| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=X64 --check-prefix=SSE --check-prefix=SSE2 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse3 | FileCheck %s --check-prefix=X64 --check-prefix=SSE --check-prefix=SSE3 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=X64 --check-prefix=SSE --check-prefix=SSSE3 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X64 --check-prefix=SSE --check-prefix=SSE41 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX1 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX2 |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX512VL |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512dq | FileCheck %s --check-prefix=X64 --check-prefix=AVX512VLBWDQ |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512cd,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512VLCD |
| ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512cd | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512CD |
| ; |
| ; Just one 32-bit run to make sure we do reasonable things for i64 lzcnt. |
| ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X32-SSE |
| |
| define <2 x i64> @testv2i64(<2 x i64> %in) nounwind { |
| ; SSE2-LABEL: testv2i64: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $16, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $32, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE2-NEXT: pxor %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubq %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323] |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: psrlq $2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: paddq %xmm2, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: psrlq $4, %xmm2 |
| ; SSE2-NEXT: paddq %xmm1, %xmm2 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE2-NEXT: pxor %xmm0, %xmm0 |
| ; SSE2-NEXT: psadbw %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv2i64: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $16, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $32, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE3-NEXT: pxor %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubq %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323] |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: psrlq $2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: paddq %xmm2, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: psrlq $4, %xmm2 |
| ; SSE3-NEXT: paddq %xmm1, %xmm2 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE3-NEXT: pxor %xmm0, %xmm0 |
| ; SSE3-NEXT: psadbw %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv2i64: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrld $16, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrld $16, %xmm1 |
| ; SSSE3-NEXT: paddd %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrlq $32, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrlq $32, %xmm1 |
| ; SSSE3-NEXT: paddq %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv2i64: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; SSE41-NEXT: psrld $16, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrld $16, %xmm1 |
| ; SSE41-NEXT: paddd %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; SSE41-NEXT: psrlq $32, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrlq $32, %xmm1 |
| ; SSE41-NEXT: paddq %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv2i64: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrld $16, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddd %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrlq $32, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrlq $32, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddq %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv2i64: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddd %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlq $32, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlq $32, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddq %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv2i64: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vplzcntq %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv2i64: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512CD-NEXT: vplzcntq %zmm0, %zmm0 |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv2i64: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrld $16, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrld $16, %xmm1 |
| ; X32-SSE-NEXT: paddd %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrlq $32, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrlq $32, %xmm1 |
| ; X32-SSE-NEXT: paddq %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| |
| %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 0) |
| ret <2 x i64> %out |
| } |
| |
| define <2 x i64> @testv2i64u(<2 x i64> %in) nounwind { |
| ; SSE2-LABEL: testv2i64u: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlq $16, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $32, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE2-NEXT: pxor %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlq $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubq %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323] |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: psrlq $2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: paddq %xmm2, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: psrlq $4, %xmm2 |
| ; SSE2-NEXT: paddq %xmm1, %xmm2 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE2-NEXT: pxor %xmm0, %xmm0 |
