmp ugt <8 x i16> %a1, %a2
  %t4 = select <8 x i1> %t3, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t5 = select <8 x i1> %t3, <8 x i16> %a2, <8 x i16> %a1
  %t6 = select <8 x i1> %t3, <8 x i16> %a1, <8 x i16> %a2
  %t7 = sub <8 x i16> %t6, %t5
  %t8 = lshr <8 x i16> %t7, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t9 = mul <8 x i16> %t8, %t4
  %a10 = add <8 x i16> %t9, %a1
  ret <8 x i16> %a10
}

; Values are loaded. Only check signed case.

define <8 x i16> @vec128_i16_signed_mem_reg(ptr %a1_addr, <8 x i16> %a2) nounwind {
; SSE-LABEL: vec128_i16_signed_mem_reg:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa (%rdi), %xmm1
; SSE-NEXT:    movdqa %xmm1, %xmm2
; SSE-NEXT:    pcmpgtw %xmm0, %xmm2
; SSE-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE-NEXT:    movdqa %xmm1, %xmm3
; SSE-NEXT:    pminsw %xmm0, %xmm3
; SSE-NEXT:    pmaxsw %xmm1, %xmm0
; SSE-NEXT:    psubw %xmm3, %xmm0
; SSE-NEXT:    psrlw $1, %xmm0
; SSE-NEXT:    pmullw %xmm2, %xmm0
; SSE-NEXT:    paddw %xmm1, %xmm0
; SSE-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i16_signed_mem_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsw %xmm0, %xmm1, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpsubw %xmm3, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i16_signed_mem_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsw %xmm0, %xmm1, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpsubw %xmm3, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i16_signed_mem_reg:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-NEXT:    vpcomgtw %xmm0, %xmm1, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-NEXT:    vpminsw %xmm0, %xmm1, %xmm3
; XOP-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; XOP-NEXT:    vpsubw %xmm3, %xmm0, %xmm0
; XOP-NEXT:    vpsrlw $1, %xmm0, %xmm0
; XOP-NEXT:    vpmacsww %xmm1, %xmm2, %xmm0, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i16_signed_mem_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsw %xmm0, %xmm1, %xmm3
; AVX512F-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpsubw %xmm3, %xmm0, %xmm0
; AVX512F-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512F-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; AVX512F-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i16_signed_mem_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsw %xmm0, %xmm1, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm0, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm0, %xmm2, %xmm0
; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i16_signed_mem_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtw %zmm0, %zmm1, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu16 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsw %xmm0, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsubw %xmm2, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpmullw %xmm3, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i16_signed_mem_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtw %xmm0, %xmm1, %k1
; AVX512VLBW-NEXT:    vpminsw %xmm0, %xmm1, %xmm2
; AVX512VLBW-NEXT:    vpmaxsw %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm0, %xmm0
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubw %xmm0, %xmm2, %xmm0 {%k1}
; AVX512VLBW-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; AVX512VLBW-NEXT:    retq
  %a1 = load <8 x i16>, ptr %a1_addr
  %t3 = icmp sgt <8 x i16> %a1, %a2 ; signed
  %t4 = select <8 x i1> %t3, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t5 = select <8 x i1> %t3, <8 x i16> %a2, <8 x i16> %a1
  %t6 = select <8 x i1> %t3, <8 x i16> %a1, <8 x i16> %a2
  %t7 = sub <8 x i16> %t6, %t5
  %t8 = lshr <8 x i16> %t7, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t9 = mul nsw <8 x i16> %t8, %t4 ; signed
  %a10 = add nsw <8 x i16> %t9, %a1 ; signed
  ret <8 x i16> %a10
}

define <8 x i16> @vec128_i16_signed_reg_mem(<8 x i16> %a1, ptr %a2_addr) nounwind {
; SSE-LABEL: vec128_i16_signed_reg_mem:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa (%rdi), %xmm1
; SSE-NEXT:    movdqa %xmm0, %xmm2
; SSE-NEXT:    pcmpgtw %xmm1, %xmm2
; SSE-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE-NEXT:    movdqa %xmm0, %xmm3
; SSE-NEXT:    pminsw %xmm1, %xmm3
