derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
2476 lines
120 KiB
LLVM
2476 lines
120 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx,+pclmul | FileCheck %s --check-prefixes=AVX1
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+pclmul | FileCheck %s --check-prefixes=AVX2,AVX2-PCLMUL
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+vpclmulqdq | FileCheck %s --check-prefixes=AVX2,AVX2-VPCLMULQDQ
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vl,+vpclmulqdq | FileCheck %s --check-prefixes=AVX512
|
|
|
|
define <32 x i8> @clmul_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
|
|
; AVX1-LABEL: clmul_v32i8:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm3
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm6
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm6, %ymm6
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} ymm3 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
|
|
; AVX1-NEXT: vandps %ymm3, %ymm6, %ymm6
|
|
; AVX1-NEXT: vandnps %xmm4, %xmm3, %xmm4
|
|
; AVX1-NEXT: vpmaddubsw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vpsllw $8, %xmm4, %xmm4
|
|
; AVX1-NEXT: vandnps %xmm5, %xmm3, %xmm5
|
|
; AVX1-NEXT: vpmaddubsw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpsllw $8, %xmm5, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vorps %ymm4, %ymm6, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm8
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm8, %ymm7
|
|
; AVX1-NEXT: vandps %ymm3, %ymm7, %ymm7
|
|
; AVX1-NEXT: vandnps %xmm5, %xmm3, %xmm5
|
|
; AVX1-NEXT: vpmaddubsw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpsllw $8, %xmm5, %xmm5
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vorps %ymm5, %ymm7, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm4, %ymm5, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm8
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm8, %ymm7
|
|
; AVX1-NEXT: vandps %ymm3, %ymm7, %ymm7
|
|
; AVX1-NEXT: vandnps %xmm5, %xmm3, %xmm5
|
|
; AVX1-NEXT: vpmaddubsw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpsllw $8, %xmm5, %xmm5
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vorps %ymm5, %ymm7, %ymm5
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm6
|
|
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm2, %xmm8
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm9
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm8, %ymm9, %ymm8
|
|
; AVX1-NEXT: vandps %ymm3, %ymm8, %ymm8
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vandnps %xmm7, %xmm3, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm6, %ymm6
|
|
; AVX1-NEXT: vorps %ymm6, %ymm8, %ymm6
|
|
; AVX1-NEXT: vxorps %ymm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm8
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm8, %ymm7
|
|
; AVX1-NEXT: vandps %ymm3, %ymm7, %ymm7
|
|
; AVX1-NEXT: vandnps %xmm5, %xmm3, %xmm5
|
|
; AVX1-NEXT: vpmaddubsw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpsllw $8, %xmm5, %xmm5
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vorps %ymm5, %ymm7, %ymm5
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm6
|
|
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm2, %xmm8
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm9
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm8, %ymm9, %ymm8
|
|
; AVX1-NEXT: vandps %ymm3, %ymm8, %ymm8
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vandnps %xmm7, %xmm3, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm6, %ymm6
|
|
; AVX1-NEXT: vorps %ymm6, %ymm8, %ymm6
|
|
; AVX1-NEXT: vxorps %ymm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm6
|
|
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm2, %xmm8
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm9
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm8, %ymm9, %ymm8
|
|
; AVX1-NEXT: vandps %ymm3, %ymm8, %ymm8
|
|
; AVX1-NEXT: vandnps %xmm6, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vandnps %xmm7, %xmm3, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm6, %ymm6
|
|
; AVX1-NEXT: vorps %ymm6, %ymm8, %ymm6
|
|
; AVX1-NEXT: vxorps %ymm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm7
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm7, %ymm6
|
|
; AVX1-NEXT: vandps %ymm3, %ymm6, %ymm6
|
|
; AVX1-NEXT: vandnps %xmm1, %xmm3, %xmm1
|
|
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
|
|
; AVX1-NEXT: vandnps %xmm5, %xmm3, %xmm1
|
|
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm2, %xmm1
|
|
; AVX1-NEXT: vpsllw $8, %xmm1, %xmm1
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vorps %ymm0, %ymm6, %ymm0
|
|
; AVX1-NEXT: vxorps %ymm0, %ymm4, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmul_v32i8:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpbroadcastw {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
|
|
; AVX2-NEXT: vpand %ymm2, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpandn %ymm3, %ymm2, %ymm3
|
|
; AVX2-NEXT: vpmaddubsw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpsllw $8, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpor %ymm3, %ymm4, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpand %ymm2, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpandn %ymm4, %ymm2, %ymm4
|
|
; AVX2-NEXT: vpmaddubsw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpsllw $8, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpor %ymm4, %ymm5, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm3, %ymm4, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpand %ymm2, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpandn %ymm4, %ymm2, %ymm4
|
|
; AVX2-NEXT: vpmaddubsw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpsllw $8, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpor %ymm4, %ymm5, %ymm4
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm0, %ymm6
|
|
; AVX2-NEXT: vpand %ymm2, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpandn %ymm5, %ymm2, %ymm5
|
|
; AVX2-NEXT: vpmaddubsw %ymm5, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpsllw $8, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpor %ymm5, %ymm6, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpand %ymm2, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpandn %ymm4, %ymm2, %ymm4
|
|
; AVX2-NEXT: vpmaddubsw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpsllw $8, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpor %ymm4, %ymm5, %ymm4
