Google Git
Sign in
llvm / llvm-project / llvm / e7e1a7ad6f753bcb646a0a80f8ac71809ba0829d / . / test / CodeGen / X86 / sshl_sat_vec.ll
blob: 531297af2a3094c955f9a9d0629d3c0eafe1fc32 [file] [log] [blame]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s --check-prefix=X64
; RUN: llc < %s -mtriple=x86_64-linux -mattr=+avx2 | FileCheck %s --check-prefix=X64-AVX2
; RUN: llc < %s -mtriple=i686 -mattr=cmov | FileCheck %s --check-prefix=X86
declare <2 x i64> @llvm.sshl.sat.v2i64(<2 x i64>, <2 x i64>)
declare <4 x i32> @llvm.sshl.sat.v4i32(<4 x i32>, <4 x i32>)
declare <8 x i16> @llvm.sshl.sat.v8i16(<8 x i16>, <8 x i16>)
declare <16 x i8> @llvm.sshl.sat.v16i8(<16 x i8>, <16 x i8>)
define <2 x i64> @vec_v2i64(<2 x i64> %x, <2 x i64> %y) nounwind {
; X64-LABEL: vec_v2i64:
; X64: # %bb.0:
; X64-NEXT: movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; X64-NEXT: movdqa %xmm2, %xmm3
; X64-NEXT: psrlq %xmm1, %xmm3
; X64-NEXT: pshufd {{.*#+}} xmm4 = xmm1[2,3,2,3]
; X64-NEXT: movdqa %xmm2, %xmm5
; X64-NEXT: psrlq %xmm4, %xmm5
; X64-NEXT: movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
; X64-NEXT: movdqa %xmm0, %xmm6
; X64-NEXT: psllq %xmm1, %xmm6
; X64-NEXT: movdqa %xmm0, %xmm3
; X64-NEXT: psllq %xmm4, %xmm3
; X64-NEXT: movdqa %xmm3, %xmm7
; X64-NEXT: movsd {{.*#+}} xmm3 = xmm6[0],xmm3[1]
; X64-NEXT: psrlq %xmm1, %xmm6
; X64-NEXT: psrlq %xmm4, %xmm7
; X64-NEXT: movsd {{.*#+}} xmm7 = xmm6[0],xmm7[1]
; X64-NEXT: xorpd %xmm5, %xmm7
; X64-NEXT: psubq %xmm5, %xmm7
; X64-NEXT: pcmpeqd %xmm0, %xmm7
; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm7[1,0,3,2]
; X64-NEXT: pand %xmm7, %xmm1
; X64-NEXT: andpd %xmm1, %xmm3
; X64-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
; X64-NEXT: pand %xmm2, %xmm0
; X64-NEXT: pxor %xmm5, %xmm5
; X64-NEXT: pcmpgtd %xmm4, %xmm5
; X64-NEXT: pcmpeqd %xmm4, %xmm4
; X64-NEXT: pxor %xmm5, %xmm4
; X64-NEXT: pandn %xmm4, %xmm2
; X64-NEXT: por %xmm0, %xmm2
; X64-NEXT: pandn %xmm2, %xmm1
; X64-NEXT: por %xmm3, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm0
; X64-NEXT: retq
;
; X64-AVX2-LABEL: vec_v2i64:
; X64-AVX2: # %bb.0:
; X64-AVX2-NEXT: vmovddup {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; X64-AVX2-NEXT: # xmm2 = mem[0,0]
; X64-AVX2-NEXT: vmovddup {{.*#+}} xmm3 = [9223372036854775807,9223372036854775807]
; X64-AVX2-NEXT: # xmm3 = mem[0,0]
; X64-AVX2-NEXT: vblendvpd %xmm0, %xmm2, %xmm3, %xmm3
; X64-AVX2-NEXT: vpsrlvq %xmm1, %xmm2, %xmm2
; X64-AVX2-NEXT: vpsllvq %xmm1, %xmm0, %xmm4
; X64-AVX2-NEXT: vpsrlvq %xmm1, %xmm4, %xmm1
; X64-AVX2-NEXT: vpxor %xmm2, %xmm1, %xmm1
; X64-AVX2-NEXT: vpsubq %xmm2, %xmm1, %xmm1
; X64-AVX2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; X64-AVX2-NEXT: vblendvpd %xmm0, %xmm4, %xmm3, %xmm0
; X64-AVX2-NEXT: retq
;
; X86-LABEL: vec_v2i64:
; X86: # %bb.0:
; X86-NEXT: pushl %ebp
; X86-NEXT: pushl %ebx
; X86-NEXT: pushl %edi
; X86-NEXT: pushl %esi
; X86-NEXT: subl $20, %esp
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movl %edx, %eax
; X86-NEXT: shll %cl, %eax
; X86-NEXT: shldl %cl, %edx, %esi
; X86-NEXT: xorl %edi, %edi
; X86-NEXT: testb $32, %cl
; X86-NEXT: cmovnel %eax, %esi
; X86-NEXT: cmovnel %edi, %eax
; X86-NEXT: movl %eax, (%esp) # 4-byte Spill
; X86-NEXT: movl %esi, %ebx
; X86-NEXT: sarl %cl, %ebx
; X86-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %esi, %eax