; SSE-NEXT:    pmaxsw %xmm0, %xmm1
; SSE-NEXT:    psubw %xmm3, %xmm1
; SSE-NEXT:    psrlw $1, %xmm1
; SSE-NEXT:    pmullw %xmm2, %xmm1
; SSE-NEXT:    paddw %xmm1, %xmm0
; SSE-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i16_signed_reg_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i16_signed_reg_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i16_signed_reg_mem:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-NEXT:    vpcomgtw %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; XOP-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; XOP-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
; XOP-NEXT:    vpmacsww %xmm0, %xmm2, %xmm1, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i16_signed_reg_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i16_signed_reg_mem:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i16_signed_reg_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtw %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu16 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i16_signed_reg_mem:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %a2 = load <8 x i16>, ptr %a2_addr
  %t3 = icmp sgt <8 x i16> %a1, %a2 ; signed
  %t4 = select <8 x i1> %t3, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t5 = select <8 x i1> %t3, <8 x i16> %a2, <8 x i16> %a1
  %t6 = select <8 x i1> %t3, <8 x i16> %a1, <8 x i16> %a2
  %t7 = sub <8 x i16> %t6, %t5
  %t8 = lshr <8 x i16> %t7, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t9 = mul nsw <8 x i16> %t8, %t4 ; signed
  %a10 = add nsw <8 x i16> %t9, %a1 ; signed
  ret <8 x i16> %a10
}

define <8 x i16> @vec128_i16_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; SSE-LABEL: vec128_i16_signed_mem_mem:
; SSE:       # %bb.0:
; SSE-NEXT:    movdqa (%rdi), %xmm1
; SSE-NEXT:    movdqa (%rsi), %xmm0
; SSE-NEXT:    movdqa %xmm1, %xmm2
; SSE-NEXT:    pcmpgtw %xmm0, %xmm2
; SSE-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE-NEXT:    movdqa %xmm1, %xmm3
; SSE-NEXT:    pminsw %xmm0, %xmm3
; SSE-NEXT:    pmaxsw %xmm1, %xmm0
; SSE-NEXT:    psubw %xmm3, %xmm0
; SSE-NEXT:    psrlw $1, %xmm0
; SSE-NEXT:    pmullw %xmm2, %xmm0
; SSE-NEXT:    paddw %xmm1, %xmm0
; SSE-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i16_signed_mem_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX1-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i16_signed_mem_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-LABEL: vec128_i16_signed_mem_mem:
; XOP:       # %bb.0:
; XOP-NEXT:    vmovdqa (%rdi), %xmm0
; XOP-NEXT:    vmovdqa (%rsi), %xmm1
; XOP-NEXT:    vpcomgtw %xmm1, %xmm0, %xmm2
; XOP-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; XOP-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; XOP-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
; XOP-NEXT:    vpmacsww %xmm0, %xmm2, %xmm1, %xmm0
; XOP-NEXT:    retq
;
; AVX512F-LABEL: vec128_i16_signed_mem_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512F-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsw %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubw %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i16_signed_mem_mem:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i16_signed_mem_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtw %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu16 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i16_signed_mem_mem:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VLBW-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsw %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %a1 = load <8 x i16>, ptr %a1_addr
  %a2 = load <8 x i16>, ptr %a2_addr
  %t3 = icmp sgt <8 x i16> %a1, %a2 ; signed
  %t4 = select <8 x i1> %t3, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t5 = select <8 x i1> %t3, <8 x i16> %a2, <8 x i16> %a1
  %t6 = select <8 x i1> %t3, <8 x i16> %a1, <8 x i16> %a2
  %t7 = sub <8 x i16> %t6, %t5
  %t8 = lshr <8 x i16> %t7, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
  %t9 = mul nsw <8 x i16> %t8, %t4 ; signed
  %a10 = add nsw <8 x i16> %t9, %a1 ; signed
  ret <8 x i16> %a10
}

; ---------------------------------------------------------------------------- ;
; 8-bit width. 128 / 8 = 16 elts.