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm0, %ymm6
|
|
; AVX2-NEXT: vpand %ymm2, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpandn %ymm5, %ymm2, %ymm5
|
|
; AVX2-NEXT: vpmaddubsw %ymm5, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpsllw $8, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpor %ymm5, %ymm6, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm0, %ymm6
|
|
; AVX2-NEXT: vpand %ymm2, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpandn %ymm5, %ymm2, %ymm5
|
|
; AVX2-NEXT: vpmaddubsw %ymm5, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpsllw $8, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpor %ymm5, %ymm6, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpand %ymm2, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpandn %ymm1, %ymm2, %ymm1
|
|
; AVX2-NEXT: vpmaddubsw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpsllw $8, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpor %ymm0, %ymm4, %ymm0
|
|
; AVX2-NEXT: vpxor %ymm0, %ymm3, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmul_v32i8:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
|
|
; AVX512-NEXT: vpandn %ymm2, %ymm4, %ymm2
|
|
; AVX512-NEXT: vpmaddubsw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpsllw $8, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm2 | (ymm3 & ymm4)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm3, %ymm4, %ymm3
|
|
; AVX512-NEXT: vpmaddubsw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpsllw $8, %ymm3, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm2 ^ (ymm3 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm2, %ymm4, %ymm2
|
|
; AVX512-NEXT: vpmaddubsw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpsllw $8, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm3 ^ (ymm2 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm3, %ymm4, %ymm3
|
|
; AVX512-NEXT: vpmaddubsw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpsllw $8, %ymm3, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm2 ^ (ymm3 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm2, %ymm4, %ymm2
|
|
; AVX512-NEXT: vpmaddubsw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpsllw $8, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm3 ^ (ymm2 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm3, %ymm4, %ymm3
|
|
; AVX512-NEXT: vpmaddubsw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpsllw $8, %ymm3, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm2 ^ (ymm3 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm5, %ymm5
|
|
; AVX512-NEXT: vpandn %ymm2, %ymm4, %ymm2
|
|
; AVX512-NEXT: vpmaddubsw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpsllw $8, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm3 ^ (ymm2 | ymm5)
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1
|
|
; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm4, %ymm3, %ymm3
|
|
; AVX512-NEXT: vpandn %ymm1, %ymm4, %ymm1
|
|
; AVX512-NEXT: vpmaddubsw %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpsllw $8, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm0 = ymm2 ^ (ymm0 | ymm3)
|
|
; AVX512-NEXT: retq
|
|
%res = call <32 x i8> @llvm.clmul.v32i8(<32 x i8> %a, <32 x i8> %b)
|
|
ret <32 x i8> %res
|
|
}
|
|
|
|
define <16 x i16> @clmul_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
|
|
; AVX1-LABEL: clmul_v16i16:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm2, %xmm4
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm0, %xmm3
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vxorps %ymm3, %ymm4, %ymm3
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vxorps %ymm4, %ymm3, %ymm3
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vxorps %ymm4, %ymm3, %ymm3
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vxorps %ymm4, %ymm3, %ymm3
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm2, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm5, %ymm5
|
|
; AVX1-NEXT: vxorps %ymm5, %ymm4, %ymm4
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: vxorps %ymm0, %ymm4, %ymm0
|
|
; AVX1-NEXT: vxorps %ymm0, %ymm3, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmul_v16i16:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm2
|
|
; AVX2-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpxor %ymm2, %ymm3, %ymm2
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpxor %ymm3, %ymm2, %ymm2
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpxor %ymm3, %ymm2, %ymm2
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpxor %ymm3, %ymm2, %ymm2
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
|
|
; AVX2-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm4
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpxor %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpxor %ymm0, %ymm3, %ymm0
|
|
; AVX2-NEXT: vpxor %ymm0, %ymm2, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmul_v16i16:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpxor %ymm2, %ymm3, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm4
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm4 = ymm4 ^ ymm2 ^ ymm3
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm3 ^ ymm4 ^ ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm4
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm4 = ymm4 ^ ymm3 ^ ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm3 ^ ymm4 ^ ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm4
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm4 = ymm4 ^ ymm3 ^ ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm3 = ymm3 ^ ymm4 ^ ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm1
|
|
; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm0 = ymm0 ^ ymm3 ^ ymm2
|
|
; AVX512-NEXT: retq
|
|
%res = call <16 x i16> @llvm.clmul.v16i16(<16 x i16> %a, <16 x i16> %b)
|
|
ret <16 x i16> %res
|
|
}
|
|
|
|
define <8 x i32> @clmul_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
|
|
; AVX1-LABEL: clmul_v8i32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm4
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX1-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX1-NEXT: vmovq %xmm5, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vmovq %xmm2, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm4, %xmm2