| ; SSE2-NEXT: psadbw %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv2i64u: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlq $16, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $32, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE3-NEXT: pxor %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlq $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubq %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323] |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: psrlq $2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: paddq %xmm2, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: psrlq $4, %xmm2 |
| ; SSE3-NEXT: paddq %xmm1, %xmm2 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE3-NEXT: pxor %xmm0, %xmm0 |
| ; SSE3-NEXT: psadbw %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv2i64u: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrld $16, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrld $16, %xmm1 |
| ; SSSE3-NEXT: paddd %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrlq $32, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrlq $32, %xmm1 |
| ; SSSE3-NEXT: paddq %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv2i64u: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; SSE41-NEXT: psrld $16, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrld $16, %xmm1 |
| ; SSE41-NEXT: paddd %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; SSE41-NEXT: psrlq $32, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrlq $32, %xmm1 |
| ; SSE41-NEXT: paddq %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv2i64u: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrld $16, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddd %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrlq $32, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrlq $32, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddq %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv2i64u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddd %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlq $32, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlq $32, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddq %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv2i64u: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vplzcntq %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv2i64u: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512CD-NEXT: vplzcntq %zmm0, %zmm0 |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv2i64u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqw %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrld $16, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrld $16, %xmm1 |
| ; X32-SSE-NEXT: paddd %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrlq $32, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrlq $32, %xmm1 |
| ; X32-SSE-NEXT: paddq %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| |
| %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 -1) |
| ret <2 x i64> %out |
| } |
| |
| define <4 x i32> @testv4i32(<4 x i32> %in) nounwind { |
| ; SSE2-LABEL: testv4i32: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrld $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrld $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $16, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: pcmpeqd %xmm2, %xmm2 |
| ; SSE2-NEXT: pxor %xmm1, %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psrld $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubd %xmm0, %xmm2 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459] |
| ; SSE2-NEXT: movdqa %xmm2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $2, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: paddd %xmm1, %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psrld $4, %xmm0 |
| ; SSE2-NEXT: paddd %xmm2, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: pxor %xmm1, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm2 |
| ; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3] |
| ; SSE2-NEXT: psadbw %xmm1, %xmm2 |
| ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] |
| ; SSE2-NEXT: psadbw %xmm1, %xmm0 |
| ; SSE2-NEXT: packuswb %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv4i32: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrld $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrld $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $16, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: pcmpeqd %xmm2, %xmm2 |
| ; SSE3-NEXT: pxor %xmm1, %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psrld $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubd %xmm0, %xmm2 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459] |
| ; SSE3-NEXT: movdqa %xmm2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $2, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: paddd %xmm1, %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psrld $4, %xmm0 |
| ; SSE3-NEXT: paddd %xmm2, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: pxor %xmm1, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm2 |
| ; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3] |
| ; SSE3-NEXT: psadbw %xmm1, %xmm2 |
| ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] |
| ; SSE3-NEXT: psadbw %xmm1, %xmm0 |
| ; SSE3-NEXT: packuswb %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv4i32: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrld $16, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrld $16, %xmm1 |