; ---------------------------------------------------------------------------- ;

; Values come from regs

define <16 x i8> @vec128_i8_signed_reg_reg(<16 x i8> %a1, <16 x i8> %a2) nounwind {
; SSE2-LABEL: vec128_i8_signed_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa %xmm0, %xmm3
; SSE2-NEXT:    pcmpgtb %xmm1, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE2-NEXT:    por %xmm3, %xmm2
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pcmpgtb %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm3, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm3
; SSE2-NEXT:    por %xmm5, %xmm3
; SSE2-NEXT:    psubb %xmm4, %xmm3
; SSE2-NEXT:    psrlw $1, %xmm3
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm1
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    movdqa %xmm2, %xmm4
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    pmullw %xmm1, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
; SSE2-NEXT:    pand %xmm1, %xmm4
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    pmullw %xmm3, %xmm2
; SSE2-NEXT:    pand %xmm1, %xmm2
; SSE2-NEXT:    packuswb %xmm4, %xmm2
; SSE2-NEXT:    paddb %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i8_signed_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtb %xmm1, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pminsb %xmm1, %xmm3
; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
; SSE41-NEXT:    psubb %xmm3, %xmm1
; SSE41-NEXT:    psrlw $1, %xmm1
; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmullw %xmm1, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
; SSE41-NEXT:    pand %xmm1, %xmm2
; SSE41-NEXT:    pmullw %xmm4, %xmm3
; SSE41-NEXT:    pand %xmm1, %xmm3
; SSE41-NEXT:    packuswb %xmm2, %xmm3
; SSE41-NEXT:    paddb %xmm3, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i8_signed_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i8_signed_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX2-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vzeroupper
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i8_signed_reg_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOP-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOP-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i8_signed_reg_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOPAVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i8_signed_reg_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; XOPAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; XOPAVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vzeroupper
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i8_signed_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512F-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
; AVX512F-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i8_signed_reg_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i8_signed_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpgtb %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu8 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpmovwb %zmm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i8_signed_reg_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %t3 = icmp sgt <16 x i8> %a1, %a2 ; signed
  %t4 = select <16 x i1> %t3, <16 x i8> <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>, <16 x i8> <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>
  %t5 = select <16 x i1> %t3, <16 x i8> %a2, <16 x i8> %a1
  %t6 = select <16 x i1> %t3, <16 x i8> %a1, <16 x i8> %a2
  %t7 = sub <16 x i8> %t6, %t5
  %t8 = lshr <16 x i8> %t7, <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>
  %t9 = mul nsw <16 x i8> %t8, %t4 ; signed
  %a10 = add nsw <16 x i8> %t9, %a1 ; signed
  ret <16 x i8> %a10
}

define <16 x i8> @vec128_i8_unsigned_reg_reg(<16 x i8> %a1, <16 x i8> %a2) nounwind {
; SSE2-LABEL: vec128_i8_unsigned_reg_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa %xmm0, %xmm3
; SSE2-NEXT:    pminub %xmm1, %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    pcmpeqb %xmm3, %xmm4
; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
; SSE2-NEXT:    pxor %xmm4, %xmm2
; SSE2-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE2-NEXT:    pmaxub %xmm0, %xmm1
; SSE2-NEXT:    psubb %xmm3, %xmm1
; SSE2-NEXT:    psrlw $1, %xmm1
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE2-NEXT:    movdqa %xmm1, %xmm3
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    movdqa %xmm2, %xmm4
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    pmullw %xmm3, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    pmullw %xmm1, %xmm2
; SSE2-NEXT:    pand %xmm3, %xmm2
; SSE2-NEXT:    packuswb %xmm4, %xmm2
; SSE2-NEXT:    paddb %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i8_unsigned_reg_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pminub %xmm1, %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pcmpeqb %xmm2, %xmm3
; SSE41-NEXT:    pcmpeqd %xmm4, %xmm4
; SSE41-NEXT:    pxor %xmm3, %xmm4
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
; SSE41-NEXT:    pmaxub %xmm0, %xmm1
; SSE41-NEXT:    psubb %xmm2, %xmm1
; SSE41-NEXT:    psrlw $1, %xmm1
; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmullw %xmm1, %xmm4
; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
; SSE41-NEXT:    pand %xmm1, %xmm4
; SSE41-NEXT:    pmullw %xmm3, %xmm2
; SSE41-NEXT:    pand %xmm1, %xmm2
; SSE41-NEXT:    packuswb %xmm4, %xmm2
; SSE41-NEXT:    paddb %xmm2, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX1-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpmullw %xmm4, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
; AVX1-FALLBACK-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
; AVX2-FALLBACK-NEXT:    vpxor %xmm4, %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX2-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vzeroupper
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vpcomgtub %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOP-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOP-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i8_unsigned_reg_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vpcomgtub %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOPAVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i8_unsigned_reg_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vpcomgtub %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; XOPAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; XOPAVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vzeroupper
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i8_unsigned_reg_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm3
; AVX512F-NEXT:    vpternlogq $15, %zmm3, %zmm3, %zmm3
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; AVX512F-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512F-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
; AVX512F-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpternlogq $15, %xmm2, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i8_unsigned_reg_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vpcmpnleub %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu8 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpmovwb %zmm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i8_unsigned_reg_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vpcmpnleub %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminub %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %t3 = icmp ugt <16 x i8> %a1, %a2
  %t4 = select <16 x i1> %t3, <16 x i8> <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>, <16 x i8> <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>
  %t5 = select <16 x i1> %t3, <16 x i8> %a2, <16 x i8> %a1
  %t6 = select <16 x i1> %t3, <16 x i8> %a1, <16 x i8> %a2
  %t7 = sub <16 x i8> %t6, %t5
  %t8 = lshr <16 x i8> %t7, <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>
  %t9 = mul <16 x i8> %t8, %t4
  %a10 = add <16 x i8> %t9, %a1
  ret <16 x i8> %a10
}

; Values are loaded. Only check signed case.