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm4 = xmm1[1,1,1,1]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm0[1,1,1,1]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX1-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm4
|
|
; AVX1-NEXT: vmovq %xmm4, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovq %xmm0, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm3, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmul_v8i32:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm4
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX2-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX2-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX2-NEXT: vmovq %xmm5, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX2-NEXT: vmovq %xmm2, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm4, %xmm2
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm1[1,1,1,1]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm5 = xmm0[1,1,1,1]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX2-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
|
|
; AVX2-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm4
|
|
; AVX2-NEXT: vmovq %xmm4, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm3, %xmm3
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm3, %xmm0
|
|
; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmul_v8i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX512-NEXT: vmovq %xmm2, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm2
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm4 = xmm1[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm4
|
|
; AVX512-NEXT: vmovq %xmm4, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm3, %xmm3
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovq %xmm0, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm3, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%res = call <8 x i32> @llvm.clmul.v8i32(<8 x i32> %a, <8 x i32> %b)
|
|
ret <8 x i32> %res
|
|
}
|
|
|
|
define <4 x i64> @clmul_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
|
|
; AVX1-LABEL: clmul_v4i64:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm4
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm3
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-PCLMUL-LABEL: clmul_v4i64:
|
|
; AVX2-PCLMUL: # %bb.0:
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm4, %ymm5, %ymm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX2-PCLMUL-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm4[0],ymm0[2],ymm4[2]
|
|
; AVX2-PCLMUL-NEXT: retq
|
|
;
|
|
; AVX2-VPCLMULQDQ-LABEL: clmul_v4i64:
|
|
; AVX2-VPCLMULQDQ: # %bb.0:
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX2-VPCLMULQDQ-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[2],ymm2[2]
|
|
; AVX2-VPCLMULQDQ-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmul_v4i64:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[2],ymm2[2]
|
|
; AVX512-NEXT: retq
|
|
%res = call <4 x i64> @llvm.clmul.v4i64(<4 x i64> %a, <4 x i64> %b)
|
|
ret <4 x i64> %res
|
|
}
|
|
|
|
define <32 x i8> @clmulr_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
|
|
; AVX1-LABEL: clmulr_v32i8:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovdqa %ymm1, %ymm3
|
|
; AVX1-NEXT: vmovdqa %ymm0, %ymm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX1-NEXT: vpand %xmm2, %xmm4, %xmm5
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm0 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX1-NEXT: vpshufb %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpsrlw $4, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpand %xmm2, %xmm4, %xmm6
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX1-NEXT: vpshufb %xmm6, %xmm4, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm5, %xmm14
|
|
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm10
|
|
; AVX1-NEXT: vpsrlw $4, %xmm10, %xmm5
|
|
; AVX1-NEXT: vpand %xmm2, %xmm5, %xmm5
|
|
; AVX1-NEXT: vpshufb %xmm5, %xmm4, %xmm11
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm11, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm14, %xmm8
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm11, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm9, %xmm14, %xmm12
|
|
; AVX1-NEXT: vpand %xmm5, %xmm12, %xmm12
|
|
; AVX1-NEXT: vpandn %xmm9, %xmm5, %xmm9
|
|
; AVX1-NEXT: vpmaddubsw %xmm9, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpsllw $8, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpor %xmm9, %xmm12, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm8, %xmm9, %xmm12
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm8 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
|
|
; AVX1-NEXT: vpand %xmm8, %xmm11, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm9, %xmm14, %xmm13
|
|
; AVX1-NEXT: vpand %xmm5, %xmm13, %xmm13
|
|
; AVX1-NEXT: vpandn %xmm9, %xmm5, %xmm9
|
|
; AVX1-NEXT: vpmaddubsw %xmm9, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpsllw $8, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpor %xmm9, %xmm13, %xmm13
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm11, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm11, %xmm14, %xmm15
|
|
; AVX1-NEXT: vpand %xmm5, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpandn %xmm11, %xmm5, %xmm11
|
|
; AVX1-NEXT: vpmaddubsw %xmm11, %xmm14, %xmm11
|
|
; AVX1-NEXT: vpsllw $8, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpor %xmm11, %xmm15, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm13, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm12, %xmm13
|
|
; AVX1-NEXT: vpand %xmm2, %xmm10, %xmm10
|
|
; AVX1-NEXT: vmovdqa %xmm0, %xmm9
|
|
; AVX1-NEXT: vpshufb %xmm10, %xmm0, %xmm15
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm10 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
|
|
; AVX1-NEXT: vpand %xmm10, %xmm15, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm11, %xmm14, %xmm12
|
|
; AVX1-NEXT: vpand %xmm5, %xmm12, %xmm12
|
|
; AVX1-NEXT: vpandn %xmm11, %xmm5, %xmm11
|
|
; AVX1-NEXT: vpmaddubsw %xmm11, %xmm14, %xmm11
|
|
; AVX1-NEXT: vpsllw $8, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpor %xmm11, %xmm12, %xmm12
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm11 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
|
|
; AVX1-NEXT: vpand %xmm11, %xmm15, %xmm0
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpandn %xmm0, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpmaddubsw %xmm0, %xmm14, %xmm0