; X86-NEXT: movl %esi, %edx
; X86-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: sarl $31, %eax
; X86-NEXT: testb $32, %cl
; X86-NEXT: cmovel %ebx, %eax
; X86-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movb {{[0-9]+}}(%esp), %ch
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movl %esi, %eax
; X86-NEXT: movb %ch, %cl
; X86-NEXT: shll %cl, %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %ebp
; X86-NEXT: movl %ebp, %ebx
; X86-NEXT: shldl %cl, %esi, %ebx
; X86-NEXT: testb $32, %ch
; X86-NEXT: cmovnel %eax, %ebx
; X86-NEXT: cmovnel %edi, %eax
; X86-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %ebx, %edi
; X86-NEXT: sarl %cl, %edi
; X86-NEXT: movl %ebx, %esi
; X86-NEXT: sarl $31, %esi
; X86-NEXT: testb $32, %ch
; X86-NEXT: cmovel %edi, %esi
; X86-NEXT: movl (%esp), %eax # 4-byte Reload
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
; X86-NEXT: shrdl %cl, %edx, %eax
; X86-NEXT: testb $32, %cl
; X86-NEXT: cmovnel {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
; X86-NEXT: movb %ch, %cl
; X86-NEXT: shrdl %cl, %ebx, %edx
; X86-NEXT: testb $32, %ch
; X86-NEXT: cmovnel %edi, %edx
; X86-NEXT: xorl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
; X86-NEXT: xorl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
; X86-NEXT: sarl $31, %edi
; X86-NEXT: movl %edi, %ecx
; X86-NEXT: xorl $2147483647, %ecx # imm = 0x7FFFFFFF
; X86-NEXT: orl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
; X86-NEXT: notl %edi
; X86-NEXT: cmovel (%esp), %edi # 4-byte Folded Reload
; X86-NEXT: movl %edi, (%esp) # 4-byte Spill
; X86-NEXT: cmovel {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
; X86-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X86-NEXT: xorl %ebp, %esi
; X86-NEXT: sarl $31, %ebp
; X86-NEXT: movl %ebp, %edi
; X86-NEXT: xorl $2147483647, %edi # imm = 0x7FFFFFFF
; X86-NEXT: orl %esi, %edx
; X86-NEXT: notl %ebp
; X86-NEXT: cmovel {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
; X86-NEXT: cmovel %ebx, %edi
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %edi, 12(%eax)
; X86-NEXT: movl %ebp, 8(%eax)
; X86-NEXT: movl %ecx, 4(%eax)
; X86-NEXT: movl (%esp), %ecx # 4-byte Reload
; X86-NEXT: movl %ecx, (%eax)
; X86-NEXT: addl $20, %esp
; X86-NEXT: popl %esi
; X86-NEXT: popl %edi
; X86-NEXT: popl %ebx
; X86-NEXT: popl %ebp
; X86-NEXT: retl $4
%tmp = call <2 x i64> @llvm.sshl.sat.v2i64(<2 x i64> %x, <2 x i64> %y)
ret <2 x i64> %tmp
}
define <4 x i32> @vec_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {
; X64-LABEL: vec_v4i32:
; X64: # %bb.0:
; X64-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
; X64-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[2,3,3,3,4,5,6,7]
; X64-NEXT: pshuflw {{.*#+}} xmm4 = xmm1[0,1,1,1,4,5,6,7]
; X64-NEXT: pslld $23, %xmm1
; X64-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; X64-NEXT: cvttps2dq %xmm1, %xmm5
; X64-NEXT: movdqa %xmm0, %xmm1
; X64-NEXT: pmuludq %xmm5, %xmm1
; X64-NEXT: pshufd {{.*#+}} xmm6 = xmm1[0,2,2,3]
; X64-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
; X64-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
; X64-NEXT: pmuludq %xmm7, %xmm5
; X64-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
; X64-NEXT: punpckldq {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
; X64-NEXT: pshuflw {{.*#+}} xmm5 = xmm2[2,3,3,3,4,5,6,7]