define <16 x i8> @vec128_i8_signed_mem_reg(ptr %a1_addr, <16 x i8> %a2) nounwind {
; SSE2-LABEL: vec128_i8_signed_mem_reg:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa %xmm0, %xmm1
; SSE2-NEXT:    movdqa (%rdi), %xmm2
; SSE2-NEXT:    movdqa %xmm2, %xmm3
; SSE2-NEXT:    pcmpgtb %xmm0, %xmm3
; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE2-NEXT:    por %xmm3, %xmm0
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    pcmpgtb %xmm2, %xmm4
; SSE2-NEXT:    movdqa %xmm2, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm2, %xmm5
; SSE2-NEXT:    pand %xmm3, %xmm5
; SSE2-NEXT:    pandn %xmm1, %xmm3
; SSE2-NEXT:    por %xmm5, %xmm3
; SSE2-NEXT:    psubb %xmm4, %xmm3
; SSE2-NEXT:    psrlw $1, %xmm3
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; SSE2-NEXT:    movdqa %xmm3, %xmm1
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    pmullw %xmm1, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
; SSE2-NEXT:    pand %xmm1, %xmm4
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    pmullw %xmm3, %xmm0
; SSE2-NEXT:    pand %xmm1, %xmm0
; SSE2-NEXT:    packuswb %xmm4, %xmm0
; SSE2-NEXT:    paddb %xmm2, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i8_signed_mem_reg:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm2
; SSE41-NEXT:    movdqa %xmm2, %xmm3
; SSE41-NEXT:    pcmpgtb %xmm0, %xmm3
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; SSE41-NEXT:    movdqa %xmm2, %xmm1
; SSE41-NEXT:    pminsb %xmm0, %xmm1
; SSE41-NEXT:    pmaxsb %xmm2, %xmm0
; SSE41-NEXT:    psubb %xmm1, %xmm0
; SSE41-NEXT:    psrlw $1, %xmm0
; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmullw %xmm0, %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,255,255,255,255]
; SSE41-NEXT:    pand %xmm0, %xmm3
; SSE41-NEXT:    pmullw %xmm4, %xmm1
; SSE41-NEXT:    pand %xmm0, %xmm1
; SSE41-NEXT:    packuswb %xmm3, %xmm1
; SSE41-NEXT:    paddb %xmm2, %xmm1
; SSE41-NEXT:    movdqa %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i8_signed_mem_reg:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i8_signed_mem_reg:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX2-FALLBACK-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX2-FALLBACK-NEXT:    vextracti128 $1, %ymm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX2-FALLBACK-NEXT:    vzeroupper
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i8_signed_mem_reg:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtb %xmm0, %xmm1, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOP-FALLBACK-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOP-FALLBACK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i8_signed_mem_reg:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX1-NEXT:    vpcomgtb %xmm0, %xmm1, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; XOPAVX1-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOPAVX1-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOPAVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i8_signed_mem_reg:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX2-NEXT:    vpcomgtb %xmm0, %xmm1, %xmm2
; XOPAVX2-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; XOPAVX2-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
; XOPAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
; XOPAVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
; XOPAVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; XOPAVX2-NEXT:    vzeroupper
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i8_signed_mem_reg:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsb %xmm0, %xmm1, %xmm3
; AVX512F-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
; AVX512F-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512F-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
; AVX512F-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i8_signed_mem_reg:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsb %xmm0, %xmm1, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm0, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm0, %xmm2, %xmm0
; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i8_signed_mem_reg:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtb %zmm0, %zmm1, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu8 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsb %xmm0, %xmm1, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsubb %xmm2, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
; AVX512BW-FALLBACK-NEXT:    vpmovwb %zmm0, %ymm0
; AVX512BW-FALLBACK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i8_signed_mem_reg:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtb %xmm0, %xmm1, %k1