|
|
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm12, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm12 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
|
|
; AVX1-NEXT: vpand %xmm12, %xmm15, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm13, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm13 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
|
|
; AVX1-NEXT: vpand %xmm13, %xmm15, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpshufb %xmm6, %xmm9, %xmm6
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm14
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm9, %xmm0
|
|
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm4, %xmm1
|
|
; AVX1-NEXT: vpor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsrlw $4, %xmm3, %xmm1
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm4, %xmm1
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm15
|
|
; AVX1-NEXT: vpand %xmm5, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm15, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand %xmm1, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm1, %xmm5, %xmm1
|
|
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpsllw $8, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpor %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm7, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm6, %xmm1
|
|
; AVX1-NEXT: vpand %xmm2, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpshufb %xmm3, %xmm9, %xmm3
|
|
; AVX1-NEXT: vpand %xmm3, %xmm10, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand %xmm3, %xmm11, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpand %xmm3, %xmm12, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpand %xmm3, %xmm13, %xmm3
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpandn %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vpmaddubsw %xmm3, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm14, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulr_v32i8:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpbroadcastb {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm4
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240,0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm4, %ymm3, %ymm4
|
|
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm5
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm0 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15,0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX2-NEXT: # ymm0 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpor %ymm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm5
|
|
; AVX2-NEXT: vpand %ymm2, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm0, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpbroadcastw {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
|
|
; AVX2-NEXT: vpand %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpandn %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpmaddubsw %ymm7, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpsllw $8, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
|
|
; AVX2-NEXT: vpmullw %ymm6, %ymm4, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm6, %ymm5, %ymm6
|
|
; AVX2-NEXT: vpmaddubsw %ymm6, %ymm4, %ymm6
|
|
; AVX2-NEXT: vpsllw $8, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpor %ymm6, %ymm9, %ymm6
|
|
; AVX2-NEXT: vpxor %ymm6, %ymm8, %ymm6
|
|
; AVX2-NEXT: vpxor %ymm6, %ymm7, %ymm6
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm3, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpand %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpandn %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpmaddubsw %ymm7, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpsllw $8, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpand %ymm5, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpandn %ymm1, %ymm5, %ymm1
|
|
; AVX2-NEXT: vpmaddubsw %ymm1, %ymm4, %ymm1
|
|
; AVX2-NEXT: vpsllw $8, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpor %ymm1, %ymm7, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm6, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm4
|
|
; AVX2-NEXT: vpshufb %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpor %ymm0, %ymm3, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulr_v32i8:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm3 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
|
|
; AVX512-NEXT: vpand %ymm3, %ymm2, %ymm4
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm5 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm5, %ymm4
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm6 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
|
|
; AVX512-NEXT: vpand %ymm6, %ymm2, %ymm7
|
|
; AVX512-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX512-NEXT: vpxor %ymm4, %ymm7, %ymm4
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm7 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
|
|
; AVX512-NEXT: vpand %ymm7, %ymm2, %ymm8
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm9 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm9, %ymm10
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm10, %ymm10
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm10 = ymm10 ^ ymm4 ^ ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
|
|
; AVX512-NEXT: vpand %ymm4, %ymm2, %ymm8
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm11 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm11, %ymm12
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm12, %ymm12
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm12 = ymm12 ^ ymm10 ^ ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm8 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm8, %ymm10
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm10, %ymm10
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm13 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm13, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm5, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm2 ^ ymm12 ^ ymm10
|
|
; AVX512-NEXT: vpsrlw $7, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
|
|
; AVX512-NEXT: vpmovdb %zmm2, %xmm2
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
|
|
; AVX512-NEXT: vpand %ymm3, %ymm1, %ymm3
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm6, %ymm1, %ymm5