| ; SSSE3-NEXT: paddd %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv4i32: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; SSE41-NEXT: psrld $16, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrld $16, %xmm1 |
| ; SSE41-NEXT: paddd %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv4i32: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrld $16, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddd %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv4i32: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddd %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv4i32: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vplzcntd %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv4i32: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv4i32: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrld $16, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrld $16, %xmm1 |
| ; X32-SSE-NEXT: paddd %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| |
| %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 0) |
| ret <4 x i32> %out |
| } |
| |
| define <4 x i32> @testv4i32u(<4 x i32> %in) nounwind { |
| ; SSE2-LABEL: testv4i32u: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrld $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrld $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $16, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: pcmpeqd %xmm2, %xmm2 |
| ; SSE2-NEXT: pxor %xmm1, %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psrld $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubd %xmm0, %xmm2 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459] |
| ; SSE2-NEXT: movdqa %xmm2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: psrld $2, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: paddd %xmm1, %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psrld $4, %xmm0 |
| ; SSE2-NEXT: paddd %xmm2, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: pxor %xmm1, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm2 |
| ; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3] |
| ; SSE2-NEXT: psadbw %xmm1, %xmm2 |
| ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] |
| ; SSE2-NEXT: psadbw %xmm1, %xmm0 |
| ; SSE2-NEXT: packuswb %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv4i32u: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrld $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrld $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $16, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: pcmpeqd %xmm2, %xmm2 |
| ; SSE3-NEXT: pxor %xmm1, %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psrld $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubd %xmm0, %xmm2 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459] |
| ; SSE3-NEXT: movdqa %xmm2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: psrld $2, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: paddd %xmm1, %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psrld $4, %xmm0 |
| ; SSE3-NEXT: paddd %xmm2, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: pxor %xmm1, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm2 |
| ; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3] |
| ; SSE3-NEXT: psadbw %xmm1, %xmm2 |
| ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] |
| ; SSE3-NEXT: psadbw %xmm1, %xmm0 |
| ; SSE3-NEXT: packuswb %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv4i32u: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm3 |
| ; SSSE3-NEXT: pand %xmm1, %xmm3 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrld $16, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrld $16, %xmm1 |
| ; SSSE3-NEXT: paddd %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv4i32u: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm3 |
| ; SSE41-NEXT: pand %xmm1, %xmm3 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; SSE41-NEXT: psrld $16, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrld $16, %xmm1 |
| ; SSE41-NEXT: paddd %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv4i32u: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrld $16, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddd %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv4i32u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm2, %xmm1, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm2, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqw %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrld $16, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddd %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv4i32u: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vplzcntd %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv4i32u: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512CD-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv4i32u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm3 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqw %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrld $16, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrld $16, %xmm1 |
| ; X32-SSE-NEXT: paddd %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| |
| %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 -1) |
| ret <4 x i32> %out |
| } |
| |
| define <8 x i16> @testv8i16(<8 x i16> %in) nounwind { |
| ; SSE2-LABEL: testv8i16: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE2-NEXT: pxor %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubw %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107] |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: psrlw $2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: paddw %xmm2, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: psrlw $4, %xmm2 |