; AVX512VLBW-NEXT:    vpminsb %xmm0, %xmm1, %xmm2
; AVX512VLBW-NEXT:    vpmaxsb %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm0, %xmm0
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm0, %xmm0
; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubb %xmm0, %xmm2, %xmm0 {%k1}
; AVX512VLBW-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; AVX512VLBW-NEXT:    retq
  %a1 = load <16 x i8>, ptr %a1_addr
  %t3 = icmp sgt <16 x i8> %a1, %a2 ; signed
  %t4 = select <16 x i1> %t3, <16 x i8> <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>, <16 x i8> <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>
  %t5 = select <16 x i1> %t3, <16 x i8> %a2, <16 x i8> %a1
  %t6 = select <16 x i1> %t3, <16 x i8> %a1, <16 x i8> %a2
  %t7 = sub <16 x i8> %t6, %t5
  %t8 = lshr <16 x i8> %t7, <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>
  %t9 = mul nsw <16 x i8> %t8, %t4 ; signed
  %a10 = add nsw <16 x i8> %t9, %a1 ; signed
  ret <16 x i8> %a10
}

define <16 x i8> @vec128_i8_signed_reg_mem(<16 x i8> %a1, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i8_signed_reg_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm3
; SSE2-NEXT:    movdqa %xmm0, %xmm2
; SSE2-NEXT:    pcmpgtb %xmm3, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm1
; SSE2-NEXT:    movdqa %xmm3, %xmm4
; SSE2-NEXT:    pcmpgtb %xmm0, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm0, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubb %xmm4, %xmm2
; SSE2-NEXT:    psrlw $1, %xmm2
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE2-NEXT:    movdqa %xmm2, %xmm3
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    movdqa %xmm1, %xmm4
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    pmullw %xmm3, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    pmullw %xmm2, %xmm1
; SSE2-NEXT:    pand %xmm3, %xmm1
; SSE2-NEXT:    packuswb %xmm4, %xmm1
; SSE2-NEXT:    paddb %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i8_signed_reg_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm1
; SSE41-NEXT:    movdqa %xmm0, %xmm2
; SSE41-NEXT:    pcmpgtb %xmm1, %xmm2
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    movdqa %xmm0, %xmm3
; SSE41-NEXT:    pminsb %xmm1, %xmm3
; SSE41-NEXT:    pmaxsb %xmm0, %xmm1
; SSE41-NEXT:    psubb %xmm3, %xmm1
; SSE41-NEXT:    psrlw $1, %xmm1
; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmullw %xmm1, %xmm2
; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
; SSE41-NEXT:    pand %xmm1, %xmm2
; SSE41-NEXT:    pmullw %xmm4, %xmm3
; SSE41-NEXT:    pand %xmm1, %xmm3
; SSE41-NEXT:    packuswb %xmm2, %xmm3
; SSE41-NEXT:    paddb %xmm3, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i8_signed_reg_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i8_signed_reg_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX2-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vzeroupper
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i8_signed_reg_mem:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOP-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOP-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i8_signed_reg_mem:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX1-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOPAVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i8_signed_reg_mem:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm1
; XOPAVX2-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; XOPAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; XOPAVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vzeroupper
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i8_signed_reg_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512F-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
; AVX512F-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i8_signed_reg_mem:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i8_signed_reg_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtb %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu8 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpmovwb %zmm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i8_signed_reg_mem:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %a2 = load <16 x i8>, ptr %a2_addr
  %t3 = icmp sgt <16 x i8> %a1, %a2 ; signed
  %t4 = select <16 x i1> %t3, <16 x i8> <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>, <16 x i8> <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>
  %t5 = select <16 x i1> %t3, <16 x i8> %a2, <16 x i8> %a1
  %t6 = select <16 x i1> %t3, <16 x i8> %a1, <16 x i8> %a2
  %t7 = sub <16 x i8> %t6, %t5
  %t8 = lshr <16 x i8> %t7, <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>