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpxor %ymm3, %ymm5, %ymm3
|
|
; AVX512-NEXT: vpand %ymm7, %ymm1, %ymm5
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm1, %ymm9, %ymm6
|
|
; AVX512-NEXT: vpmullw %ymm6, %ymm0, %ymm6
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm6 = ymm6 ^ ymm3 ^ ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm11, %ymm4
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm4 = ymm4 ^ ymm6 ^ ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm8, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm13, %ymm1
|
|
; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm0 = ymm0 ^ ymm4 ^ ymm3
|
|
; AVX512-NEXT: vpsrlw $7, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
|
|
; AVX512-NEXT: vpmovdb %zmm0, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm2, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <32 x i8> %a to <32 x i16>
|
|
%b.ext = zext <32 x i8> %b to <32 x i16>
|
|
%clmul = call <32 x i16> @llvm.clmul.v32i16(<32 x i16> %a.ext, <32 x i16> %b.ext)
|
|
%res.ext = lshr <32 x i16> %clmul, splat (i16 7)
|
|
%res = trunc <32 x i16> %res.ext to <32 x i8>
|
|
ret <32 x i8> %res
|
|
}
|
|
|
|
define <16 x i16> @clmulr_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
|
|
; AVX1-LABEL: clmulr_v16i16:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovdqa %ymm1, %ymm5
|
|
; AVX1-NEXT: vmovdqa %ymm0, %ymm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm12 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm3
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm8 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX1-NEXT: vpshufb %xmm3, %xmm8, %xmm3
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm3, %xmm3
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm2, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpshufb %xmm7, %xmm6, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm14 = [2,2,2,2,2,2,2,2]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm9, %xmm10, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm8, %xmm0
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm1
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm13 = [2048,2048,2048,2048,2048,2048,2048,2048]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm13, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm3, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm11 = [4096,4096,4096,4096,4096,4096,4096,4096]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm11, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm9, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm10 = [8192,8192,8192,8192,8192,8192,8192,8192]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm10, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm9, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm9 = [16384,16384,16384,16384,16384,16384,16384,16384]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm9, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm15, %xmm15
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm7 = [32768,32768,32768,32768,32768,32768,32768,32768]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm3, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm15, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm3
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm5, %xmm4
|
|
; AVX1-NEXT: vpsrlw $4, %xmm4, %xmm1
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
|
|
; AVX1-NEXT: vpand %xmm1, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm15, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm14, %xmm15, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand %xmm2, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpshufb %xmm4, %xmm8, %xmm4
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand %xmm1, %xmm13, %xmm1
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm11, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm10, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpand %xmm7, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulr_v16i16:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm2 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
|
|
; AVX2-NEXT: # ymm2 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpbroadcastb {{.*#+}} ymm0 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX2-NEXT: vpand %ymm0, %ymm4, %ymm5
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240,0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm3, %ymm5
|
|
; AVX2-NEXT: vpsrlw $4, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpand %ymm0, %ymm4, %ymm6
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15,0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX2-NEXT: # ymm4 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm6, %ymm4, %ymm6
|
|
; AVX2-NEXT: vpor %ymm6, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm1, %ymm6
|
|
; AVX2-NEXT: vpsrlw $4, %ymm6, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm4, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand %ymm0, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpshufb %ymm6, %ymm3, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm5, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm8, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm7, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
|
|
; AVX2-NEXT: vpmullw %ymm6, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm7, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm2
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm0
|
|
; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
|
|
; AVX2-NEXT: vpor %ymm0, %ymm2, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulr_v16i16:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
|
|
; AVX512-NEXT: vextracti32x4 $3, %zmm1, %xmm2
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
|
|
; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm3
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX512-NEXT: vmovq %xmm2, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm2
|
|
; AVX512-NEXT: vextracti32x4 $2, %zmm1, %xmm3
|
|
; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm4
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm4[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm3, %xmm4, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm3
|
|