| ; SSE2-NEXT: paddw %xmm1, %xmm2 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psllw $8, %xmm0 |
| ; SSE2-NEXT: paddb %xmm2, %xmm0 |
| ; SSE2-NEXT: psrlw $8, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv8i16: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE3-NEXT: pxor %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubw %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107] |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: psrlw $2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: paddw %xmm2, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: psrlw $4, %xmm2 |
| ; SSE3-NEXT: paddw %xmm1, %xmm2 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psllw $8, %xmm0 |
| ; SSE3-NEXT: paddb %xmm2, %xmm0 |
| ; SSE3-NEXT: psrlw $8, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv8i16: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrlw $8, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv8i16: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; SSE41-NEXT: psrlw $8, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv8i16: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrlw $8, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv8i16: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv8i16: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vpmovzxwd {{.*#+}} 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 |
| ; AVX512VLCD-NEXT: vplzcntd %ymm0, %ymm0 |
| ; AVX512VLCD-NEXT: vpmovdw %ymm0, %xmm0 |
| ; AVX512VLCD-NEXT: vpsubw {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: vzeroupper |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv8i16: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: vpmovzxwd {{.*#+}} 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 |
| ; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512CD-NEXT: vpmovdw %zmm0, %ymm0 |
| ; AVX512CD-NEXT: vpsubw {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv8i16: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrlw $8, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 0) |
| ret <8 x i16> %out |
| } |
| |
| define <8 x i16> @testv8i16u(<8 x i16> %in) nounwind { |
| ; SSE2-LABEL: testv8i16u: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $1, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $2, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $4, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $8, %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE2-NEXT: pxor %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubw %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107] |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: pand %xmm0, %xmm2 |
| ; SSE2-NEXT: psrlw $2, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: paddw %xmm2, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE2-NEXT: psrlw $4, %xmm2 |
| ; SSE2-NEXT: paddw %xmm1, %xmm2 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE2-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE2-NEXT: psllw $8, %xmm0 |
| ; SSE2-NEXT: paddb %xmm2, %xmm0 |
| ; SSE2-NEXT: psrlw $8, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv8i16u: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $1, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $2, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $4, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $8, %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: pcmpeqd %xmm1, %xmm1 |
| ; SSE3-NEXT: pxor %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubw %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107] |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: pand %xmm0, %xmm2 |
| ; SSE3-NEXT: psrlw $2, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: paddw %xmm2, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm2 |
| ; SSE3-NEXT: psrlw $4, %xmm2 |
| ; SSE3-NEXT: paddw %xmm1, %xmm2 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm2 |
| ; SSE3-NEXT: movdqa %xmm2, %xmm0 |
| ; SSE3-NEXT: psllw $8, %xmm0 |
| ; SSE3-NEXT: paddb %xmm2, %xmm0 |
| ; SSE3-NEXT: psrlw $8, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv8i16u: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm3, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm4 |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSSE3-NEXT: psrlw $4, %xmm1 |
| ; SSSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm1, %xmm3 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSSE3-NEXT: pand %xmm4, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm3, %xmm1 |
| ; SSSE3-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; SSSE3-NEXT: psrlw $8, %xmm0 |
| ; SSSE3-NEXT: pand %xmm1, %xmm0 |
| ; SSSE3-NEXT: psrlw $8, %xmm1 |
| ; SSSE3-NEXT: paddw %xmm0, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv8i16u: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm3, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm4 |
| ; SSE41-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE41-NEXT: psrlw $4, %xmm1 |
| ; SSE41-NEXT: pand %xmm2, %xmm1 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm1, %xmm3 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; SSE41-NEXT: pand %xmm4, %xmm1 |
| ; SSE41-NEXT: paddb %xmm3, %xmm1 |
| ; SSE41-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; SSE41-NEXT: psrlw $8, %xmm0 |
| ; SSE41-NEXT: pand %xmm1, %xmm0 |
| ; SSE41-NEXT: psrlw $8, %xmm1 |