  %t9 = mul nsw <16 x i8> %t8, %t4 ; signed
  %a10 = add nsw <16 x i8> %t9, %a1 ; signed
  ret <16 x i8> %a10
}

define <16 x i8> @vec128_i8_signed_mem_mem(ptr %a1_addr, ptr %a2_addr) nounwind {
; SSE2-LABEL: vec128_i8_signed_mem_mem:
; SSE2:       # %bb.0:
; SSE2-NEXT:    movdqa (%rdi), %xmm1
; SSE2-NEXT:    movdqa (%rsi), %xmm3
; SSE2-NEXT:    movdqa %xmm1, %xmm2
; SSE2-NEXT:    pcmpgtb %xmm3, %xmm2
; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; SSE2-NEXT:    por %xmm2, %xmm0
; SSE2-NEXT:    movdqa %xmm3, %xmm4
; SSE2-NEXT:    pcmpgtb %xmm1, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm4, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm4
; SSE2-NEXT:    por %xmm5, %xmm4
; SSE2-NEXT:    movdqa %xmm1, %xmm5
; SSE2-NEXT:    pand %xmm2, %xmm5
; SSE2-NEXT:    pandn %xmm3, %xmm2
; SSE2-NEXT:    por %xmm5, %xmm2
; SSE2-NEXT:    psubb %xmm4, %xmm2
; SSE2-NEXT:    psrlw $1, %xmm2
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE2-NEXT:    movdqa %xmm2, %xmm3
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    movdqa %xmm0, %xmm4
; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE2-NEXT:    pmullw %xmm3, %xmm4
; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
; SSE2-NEXT:    pand %xmm3, %xmm4
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; SSE2-NEXT:    pmullw %xmm2, %xmm0
; SSE2-NEXT:    pand %xmm3, %xmm0
; SSE2-NEXT:    packuswb %xmm4, %xmm0
; SSE2-NEXT:    paddb %xmm1, %xmm0
; SSE2-NEXT:    retq
;
; SSE41-LABEL: vec128_i8_signed_mem_mem:
; SSE41:       # %bb.0:
; SSE41-NEXT:    movdqa (%rdi), %xmm1
; SSE41-NEXT:    movdqa (%rsi), %xmm2
; SSE41-NEXT:    movdqa %xmm1, %xmm3
; SSE41-NEXT:    pcmpgtb %xmm2, %xmm3
; SSE41-NEXT:    por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; SSE41-NEXT:    movdqa %xmm1, %xmm0
; SSE41-NEXT:    pminsb %xmm2, %xmm0
; SSE41-NEXT:    pmaxsb %xmm1, %xmm2
; SSE41-NEXT:    psubb %xmm0, %xmm2
; SSE41-NEXT:    psrlw $1, %xmm2
; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; SSE41-NEXT:    pmullw %xmm2, %xmm3
; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
; SSE41-NEXT:    pand %xmm2, %xmm3
; SSE41-NEXT:    pmullw %xmm4, %xmm0
; SSE41-NEXT:    pand %xmm2, %xmm0
; SSE41-NEXT:    packuswb %xmm3, %xmm0
; SSE41-NEXT:    paddb %xmm1, %xmm0
; SSE41-NEXT:    retq
;
; AVX1-FALLBACK-LABEL: vec128_i8_signed_mem_mem:
; AVX1-FALLBACK:       # %bb.0:
; AVX1-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX1-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX1-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX1-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX1-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX1-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX1-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; AVX1-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm3, %xmm3
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX1-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX1-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpand %xmm4, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
; AVX1-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX1-FALLBACK-NEXT:    retq
;
; AVX2-FALLBACK-LABEL: vec128_i8_signed_mem_mem:
; AVX2-FALLBACK:       # %bb.0:
; AVX2-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX2-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX2-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX2-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX2-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX2-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX2-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX2-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; AVX2-FALLBACK-NEXT:    vextracti128 $1, %ymm1, %xmm2
; AVX2-FALLBACK-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; AVX2-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX2-FALLBACK-NEXT:    vzeroupper
; AVX2-FALLBACK-NEXT:    retq
;
; XOP-FALLBACK-LABEL: vec128_i8_signed_mem_mem:
; XOP-FALLBACK:       # %bb.0:
; XOP-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; XOP-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; XOP-FALLBACK-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOP-FALLBACK-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOP-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOP-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOP-FALLBACK-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOP-FALLBACK-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOP-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOP-FALLBACK-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOP-FALLBACK-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOP-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOP-FALLBACK-NEXT:    retq