; AVX512-NEXT: vmovq %xmm3, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm3
|
|
; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm3, %ymm2
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm3
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm4[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm3, %xmm4, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm3
|
|
; AVX512-NEXT: vmovq %xmm3, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm3
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovq %xmm0, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX512-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
|
|
; AVX512-NEXT: vpsrld $15, %zmm0, %zmm0
|
|
; AVX512-NEXT: vpmovdw %zmm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <16 x i16> %a to <16 x i32>
|
|
%b.ext = zext <16 x i16> %b to <16 x i32>
|
|
%clmul = call <16 x i32> @llvm.clmul.v16i32(<16 x i32> %a.ext, <16 x i32> %b.ext)
|
|
%res.ext = lshr <16 x i32> %clmul, splat (i32 15)
|
|
%res = trunc <16 x i32> %res.ext to <16 x i16>
|
|
ret <16 x i16> %res
|
|
}
|
|
|
|
define <8 x i32> @clmulr_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
|
|
; AVX1-LABEL: clmulr_v8i32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1],xmm4[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm0[0,1],xmm4[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vmovq %xmm2, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vmovd %eax, %xmm2
|
|
; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm3
|
|
; AVX1-NEXT: vpsrlq $32, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vmovq %xmm3, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $1, %eax, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vmovq %xmm3, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm5 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vmovq %xmm3, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpaddd %xmm2, %xmm2, %xmm5
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm6 = xmm2[0,1],xmm4[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm7 = xmm3[0,1],xmm4[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vmovq %xmm6, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vmovd %eax, %xmm6
|
|
; AVX1-NEXT: vpsrlq $32, %xmm2, %xmm7
|
|
; AVX1-NEXT: vpsrlq $32, %xmm3, %xmm8
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vmovq %xmm7, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $1, %eax, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm4 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm7, %xmm4, %xmm4
|
|
; AVX1-NEXT: vmovq %xmm4, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm6, %xmm4
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm6 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm7 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vmovq %xmm6, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpaddd %xmm4, %xmm4, %xmm4
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm5, %ymm4
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm6 = xmm1[1,1,1,1]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm7 = xmm0[1,1,1,1]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm6
|
|
; AVX1-NEXT: vmovq %xmm6, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovq %xmm0, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm1
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX1-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX1-NEXT: vmovq %xmm5, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vmovq %xmm2, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpsrld $31, %xmm1, %xmm1
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vorps %ymm4, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulr_v8i32:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-NEXT: vpxor %xmm4, %xmm4, %xmm4
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm5 = xmm2[0],xmm4[1,2,3]
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm6 = xmm3[0],xmm4[1,2,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX2-NEXT: vmovq %xmm5, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vmovd %eax, %xmm5
|
|
; AVX2-NEXT: vpsrlq $32, %xmm2, %xmm6
|
|
; AVX2-NEXT: vpsrlq $32, %xmm3, %xmm7
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $1, %eax, %xmm5, %xmm5
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm6 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm7 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm5, %xmm5
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm6 = xmm1[0],xmm4[1,2,3]
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm7 = xmm0[0],xmm4[1,2,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vmovd %eax, %xmm6
|
|
; AVX2-NEXT: vpsrlq $32, %xmm1, %xmm7
|
|
; AVX2-NEXT: vpsrlq $32, %xmm0, %xmm8
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX2-NEXT: vmovq %xmm7, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $1, %eax, %xmm6, %xmm6
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm4 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm7, %xmm4, %xmm4
|
|
; AVX2-NEXT: vmovq %xmm4, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm6, %xmm4
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm6 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm7 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm4, %xmm4
|
|
; AVX2-NEXT: vinserti128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpaddd %ymm4, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm5
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm6 = xmm2[1,1,1,1]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm7 = xmm3[1,1,1,1]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX2-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX2-NEXT: vmovq %xmm2, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm5, %xmm2
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[1,1,1,1]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm6 = xmm0[1,1,1,1]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX2-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
|
|
; AVX2-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX2-NEXT: vmovq %xmm5, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm3, %xmm3
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm3, %xmm0
|
|
; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpsrld $31, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpor %ymm4, %ymm0, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulr_v8i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX512-NEXT: vpxor %xmm4, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm5 = xmm2[0],xmm4[1,2,3]
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm6 = xmm3[0],xmm4[1,2,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vmovd %eax, %xmm5
|
|
; AVX512-NEXT: vpsrlq $32, %xmm2, %xmm6
|
|
; AVX512-NEXT: vpsrlq $32, %xmm3, %xmm7
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $1, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm2[2],xmm6[2],xmm2[3],xmm6[3]
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm8 = xmm3[2],xmm6[2],xmm3[3],xmm6[3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX512-NEXT: vmovq %xmm7, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm7 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm8 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX512-NEXT: vmovq %xmm7, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm7 = xmm1[0],xmm4[1,2,3]
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm4 = xmm0[0],xmm4[1,2,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm4, %xmm4
|
|
; AVX512-NEXT: vmovq %xmm4, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vmovd %eax, %xmm4
|
|
; AVX512-NEXT: vpsrlq $32, %xmm1, %xmm7
|
|
; AVX512-NEXT: vpsrlq $32, %xmm0, %xmm8
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX512-NEXT: vmovq %xmm7, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm1[2],xmm6[2],xmm1[3],xmm6[3]
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm0[2],xmm6[2],xmm0[3],xmm6[3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm6, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm6 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm7 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vinserti128 $1, %xmm5, %ymm4, %ymm4
|
|
; AVX512-NEXT: vpaddd %ymm4, %ymm4, %ymm4
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm2[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX512-NEXT: vmovq %xmm2, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm2
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm3, %xmm3
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovq %xmm0, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm3, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpor %ymm4, %ymm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <8 x i32> %a to <8 x i64>
|
|
%b.ext = zext <8 x i32> %b to <8 x i64>
|
|
%clmul = call <8 x i64> @llvm.clmul.v8i64(<8 x i64> %a.ext, <8 x i64> %b.ext)
|
|
%res.ext = lshr <8 x i64> %clmul, splat (i64 31)
|
|
%res = trunc <8 x i64> %res.ext to <8 x i32>
|
|
ret <8 x i32> %res
|
|
}
|
|
|
|
define <4 x i64> @clmulr_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
|
|
; AVX1-LABEL: clmulr_v4i64:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm2
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm3
|
|
; AVX1-NEXT: vpunpckhqdq {{.*#+}} xmm4 = xmm3[1],xmm2[1]
|
|
; AVX1-NEXT: vpaddq %xmm4, %xmm4, %xmm4
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpunpckhqdq {{.*#+}} xmm1 = xmm0[1],xmm5[1]
|
|
; AVX1-NEXT: vpaddq %xmm1, %xmm1, %xmm1
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm4, %ymm1
|
|
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
|
|
; AVX1-NEXT: vpsrlq $63, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm5[0]
|
|
; AVX1-NEXT: vpsrlq $63, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
|
|
; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-PCLMUL-LABEL: clmulr_v4i64:
|
|
; AVX2-PCLMUL: # %bb.0:
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm4, %ymm5, %ymm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX2-PCLMUL-NEXT: vpunpckhqdq {{.*#+}} ymm1 = ymm0[1],ymm4[1],ymm0[3],ymm4[3]
|
|
; AVX2-PCLMUL-NEXT: vpaddq %ymm1, %ymm1, %ymm1
|
|
; AVX2-PCLMUL-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm4[0],ymm0[2],ymm4[2]
|
|
; AVX2-PCLMUL-NEXT: vpsrlq $63, %ymm0, %ymm0
|
|
; AVX2-PCLMUL-NEXT: vpor %ymm1, %ymm0, %ymm0
|
|
; AVX2-PCLMUL-NEXT: retq
|
|
;
|
|
; AVX2-VPCLMULQDQ-LABEL: clmulr_v4i64:
|
|
; AVX2-VPCLMULQDQ: # %bb.0:
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX2-VPCLMULQDQ-NEXT: vpunpckhqdq {{.*#+}} ymm1 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
|
|
; AVX2-VPCLMULQDQ-NEXT: vpaddq %ymm1, %ymm1, %ymm1
|
|
; AVX2-VPCLMULQDQ-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[2],ymm2[2]
|
|
; AVX2-VPCLMULQDQ-NEXT: vpsrlq $63, %ymm0, %ymm0
|
|
; AVX2-VPCLMULQDQ-NEXT: vpor %ymm1, %ymm0, %ymm0
|
|
; AVX2-VPCLMULQDQ-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulr_v4i64:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpunpckhqdq {{.*#+}} ymm1 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
|
|
; AVX512-NEXT: vpaddq %ymm1, %ymm1, %ymm1
|
|
; AVX512-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[2],ymm2[2]
|
|
; AVX512-NEXT: vpsrlq $63, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpor %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <4 x i64> %a to <4 x i128>
|
|
%b.ext = zext <4 x i64> %b to <4 x i128>
|
|
%clmul = call <4 x i128> @llvm.clmul.v4i128(<4 x i128> %a.ext, <4 x i128> %b.ext)
|
|
%res.ext = lshr <4 x i128> %clmul, splat (i128 63)
|
|
%res = trunc <4 x i128> %res.ext to <4 x i64>
|
|
ret <4 x i64> %res
|
|
}
|
|
|
|
define <32 x i8> @clmulh_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
|
|
; AVX1-LABEL: clmulh_v32i8:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovdqa %ymm1, %ymm3
|
|
; AVX1-NEXT: vmovdqa %ymm0, %ymm1
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm4
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm0 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX1-NEXT: vpshufb %xmm4, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm4
|
|
; AVX1-NEXT: vpand %xmm2, %xmm4, %xmm6
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX1-NEXT: vpshufb %xmm6, %xmm4, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm5, %xmm14
|
|
; AVX1-NEXT: vpsrlw $4, %xmm3, %xmm5
|
|
; AVX1-NEXT: vpand %xmm2, %xmm5, %xmm5
|
|
; AVX1-NEXT: vpshufb %xmm5, %xmm4, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm10, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm14, %xmm8
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm10, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm9, %xmm14, %xmm11
|
|
; AVX1-NEXT: vpand %xmm5, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpandn %xmm9, %xmm5, %xmm9