| ; SSE41-NEXT: paddw %xmm0, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv8i16u: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpsrlw $8, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX-NEXT: vpaddw %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv8i16u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm4, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm4, %xmm4, %xmm4 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm1, %xmm5 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm5, %xmm2, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm1, %xmm3, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $8, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpaddw %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512VLCD-LABEL: testv8i16u: |
| ; AVX512VLCD: # %bb.0: |
| ; AVX512VLCD-NEXT: vpmovzxwd {{.*#+}} 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 |
| ; AVX512VLCD-NEXT: vplzcntd %ymm0, %ymm0 |
| ; AVX512VLCD-NEXT: vpmovdw %ymm0, %xmm0 |
| ; AVX512VLCD-NEXT: vpsubw {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512VLCD-NEXT: vzeroupper |
| ; AVX512VLCD-NEXT: retq |
| ; |
| ; AVX512CD-LABEL: testv8i16u: |
| ; AVX512CD: # %bb.0: |
| ; AVX512CD-NEXT: vpmovzxwd {{.*#+}} 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 |
| ; AVX512CD-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512CD-NEXT: vpmovdw %zmm0, %ymm0 |
| ; AVX512CD-NEXT: vpsubw {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512CD-NEXT: vzeroupper |
| ; AVX512CD-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv8i16u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm3, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm4 |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm1 |
| ; X32-SSE-NEXT: psrlw $4, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm1, %xmm3 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm3, %xmm1 |
| ; X32-SSE-NEXT: pcmpeqb %xmm2, %xmm0 |
| ; X32-SSE-NEXT: psrlw $8, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm1, %xmm0 |
| ; X32-SSE-NEXT: psrlw $8, %xmm1 |
| ; X32-SSE-NEXT: paddw %xmm0, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 -1) |
| ret <8 x i16> %out |
| } |
| |
| define <16 x i8> @testv16i8(<16 x i8> %in) nounwind { |
| ; SSE2-LABEL: testv16i8: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $1, %xmm1 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $2, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $4, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE2-NEXT: pand %xmm2, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: pcmpeqd %xmm3, %xmm3 |
| ; SSE2-NEXT: pxor %xmm1, %xmm3 |
| ; SSE2-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE2-NEXT: psrlw $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubb %xmm0, %xmm3 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51] |
| ; SSE2-NEXT: movdqa %xmm3, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $2, %xmm3 |
| ; SSE2-NEXT: pand %xmm0, %xmm3 |
| ; SSE2-NEXT: paddb %xmm1, %xmm3 |
| ; SSE2-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE2-NEXT: psrlw $4, %xmm0 |
| ; SSE2-NEXT: paddb %xmm3, %xmm0 |
| ; SSE2-NEXT: pand %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv16i8: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $1, %xmm1 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $2, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $4, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: pcmpeqd %xmm3, %xmm3 |
| ; SSE3-NEXT: pxor %xmm1, %xmm3 |
| ; SSE3-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE3-NEXT: psrlw $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubb %xmm0, %xmm3 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51] |
| ; SSE3-NEXT: movdqa %xmm3, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $2, %xmm3 |
| ; SSE3-NEXT: pand %xmm0, %xmm3 |
| ; SSE3-NEXT: paddb %xmm1, %xmm3 |
| ; SSE3-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE3-NEXT: psrlw $4, %xmm0 |
| ; SSE3-NEXT: paddb %xmm3, %xmm0 |
| ; SSE3-NEXT: pand %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv16i8: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pand %xmm2, %xmm3 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm3, %xmm4 |
| ; SSSE3-NEXT: psrlw $4, %xmm0 |
| ; SSSE3-NEXT: pand %xmm2, %xmm0 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; SSSE3-NEXT: pand %xmm4, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm0, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv16i8: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pand %xmm2, %xmm3 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm1, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm3, %xmm4 |
| ; SSE41-NEXT: psrlw $4, %xmm0 |
| ; SSE41-NEXT: pand %xmm2, %xmm0 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; SSE41-NEXT: pand %xmm4, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm0, %xmm1 |
| ; SSE41-NEXT: paddb %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv16i8: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0 |
| ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm1 |
| ; AVX-NEXT: vpand %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpshufb %xmm0, %xmm3, %xmm0 |
| ; AVX-NEXT: vpaddb %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv16i8: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm0, %xmm3, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512-LABEL: testv16i8: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero |
| ; AVX512-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovdb %zmm0, %xmm0 |
| ; AVX512-NEXT: vpsubb {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512-NEXT: vzeroupper |