; X64-NEXT: movdqa %xmm6, %xmm7
; X64-NEXT: psrad %xmm5, %xmm7
; X64-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7]
; X64-NEXT: movdqa %xmm1, %xmm5
; X64-NEXT: psrad %xmm2, %xmm5
; X64-NEXT: punpckhqdq {{.*#+}} xmm5 = xmm5[1],xmm7[1]
; X64-NEXT: movdqa %xmm6, %xmm2
; X64-NEXT: psrad %xmm3, %xmm2
; X64-NEXT: psrad %xmm4, %xmm1
; X64-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
; X64-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm5[0,3]
; X64-NEXT: pcmpeqd %xmm0, %xmm1
; X64-NEXT: pand %xmm1, %xmm6
; X64-NEXT: pxor %xmm2, %xmm2
; X64-NEXT: pcmpgtd %xmm0, %xmm2
; X64-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; X64-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; X64-NEXT: por %xmm2, %xmm0
; X64-NEXT: pandn %xmm0, %xmm1
; X64-NEXT: por %xmm6, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm0
; X64-NEXT: retq
;
; X64-AVX2-LABEL: vec_v4i32:
; X64-AVX2: # %bb.0:
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} xmm3 = [2147483647,2147483647,2147483647,2147483647]
; X64-AVX2-NEXT: vblendvps %xmm0, %xmm2, %xmm3, %xmm2
; X64-AVX2-NEXT: vpsllvd %xmm1, %xmm0, %xmm3
; X64-AVX2-NEXT: vpsravd %xmm1, %xmm3, %xmm1
; X64-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; X64-AVX2-NEXT: vblendvps %xmm0, %xmm3, %xmm2, %xmm0
; X64-AVX2-NEXT: retq
;
; X86-LABEL: vec_v4i32:
; X86: # %bb.0:
; X86-NEXT: pushl %ebp
; X86-NEXT: pushl %ebx
; X86-NEXT: pushl %edi
; X86-NEXT: pushl %esi
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movb {{[0-9]+}}(%esp), %ch
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movl %edx, %ebp
; X86-NEXT: shll %cl, %ebp
; X86-NEXT: movl %ebp, %edi
; X86-NEXT: sarl %cl, %edi
; X86-NEXT: xorl %ebx, %ebx
; X86-NEXT: testl %edx, %edx
; X86-NEXT: sets %bl
; X86-NEXT: addl $2147483647, %ebx # imm = 0x7FFFFFFF
; X86-NEXT: cmpl %edi, %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
; X86-NEXT: cmovel %ebp, %ebx
; X86-NEXT: movl %edi, %ebp
; X86-NEXT: movb %ch, %cl
; X86-NEXT: shll %cl, %ebp
; X86-NEXT: movl %ebp, %eax
; X86-NEXT: sarl %cl, %eax
; X86-NEXT: xorl %edx, %edx
; X86-NEXT: testl %edi, %edi
; X86-NEXT: sets %dl
; X86-NEXT: addl $2147483647, %edx # imm = 0x7FFFFFFF
; X86-NEXT: cmpl %eax, %edi
; X86-NEXT: cmovel %ebp, %edx
; X86-NEXT: movl %esi, %edi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shll %cl, %edi
; X86-NEXT: movl %edi, %ebp
; X86-NEXT: sarl %cl, %ebp
; X86-NEXT: xorl %eax, %eax
; X86-NEXT: testl %esi, %esi
; X86-NEXT: sets %al
; X86-NEXT: addl $2147483647, %eax # imm = 0x7FFFFFFF
; X86-NEXT: cmpl %ebp, %esi
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: cmovel %edi, %eax
; X86-NEXT: movl %esi, %edi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shll %cl, %edi
; X86-NEXT: movl %edi, %ebp
; X86-NEXT: sarl %cl, %ebp
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testl %esi, %esi
; X86-NEXT: sets %cl
; X86-NEXT: addl $2147483647, %ecx # imm = 0x7FFFFFFF
; X86-NEXT: cmpl %ebp, %esi
; X86-NEXT: cmovel %edi, %ecx
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movl %ecx, 12(%esi)
; X86-NEXT: movl %eax, 8(%esi)
; X86-NEXT: movl %edx, 4(%esi)
; X86-NEXT: movl %ebx, (%esi)
; X86-NEXT: movl %esi, %eax
; X86-NEXT: popl %esi
; X86-NEXT: popl %edi
; X86-NEXT: popl %ebx
; X86-NEXT: popl %ebp
; X86-NEXT: retl $4
%tmp = call <4 x i32> @llvm.sshl.sat.v4i32(<4 x i32> %x, <4 x i32> %y)