;
; XOPAVX1-LABEL: vec128_i8_signed_mem_mem:
; XOPAVX1:       # %bb.0:
; XOPAVX1-NEXT:    vmovdqa (%rdi), %xmm0
; XOPAVX1-NEXT:    vmovdqa (%rsi), %xmm1
; XOPAVX1-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX1-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; XOPAVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; XOPAVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; XOPAVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
; XOPAVX1-NEXT:    vpperm {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14],xmm3[0,2,4,6,8,10,12,14]
; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX1-NEXT:    retq
;
; XOPAVX2-LABEL: vec128_i8_signed_mem_mem:
; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    vmovdqa (%rdi), %xmm0
; XOPAVX2-NEXT:    vmovdqa (%rsi), %xmm1
; XOPAVX2-NEXT:    vpcomgtb %xmm1, %xmm0, %xmm2
; XOPAVX2-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; XOPAVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; XOPAVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; XOPAVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; XOPAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
; XOPAVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
; XOPAVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; XOPAVX2-NEXT:    vzeroupper
; XOPAVX2-NEXT:    retq
;
; AVX512F-LABEL: vec128_i8_signed_mem_mem:
; AVX512F:       # %bb.0:
; AVX512F-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512F-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm2
; AVX512F-NEXT:    vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT:    vpminsb %xmm1, %xmm0, %xmm3
; AVX512F-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512F-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512F-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512F-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
; AVX512F-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512F-NEXT:    vzeroupper
; AVX512F-NEXT:    retq
;
; AVX512VL-FALLBACK-LABEL: vec128_i8_signed_mem_mem:
; AVX512VL-FALLBACK:       # %bb.0:
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VL-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VL-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm3
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsrlw $1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpxor %xmm1, %xmm2, %xmm2
; AVX512VL-FALLBACK-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
; AVX512VL-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VL-FALLBACK-NEXT:    retq
;
; AVX512BW-FALLBACK-LABEL: vec128_i8_signed_mem_mem:
; AVX512BW-FALLBACK:       # %bb.0:
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512BW-FALLBACK-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512BW-FALLBACK-NEXT:    vpcmpgtb %zmm1, %zmm0, %k1
; AVX512BW-FALLBACK-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-FALLBACK-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
; AVX512BW-FALLBACK-NEXT:    vmovdqu8 %zmm2, %zmm3 {%k1}
; AVX512BW-FALLBACK-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512BW-FALLBACK-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
; AVX512BW-FALLBACK-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpmovwb %zmm1, %ymm1
; AVX512BW-FALLBACK-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512BW-FALLBACK-NEXT:    vzeroupper
; AVX512BW-FALLBACK-NEXT:    retq
;
; AVX512VLBW-LABEL: vec128_i8_signed_mem_mem:
; AVX512VLBW:       # %bb.0:
; AVX512VLBW-NEXT:    vmovdqa (%rdi), %xmm0
; AVX512VLBW-NEXT:    vmovdqa (%rsi), %xmm1
; AVX512VLBW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k1
; AVX512VLBW-NEXT:    vpminsb %xmm1, %xmm0, %xmm2
; AVX512VLBW-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm1
; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm1 {%k1}
; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT:    retq
  %a1 = load <16 x i8>, ptr %a1_addr
  %a2 = load <16 x i8>, ptr %a2_addr
  %t3 = icmp sgt <16 x i8> %a1, %a2 ; signed
  %t4 = select <16 x i1> %t3, <16 x i8> <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>, <16 x i8> <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>
  %t5 = select <16 x i1> %t3, <16 x i8> %a2, <16 x i8> %a1
  %t6 = select <16 x i1> %t3, <16 x i8> %a1, <16 x i8> %a2
  %t7 = sub <16 x i8> %t6, %t5
  %t8 = lshr <16 x i8> %t7, <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>
  %t9 = mul nsw <16 x i8> %t8, %t4 ; signed
  %a10 = add nsw <16 x i8> %t9, %a1 ; signed
  ret <16 x i8> %a10
}