| ; AVX512-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv16i8: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm3 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm3, %xmm4 |
| ; X32-SSE-NEXT: psrlw $4, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm0 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm0, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 0) |
| ret <16 x i8> %out |
| } |
| |
| define <16 x i8> @testv16i8u(<16 x i8> %in) nounwind { |
| ; SSE2-LABEL: testv16i8u: |
| ; SSE2: # %bb.0: |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $1, %xmm1 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE2-NEXT: psrlw $2, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: por %xmm1, %xmm0 |
| ; SSE2-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $4, %xmm1 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE2-NEXT: pand %xmm2, %xmm1 |
| ; SSE2-NEXT: por %xmm0, %xmm1 |
| ; SSE2-NEXT: pcmpeqd %xmm3, %xmm3 |
| ; SSE2-NEXT: pxor %xmm1, %xmm3 |
| ; SSE2-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE2-NEXT: psrlw $1, %xmm0 |
| ; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE2-NEXT: psubb %xmm0, %xmm3 |
| ; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51] |
| ; SSE2-NEXT: movdqa %xmm3, %xmm1 |
| ; SSE2-NEXT: pand %xmm0, %xmm1 |
| ; SSE2-NEXT: psrlw $2, %xmm3 |
| ; SSE2-NEXT: pand %xmm0, %xmm3 |
| ; SSE2-NEXT: paddb %xmm1, %xmm3 |
| ; SSE2-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE2-NEXT: psrlw $4, %xmm0 |
| ; SSE2-NEXT: paddb %xmm3, %xmm0 |
| ; SSE2-NEXT: pand %xmm2, %xmm0 |
| ; SSE2-NEXT: retq |
| ; |
| ; SSE3-LABEL: testv16i8u: |
| ; SSE3: # %bb.0: |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $1, %xmm1 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE3-NEXT: psrlw $2, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: por %xmm1, %xmm0 |
| ; SSE3-NEXT: movdqa %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $4, %xmm1 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE3-NEXT: pand %xmm2, %xmm1 |
| ; SSE3-NEXT: por %xmm0, %xmm1 |
| ; SSE3-NEXT: pcmpeqd %xmm3, %xmm3 |
| ; SSE3-NEXT: pxor %xmm1, %xmm3 |
| ; SSE3-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE3-NEXT: psrlw $1, %xmm0 |
| ; SSE3-NEXT: pand {{.*}}(%rip), %xmm0 |
| ; SSE3-NEXT: psubb %xmm0, %xmm3 |
| ; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51] |
| ; SSE3-NEXT: movdqa %xmm3, %xmm1 |
| ; SSE3-NEXT: pand %xmm0, %xmm1 |
| ; SSE3-NEXT: psrlw $2, %xmm3 |
| ; SSE3-NEXT: pand %xmm0, %xmm3 |
| ; SSE3-NEXT: paddb %xmm1, %xmm3 |
| ; SSE3-NEXT: movdqa %xmm3, %xmm0 |
| ; SSE3-NEXT: psrlw $4, %xmm0 |
| ; SSE3-NEXT: paddb %xmm3, %xmm0 |
| ; SSE3-NEXT: pand %xmm2, %xmm0 |
| ; SSE3-NEXT: retq |
| ; |
| ; SSSE3-LABEL: testv16i8u: |
| ; SSSE3: # %bb.0: |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSSE3-NEXT: movdqa %xmm0, %xmm3 |
| ; SSSE3-NEXT: pand %xmm2, %xmm3 |
| ; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm4 |
| ; SSSE3-NEXT: pshufb %xmm3, %xmm4 |
| ; SSSE3-NEXT: psrlw $4, %xmm0 |
| ; SSSE3-NEXT: pand %xmm2, %xmm0 |
| ; SSSE3-NEXT: pxor %xmm2, %xmm2 |
| ; SSSE3-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; SSSE3-NEXT: pand %xmm4, %xmm2 |
| ; SSSE3-NEXT: pshufb %xmm0, %xmm1 |
| ; SSSE3-NEXT: paddb %xmm2, %xmm1 |
| ; SSSE3-NEXT: movdqa %xmm1, %xmm0 |
| ; SSSE3-NEXT: retq |
| ; |
| ; SSE41-LABEL: testv16i8u: |
| ; SSE41: # %bb.0: |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; SSE41-NEXT: movdqa %xmm0, %xmm3 |
| ; SSE41-NEXT: pand %xmm2, %xmm3 |
| ; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; SSE41-NEXT: movdqa %xmm1, %xmm4 |
| ; SSE41-NEXT: pshufb %xmm3, %xmm4 |
| ; SSE41-NEXT: psrlw $4, %xmm0 |
| ; SSE41-NEXT: pand %xmm2, %xmm0 |
| ; SSE41-NEXT: pxor %xmm2, %xmm2 |
| ; SSE41-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; SSE41-NEXT: pand %xmm4, %xmm2 |
| ; SSE41-NEXT: pshufb %xmm0, %xmm1 |
| ; SSE41-NEXT: paddb %xmm2, %xmm1 |
| ; SSE41-NEXT: movdqa %xmm1, %xmm0 |
| ; SSE41-NEXT: retq |
| ; |
| ; AVX-LABEL: testv16i8u: |
| ; AVX: # %bb.0: |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX-NEXT: vpsrlw $4, %xmm0, %xmm0 |
| ; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0 |
| ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 |
| ; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm1 |
| ; AVX-NEXT: vpand %xmm1, %xmm2, %xmm1 |
| ; AVX-NEXT: vpshufb %xmm0, %xmm3, %xmm0 |
| ; AVX-NEXT: vpaddb %xmm0, %xmm1, %xmm0 |
| ; AVX-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: testv16i8u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm2, %xmm3, %xmm2 |
| ; AVX512VLBWDQ-NEXT: vpsrlw $4, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm0, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpand %xmm1, %xmm2, %xmm1 |
| ; AVX512VLBWDQ-NEXT: vpshufb %xmm0, %xmm3, %xmm0 |
| ; AVX512VLBWDQ-NEXT: vpaddb %xmm0, %xmm1, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; AVX512-LABEL: testv16i8u: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero |
| ; AVX512-NEXT: vplzcntd %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovdb %zmm0, %xmm0 |
| ; AVX512-NEXT: vpsubb {{.*}}(%rip), %xmm0, %xmm0 |
| ; AVX512-NEXT: vzeroupper |
| ; AVX512-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: testv16i8u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] |
| ; X32-SSE-NEXT: movdqa %xmm0, %xmm3 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm3 |
| ; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0] |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm4 |
| ; X32-SSE-NEXT: pshufb %xmm3, %xmm4 |
| ; X32-SSE-NEXT: psrlw $4, %xmm0 |
| ; X32-SSE-NEXT: pand %xmm2, %xmm0 |