ret <4 x i32> %tmp
}
define <8 x i16> @vec_v8i16(<8 x i16> %x, <8 x i16> %y) nounwind {
; X64-LABEL: vec_v8i16:
; X64: # %bb.0:
; X64-NEXT: movdqa %xmm1, %xmm2
; X64-NEXT: punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
; X64-NEXT: pslld $23, %xmm2
; X64-NEXT: movdqa {{.*#+}} xmm3 = [1065353216,1065353216,1065353216,1065353216]
; X64-NEXT: paddd %xmm3, %xmm2
; X64-NEXT: cvttps2dq %xmm2, %xmm2
; X64-NEXT: pslld $16, %xmm2
; X64-NEXT: psrad $16, %xmm2
; X64-NEXT: movdqa %xmm1, %xmm4
; X64-NEXT: punpcklwd {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3]
; X64-NEXT: pslld $23, %xmm4
; X64-NEXT: paddd %xmm3, %xmm4
; X64-NEXT: cvttps2dq %xmm4, %xmm3
; X64-NEXT: pslld $16, %xmm3
; X64-NEXT: psrad $16, %xmm3
; X64-NEXT: packssdw %xmm2, %xmm3
; X64-NEXT: pmullw %xmm0, %xmm3
; X64-NEXT: psllw $12, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm2
; X64-NEXT: psraw $15, %xmm2
; X64-NEXT: movdqa %xmm3, %xmm4
; X64-NEXT: psraw $8, %xmm4
; X64-NEXT: pand %xmm2, %xmm4
; X64-NEXT: pandn %xmm3, %xmm2
; X64-NEXT: por %xmm4, %xmm2
; X64-NEXT: paddw %xmm1, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm4
; X64-NEXT: psraw $15, %xmm4
; X64-NEXT: movdqa %xmm4, %xmm5
; X64-NEXT: pandn %xmm2, %xmm5
; X64-NEXT: psraw $4, %xmm2
; X64-NEXT: pand %xmm4, %xmm2
; X64-NEXT: por %xmm5, %xmm2
; X64-NEXT: paddw %xmm1, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm4
; X64-NEXT: psraw $15, %xmm4
; X64-NEXT: movdqa %xmm4, %xmm5
; X64-NEXT: pandn %xmm2, %xmm5
; X64-NEXT: psraw $2, %xmm2
; X64-NEXT: pand %xmm4, %xmm2
; X64-NEXT: por %xmm5, %xmm2
; X64-NEXT: paddw %xmm1, %xmm1
; X64-NEXT: psraw $15, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm4
; X64-NEXT: pandn %xmm2, %xmm4
; X64-NEXT: psraw $1, %xmm2
; X64-NEXT: pand %xmm1, %xmm2
; X64-NEXT: por %xmm4, %xmm2
; X64-NEXT: pcmpeqw %xmm0, %xmm2
; X64-NEXT: pand %xmm2, %xmm3
; X64-NEXT: pxor %xmm1, %xmm1
; X64-NEXT: pcmpgtw %xmm0, %xmm1
; X64-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
; X64-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; X64-NEXT: por %xmm1, %xmm0
; X64-NEXT: pandn %xmm0, %xmm2
; X64-NEXT: por %xmm3, %xmm2
; X64-NEXT: movdqa %xmm2, %xmm0
; X64-NEXT: retq
;
; X64-AVX2-LABEL: vec_v8i16:
; X64-AVX2: # %bb.0:
; X64-AVX2-NEXT: vpmovzxwd {{.*#+}} 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
; X64-AVX2-NEXT: vpmovzxwd {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; X64-AVX2-NEXT: vpsllvd %ymm1, %ymm2, %ymm2
; X64-AVX2-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
; X64-AVX2-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
; X64-AVX2-NEXT: vpmovsxwd %xmm2, %ymm3
; X64-AVX2-NEXT: vpsravd %ymm1, %ymm3, %ymm1
; X64-AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
; X64-AVX2-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
; X64-AVX2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm1
; X64-AVX2-NEXT: vpxor %xmm3, %xmm3, %xmm3
; X64-AVX2-NEXT: vpcmpgtw %xmm0, %xmm3, %xmm0
; X64-AVX2-NEXT: vpbroadcastw {{.*#+}} xmm3 = [32767,32767,32767,32767,32767,32767,32767,32767]
; X64-AVX2-NEXT: vpblendvb %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm0
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
; X64-AVX2-NEXT: vzeroupper
; X64-AVX2-NEXT: retq
;
; X86-LABEL: vec_v8i16:
; X86: # %bb.0:
; X86-NEXT: pushl %ebp
; X86-NEXT: pushl %ebx
; X86-NEXT: pushl %edi
; X86-NEXT: pushl %esi
; X86-NEXT: subl $16, %esp
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movl {{[0-9]+}}(%esp), %edi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movl %edi, %ebx
; X86-NEXT: shll %cl, %ebx
; X86-NEXT: movswl %bx, %ebp
; X86-NEXT: sarl %cl, %ebp
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testw %di, %di
; X86-NEXT: sets %cl
; X86-NEXT: addl $32767, %ecx # imm = 0x7FFF
; X86-NEXT: cmpw %bp, %di
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: cmovel %ebx, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %esi, %edi
; X86-NEXT: movl %eax, %ecx
; X86-NEXT: shll %cl, %edi
; X86-NEXT: movswl %di, %ebx
; X86-NEXT: sarl %cl, %ebx
; X86-NEXT: xorl %eax, %eax
; X86-NEXT: testw %si, %si
; X86-NEXT: sets %al
; X86-NEXT: addl $32767, %eax # imm = 0x7FFF
; X86-NEXT: cmpw %bx, %si
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: cmovel %edi, %eax
; X86-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %edx, %esi
; X86-NEXT: shll %cl, %esi
; X86-NEXT: movswl %si, %edi
; X86-NEXT: sarl %cl, %edi
; X86-NEXT: xorl %eax, %eax
; X86-NEXT: testw %dx, %dx
; X86-NEXT: sets %al
; X86-NEXT: addl $32767, %eax # imm = 0x7FFF
; X86-NEXT: cmpw %di, %dx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: cmovel %esi, %eax
; X86-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shll %cl, %edx
; X86-NEXT: movswl %dx, %esi
; X86-NEXT: sarl %cl, %esi
; X86-NEXT: xorl %ebx, %ebx
; X86-NEXT: testw %ax, %ax
; X86-NEXT: sets %bl
; X86-NEXT: addl $32767, %ebx # imm = 0x7FFF
; X86-NEXT: cmpw %si, %ax
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: cmovel %edx, %ebx
; X86-NEXT: movl %ebx, (%esp) # 4-byte Spill
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shll %cl, %edx
; X86-NEXT: movswl %dx, %esi
; X86-NEXT: sarl %cl, %esi
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testw %ax, %ax
; X86-NEXT: sets %cl
; X86-NEXT: addl $32767, %ecx # imm = 0x7FFF
; X86-NEXT: cmpw %si, %ax
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: cmovel %edx, %ecx
; X86-NEXT: movl %ecx, %ebp
; X86-NEXT: movl %eax, %edx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shll %cl, %edx
; X86-NEXT: movswl %dx, %esi
; X86-NEXT: sarl %cl, %esi
; X86-NEXT: xorl %ebx, %ebx
; X86-NEXT: testw %ax, %ax
; X86-NEXT: sets %bl
; X86-NEXT: addl $32767, %ebx # imm = 0x7FFF
; X86-NEXT: cmpw %si, %ax
; X86-NEXT: cmovel %edx, %ebx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %esi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shll %cl, %esi
; X86-NEXT: movswl %si, %edi
; X86-NEXT: sarl %cl, %edi
; X86-NEXT: xorl %edx, %edx
; X86-NEXT: testw %ax, %ax
; X86-NEXT: sets %dl
; X86-NEXT: addl $32767, %edx # imm = 0x7FFF
; X86-NEXT: cmpw %di, %ax
; X86-NEXT: cmovel %esi, %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %esi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shll %cl, %esi
; X86-NEXT: movswl %si, %edi
; X86-NEXT: sarl %cl, %edi
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testw %ax, %ax
; X86-NEXT: sets %cl
; X86-NEXT: addl $32767, %ecx # imm = 0x7FFF
; X86-NEXT: cmpw %di, %ax
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movw %cx, 14(%eax)
; X86-NEXT: movw %dx, 12(%eax)
; X86-NEXT: movw %bx, 10(%eax)
; X86-NEXT: movw %bp, 8(%eax)
; X86-NEXT: movl (%esp), %ecx # 4-byte Reload