| ; X32-SSE-NEXT: pxor %xmm2, %xmm2 |
| ; X32-SSE-NEXT: pcmpeqb %xmm0, %xmm2 |
| ; X32-SSE-NEXT: pand %xmm4, %xmm2 |
| ; X32-SSE-NEXT: pshufb %xmm0, %xmm1 |
| ; X32-SSE-NEXT: paddb %xmm2, %xmm1 |
| ; X32-SSE-NEXT: movdqa %xmm1, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 -1) |
| ret <16 x i8> %out |
| } |
| |
| define <2 x i64> @foldv2i64() nounwind { |
| ; SSE-LABEL: foldv2i64: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movl $55, %eax |
| ; SSE-NEXT: movq %rax, %xmm0 |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv2i64: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: movl $55, %eax |
| ; NOBW-NEXT: vmovq %rax, %xmm0 |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv2i64: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: movl $55, %eax |
| ; AVX512VLBWDQ-NEXT: vmovq %rax, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv2i64: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movl $55, %eax |
| ; X32-SSE-NEXT: movd %eax, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> <i64 256, i64 -1>, i1 0) |
| ret <2 x i64> %out |
| } |
| |
| define <2 x i64> @foldv2i64u() nounwind { |
| ; SSE-LABEL: foldv2i64u: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movl $55, %eax |
| ; SSE-NEXT: movq %rax, %xmm0 |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv2i64u: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: movl $55, %eax |
| ; NOBW-NEXT: vmovq %rax, %xmm0 |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv2i64u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: movl $55, %eax |
| ; AVX512VLBWDQ-NEXT: vmovq %rax, %xmm0 |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv2i64u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movl $55, %eax |
| ; X32-SSE-NEXT: movd %eax, %xmm0 |
| ; X32-SSE-NEXT: retl |
| %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> <i64 256, i64 -1>, i1 -1) |
| ret <2 x i64> %out |
| } |
| |
| define <4 x i32> @foldv4i32() nounwind { |
| ; SSE-LABEL: foldv4i32: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv4i32: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv4i32: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv4i32: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; X32-SSE-NEXT: retl |
| %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> <i32 256, i32 -1, i32 0, i32 255>, i1 0) |
| ret <4 x i32> %out |
| } |
| |
| define <4 x i32> @foldv4i32u() nounwind { |
| ; SSE-LABEL: foldv4i32u: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv4i32u: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv4i32u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv4i32u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [23,0,32,24] |
| ; X32-SSE-NEXT: retl |
| %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> <i32 256, i32 -1, i32 0, i32 255>, i1 -1) |
| ret <4 x i32> %out |
| } |
| |
| define <8 x i16> @foldv8i16() nounwind { |
| ; SSE-LABEL: foldv8i16: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv8i16: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv8i16: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv8i16: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; X32-SSE-NEXT: retl |
| %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> <i16 256, i16 -1, i16 0, i16 255, i16 -65536, i16 7, i16 24, i16 88>, i1 0) |
| ret <8 x i16> %out |
| } |
| |
| define <8 x i16> @foldv8i16u() nounwind { |
| ; SSE-LABEL: foldv8i16u: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv8i16u: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv8i16u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv8i16u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [7,0,16,8,16,13,11,9] |
| ; X32-SSE-NEXT: retl |
| %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> <i16 256, i16 -1, i16 0, i16 255, i16 -65536, i16 7, i16 24, i16 88>, i1 -1) |
| ret <8 x i16> %out |
| } |
| |
| define <16 x i8> @foldv16i8() nounwind { |
| ; SSE-LABEL: foldv16i8: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv16i8: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv16i8: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv16i8: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; X32-SSE-NEXT: retl |
| %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> <i8 256, i8 -1, i8 0, i8 255, i8 -65536, i8 7, i8 24, i8 88, i8 -2, i8 254, i8 1, i8 2, i8 4, i8 8, i8 16, i8 32>, i1 0) |
| ret <16 x i8> %out |
| } |
| |
| define <16 x i8> @foldv16i8u() nounwind { |
| ; SSE-LABEL: foldv16i8u: |
| ; SSE: # %bb.0: |
| ; SSE-NEXT: movaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; SSE-NEXT: retq |
| ; |
| ; NOBW-LABEL: foldv16i8u: |
| ; NOBW: # %bb.0: |
| ; NOBW-NEXT: vmovaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; NOBW-NEXT: retq |
| ; |
| ; AVX512VLBWDQ-LABEL: foldv16i8u: |
| ; AVX512VLBWDQ: # %bb.0: |
| ; AVX512VLBWDQ-NEXT: vmovaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; AVX512VLBWDQ-NEXT: retq |
| ; |
| ; X32-SSE-LABEL: foldv16i8u: |
| ; X32-SSE: # %bb.0: |
| ; X32-SSE-NEXT: movaps {{.*#+}} xmm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2] |
| ; X32-SSE-NEXT: retl |
| %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> <i8 256, i8 -1, i8 0, i8 255, i8 -65536, i8 7, i8 24, i8 88, i8 -2, i8 254, i8 1, i8 2, i8 4, i8 8, i8 16, i8 32>, i1 -1) |
| ret <16 x i8> %out |
| } |
| |
| declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1) |
| declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1) |
| declare <8 x i16> @llvm.ctlz.v8i16(<8 x i16>, i1) |
| declare <16 x i8> @llvm.ctlz.v16i8(<16 x i8>, i1) |