; X86-NEXT: movw %cx, 6(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movw %cx, 4(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movw %cx, 2(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movw %cx, (%eax)
; X86-NEXT: addl $16, %esp
; X86-NEXT: popl %esi
; X86-NEXT: popl %edi
; X86-NEXT: popl %ebx
; X86-NEXT: popl %ebp
; X86-NEXT: retl $4
%tmp = call <8 x i16> @llvm.sshl.sat.v8i16(<8 x i16> %x, <8 x i16> %y)
ret <8 x i16> %tmp
}
define <16 x i8> @vec_v16i8(<16 x i8> %x, <16 x i8> %y) nounwind {
; X64-LABEL: vec_v16i8:
; X64: # %bb.0:
; X64-NEXT: psllw $5, %xmm1
; X64-NEXT: pxor %xmm3, %xmm3
; X64-NEXT: pxor %xmm4, %xmm4
; X64-NEXT: pcmpgtb %xmm1, %xmm4
; X64-NEXT: movdqa %xmm0, %xmm2
; X64-NEXT: psllw $4, %xmm2
; X64-NEXT: pand %xmm4, %xmm2
; X64-NEXT: pandn %xmm0, %xmm4
; X64-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; X64-NEXT: por %xmm4, %xmm2
; X64-NEXT: punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm1[8],xmm4[9],xmm1[9],xmm4[10],xmm1[10],xmm4[11],xmm1[11],xmm4[12],xmm1[12],xmm4[13],xmm1[13],xmm4[14],xmm1[14],xmm4[15],xmm1[15]
; X64-NEXT: punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm1[0],xmm5[1],xmm1[1],xmm5[2],xmm1[2],xmm5[3],xmm1[3],xmm5[4],xmm1[4],xmm5[5],xmm1[5],xmm5[6],xmm1[6],xmm5[7],xmm1[7]
; X64-NEXT: paddb %xmm1, %xmm1
; X64-NEXT: pxor %xmm6, %xmm6
; X64-NEXT: pcmpgtb %xmm1, %xmm6
; X64-NEXT: movdqa %xmm6, %xmm7
; X64-NEXT: pandn %xmm2, %xmm7
; X64-NEXT: psllw $2, %xmm2
; X64-NEXT: pand %xmm6, %xmm2
; X64-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; X64-NEXT: por %xmm7, %xmm2
; X64-NEXT: paddb %xmm1, %xmm1
; X64-NEXT: pxor %xmm6, %xmm6
; X64-NEXT: pcmpgtb %xmm1, %xmm6
; X64-NEXT: movdqa %xmm6, %xmm1
; X64-NEXT: pandn %xmm2, %xmm1
; X64-NEXT: paddb %xmm2, %xmm2
; X64-NEXT: pand %xmm6, %xmm2
; X64-NEXT: por %xmm1, %xmm2
; X64-NEXT: punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm2[8],xmm6[9],xmm2[9],xmm6[10],xmm2[10],xmm6[11],xmm2[11],xmm6[12],xmm2[12],xmm6[13],xmm2[13],xmm6[14],xmm2[14],xmm6[15],xmm2[15]
; X64-NEXT: pxor %xmm1, %xmm1
; X64-NEXT: pcmpgtw %xmm4, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm7
; X64-NEXT: pandn %xmm6, %xmm7
; X64-NEXT: psraw $4, %xmm6
; X64-NEXT: pand %xmm1, %xmm6
; X64-NEXT: por %xmm7, %xmm6
; X64-NEXT: paddw %xmm4, %xmm4
; X64-NEXT: pxor %xmm1, %xmm1
; X64-NEXT: pcmpgtw %xmm4, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm7
; X64-NEXT: pandn %xmm6, %xmm7
; X64-NEXT: psraw $2, %xmm6
; X64-NEXT: pand %xmm1, %xmm6
; X64-NEXT: por %xmm7, %xmm6
; X64-NEXT: paddw %xmm4, %xmm4
; X64-NEXT: pxor %xmm1, %xmm1
; X64-NEXT: pcmpgtw %xmm4, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm4
; X64-NEXT: pandn %xmm6, %xmm4
; X64-NEXT: psraw $1, %xmm6
; X64-NEXT: pand %xmm1, %xmm6
; X64-NEXT: por %xmm4, %xmm6
; X64-NEXT: psrlw $8, %xmm6
; X64-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
; X64-NEXT: pxor %xmm4, %xmm4
; X64-NEXT: pcmpgtw %xmm5, %xmm4
; X64-NEXT: movdqa %xmm4, %xmm7
; X64-NEXT: pandn %xmm1, %xmm7
; X64-NEXT: psraw $4, %xmm1
; X64-NEXT: pand %xmm4, %xmm1
; X64-NEXT: por %xmm7, %xmm1
; X64-NEXT: paddw %xmm5, %xmm5
; X64-NEXT: pxor %xmm4, %xmm4
; X64-NEXT: pcmpgtw %xmm5, %xmm4
; X64-NEXT: movdqa %xmm4, %xmm7
; X64-NEXT: pandn %xmm1, %xmm7
; X64-NEXT: psraw $2, %xmm1
; X64-NEXT: pand %xmm4, %xmm1
; X64-NEXT: por %xmm7, %xmm1
; X64-NEXT: paddw %xmm5, %xmm5
; X64-NEXT: pxor %xmm4, %xmm4
; X64-NEXT: pcmpgtw %xmm5, %xmm4
; X64-NEXT: movdqa %xmm4, %xmm5
; X64-NEXT: pandn %xmm1, %xmm5
; X64-NEXT: psraw $1, %xmm1
; X64-NEXT: pand %xmm4, %xmm1
; X64-NEXT: por %xmm5, %xmm1
; X64-NEXT: psrlw $8, %xmm1
; X64-NEXT: packuswb %xmm6, %xmm1
; X64-NEXT: pcmpeqb %xmm0, %xmm1
; X64-NEXT: pand %xmm1, %xmm2
; X64-NEXT: pcmpgtb %xmm0, %xmm3
; X64-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
; X64-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; X64-NEXT: por %xmm3, %xmm0
; X64-NEXT: pandn %xmm0, %xmm1
; X64-NEXT: por %xmm2, %xmm1
; X64-NEXT: movdqa %xmm1, %xmm0
; X64-NEXT: retq
;
; X64-AVX2-LABEL: vec_v16i8:
; X64-AVX2: # %bb.0:
; X64-AVX2-NEXT: vpsllw $5, %xmm1, %xmm1
; X64-AVX2-NEXT: vpsllw $4, %xmm0, %xmm2
; X64-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm2
; X64-AVX2-NEXT: vpsllw $2, %xmm2, %xmm3
; X64-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
; X64-AVX2-NEXT: vpaddb %xmm1, %xmm1, %xmm4
; X64-AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
; X64-AVX2-NEXT: vpaddb %xmm2, %xmm2, %xmm3
; X64-AVX2-NEXT: vpaddb %xmm4, %xmm4, %xmm4
; X64-AVX2-NEXT: vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
; X64-AVX2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; X64-AVX2-NEXT: vpsraw $4, %xmm3, %xmm4
; X64-AVX2-NEXT: vpunpckhbw {{.*#+}} xmm5 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
; X64-AVX2-NEXT: vpblendvb %xmm5, %xmm4, %xmm3, %xmm3
; X64-AVX2-NEXT: vpsraw $2, %xmm3, %xmm4
; X64-AVX2-NEXT: vpaddw %xmm5, %xmm5, %xmm5
; X64-AVX2-NEXT: vpblendvb %xmm5, %xmm4, %xmm3, %xmm3
; X64-AVX2-NEXT: vpsraw $1, %xmm3, %xmm4
; X64-AVX2-NEXT: vpaddw %xmm5, %xmm5, %xmm5
; X64-AVX2-NEXT: vpblendvb %xmm5, %xmm4, %xmm3, %xmm3
; X64-AVX2-NEXT: vpsrlw $8, %xmm3, %xmm3
; X64-AVX2-NEXT: vpunpcklbw {{.*#+}} xmm4 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; X64-AVX2-NEXT: vpsraw $4, %xmm4, %xmm5
; X64-AVX2-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm5, %xmm4, %xmm4
; X64-AVX2-NEXT: vpsraw $2, %xmm4, %xmm5
; X64-AVX2-NEXT: vpaddw %xmm1, %xmm1, %xmm1
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm5, %xmm4, %xmm4
; X64-AVX2-NEXT: vpsraw $1, %xmm4, %xmm5
; X64-AVX2-NEXT: vpaddw %xmm1, %xmm1, %xmm1
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm5, %xmm4, %xmm1
; X64-AVX2-NEXT: vpsrlw $8, %xmm1, %xmm1
; X64-AVX2-NEXT: vpackuswb %xmm3, %xmm1, %xmm1
; X64-AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm1
; X64-AVX2-NEXT: vpbroadcastb {{.*#+}} xmm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
; X64-AVX2-NEXT: vpblendvb %xmm0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm0
; X64-AVX2-NEXT: vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
; X64-AVX2-NEXT: retq
;
; X86-LABEL: vec_v16i8:
; X86: # %bb.0:
; X86-NEXT: pushl %ebp
; X86-NEXT: pushl %ebx
; X86-NEXT: pushl %edi
; X86-NEXT: pushl %esi
; X86-NEXT: subl $44, %esp
; X86-NEXT: movb {{[0-9]+}}(%esp), %dh
; X86-NEXT: movb {{[0-9]+}}(%esp), %dl
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ebx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movb {{[0-9]+}}(%esp), %ch
; X86-NEXT: movb %ch, %bh
; X86-NEXT: shlb %cl, %bh
; X86-NEXT: movzbl %bh, %esi
; X86-NEXT: sarb %cl, %bh
; X86-NEXT: xorl %eax, %eax
; X86-NEXT: testb %ch, %ch
; X86-NEXT: sets %al
; X86-NEXT: addl $127, %eax
; X86-NEXT: cmpb %bh, %ch
; X86-NEXT: cmovel %esi, %eax
; X86-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %ebx, %eax
; X86-NEXT: movl %edx, %ecx
; X86-NEXT: shlb %cl, %al
; X86-NEXT: movzbl %al, %esi
; X86-NEXT: sarb %cl, %al
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %bl, %bl
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %al, %bl
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movb %dh, %al
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shlb %cl, %al
; X86-NEXT: movzbl %al, %esi
; X86-NEXT: sarb %cl, %al
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %dh, %dh
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %al, %dh
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movb {{[0-9]+}}(%esp), %ah
; X86-NEXT: movb %ah, %al
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: shlb %cl, %al
; X86-NEXT: movzbl %al, %esi
; X86-NEXT: sarb %cl, %al
; X86-NEXT: xorl %edx, %edx
; X86-NEXT: testb %ah, %ah
; X86-NEXT: sets %dl
; X86-NEXT: addl $127, %edx
; X86-NEXT: cmpb %al, %ah
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: cmovel %esi, %edx
; X86-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, (%esp) # 4-byte Spill
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, %ebp
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl %ecx, %edi
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, %edx
; X86-NEXT: shlb %cl, %dl
; X86-NEXT: movzbl %dl, %esi
; X86-NEXT: sarb %cl, %dl
; X86-NEXT: xorl %ebx, %ebx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %bl
; X86-NEXT: addl $127, %ebx
; X86-NEXT: cmpb %dl, %al
; X86-NEXT: cmovel %esi, %ebx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movb %al, %ah
; X86-NEXT: shlb %cl, %ah
; X86-NEXT: movzbl %ah, %esi
; X86-NEXT: sarb %cl, %ah
; X86-NEXT: xorl %edx, %edx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %dl
; X86-NEXT: addl $127, %edx
; X86-NEXT: cmpb %ah, %al
; X86-NEXT: cmovel %esi, %edx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movb %al, %ah
; X86-NEXT: shlb %cl, %ah
; X86-NEXT: movzbl %ah, %esi
; X86-NEXT: sarb %cl, %ah
; X86-NEXT: xorl %ecx, %ecx
; X86-NEXT: testb %al, %al
; X86-NEXT: sets %cl
; X86-NEXT: addl $127, %ecx
; X86-NEXT: cmpb %ah, %al
; X86-NEXT: cmovel %esi, %ecx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movb %cl, 15(%eax)
; X86-NEXT: movb %dl, 14(%eax)
; X86-NEXT: movb %bl, 13(%eax)
; X86-NEXT: movl %edi, %ecx
; X86-NEXT: movb %cl, 12(%eax)
; X86-NEXT: movl %ebp, %ecx
; X86-NEXT: movb %cl, 11(%eax)
; X86-NEXT: movl (%esp), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 10(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 9(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 8(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 7(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 6(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 5(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 4(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 3(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 2(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, 1(%eax)
; X86-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
; X86-NEXT: movb %cl, (%eax)
; X86-NEXT: addl $44, %esp
; X86-NEXT: popl %esi
; X86-NEXT: popl %edi
; X86-NEXT: popl %ebx
; X86-NEXT: popl %ebp
; X86-NEXT: retl $4
%tmp = call <16 x i8> @llvm.sshl.sat.v16i8(<16 x i8> %x, <16 x i8> %y)
ret <16 x i8> %tmp
}