|
|
; AVX1-NEXT: vpmaddubsw %xmm9, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpsllw $8, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpor %xmm9, %xmm11, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm8, %xmm9, %xmm11
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm8 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
|
|
; AVX1-NEXT: vpand %xmm8, %xmm10, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm9, %xmm14, %xmm12
|
|
; AVX1-NEXT: vpand %xmm5, %xmm12, %xmm12
|
|
; AVX1-NEXT: vpandn %xmm9, %xmm5, %xmm9
|
|
; AVX1-NEXT: vpmaddubsw %xmm9, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpsllw $8, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpor %xmm9, %xmm12, %xmm12
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm10, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm10, %xmm14, %xmm13
|
|
; AVX1-NEXT: vpand %xmm5, %xmm13, %xmm13
|
|
; AVX1-NEXT: vpandn %xmm10, %xmm5, %xmm10
|
|
; AVX1-NEXT: vpmaddubsw %xmm10, %xmm14, %xmm10
|
|
; AVX1-NEXT: vpsllw $8, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpor %xmm10, %xmm13, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm12, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm11, %xmm13
|
|
; AVX1-NEXT: vpand %xmm2, %xmm3, %xmm10
|
|
; AVX1-NEXT: vmovdqa %xmm0, %xmm9
|
|
; AVX1-NEXT: vpshufb %xmm10, %xmm0, %xmm15
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm10 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
|
|
; AVX1-NEXT: vpand %xmm10, %xmm15, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm11, %xmm14, %xmm12
|
|
; AVX1-NEXT: vpand %xmm5, %xmm12, %xmm12
|
|
; AVX1-NEXT: vpandn %xmm11, %xmm5, %xmm11
|
|
; AVX1-NEXT: vpmaddubsw %xmm11, %xmm14, %xmm11
|
|
; AVX1-NEXT: vpsllw $8, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpor %xmm11, %xmm12, %xmm12
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm11 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
|
|
; AVX1-NEXT: vpand %xmm11, %xmm15, %xmm0
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpandn %xmm0, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpmaddubsw %xmm0, %xmm14, %xmm0
|
|
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm12, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm12 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
|
|
; AVX1-NEXT: vpand %xmm12, %xmm15, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm13, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm13 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
|
|
; AVX1-NEXT: vpand %xmm13, %xmm15, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm14, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm14, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpshufb %xmm6, %xmm9, %xmm6
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm14
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm3
|
|
; AVX1-NEXT: vpand %xmm2, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm3
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm15
|
|
; AVX1-NEXT: vpand %xmm5, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm15, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand %xmm3, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vpmaddubsw %xmm3, %xmm0, %xmm3
|
|
; AVX1-NEXT: vpsllw $8, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpor %xmm3, %xmm8, %xmm3
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm7, %xmm3
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm6, %xmm3
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpand %xmm1, %xmm10, %xmm6
|
|
; AVX1-NEXT: vpmullw %xmm6, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm5, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpandn %xmm6, %xmm5, %xmm6
|
|
; AVX1-NEXT: vpmaddubsw %xmm6, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpsllw $8, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpor %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vpand %xmm1, %xmm11, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpand %xmm1, %xmm12, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm0, %xmm8
|
|
; AVX1-NEXT: vpand %xmm5, %xmm8, %xmm8
|
|
; AVX1-NEXT: vpandn %xmm7, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpmaddubsw %xmm7, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpsllw $8, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpor %xmm7, %xmm8, %xmm7
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpxor %xmm6, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpand %xmm1, %xmm13, %xmm1
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpand %xmm5, %xmm6, %xmm6
|
|
; AVX1-NEXT: vpandn %xmm1, %xmm5, %xmm1
|
|
; AVX1-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm3, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpsrlw $1, %xmm14, %xmm1
|
|
; AVX1-NEXT: vpsrlw $1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
|
|
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulh_v32i8:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpbroadcastb {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm4
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240,0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm4, %ymm3, %ymm4
|
|
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm5
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm0 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15,0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX2-NEXT: # ymm0 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm0, %ymm5
|
|
; AVX2-NEXT: vpor %ymm5, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm5
|
|
; AVX2-NEXT: vpand %ymm2, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm0, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpbroadcastw {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
|
|
; AVX2-NEXT: vpand %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpandn %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpmaddubsw %ymm7, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpsllw $8, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
|
|
; AVX2-NEXT: vpmullw %ymm6, %ymm4, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm6, %ymm5, %ymm6
|
|
; AVX2-NEXT: vpmaddubsw %ymm6, %ymm4, %ymm6
|
|
; AVX2-NEXT: vpsllw $8, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpor %ymm6, %ymm9, %ymm6
|
|
; AVX2-NEXT: vpxor %ymm6, %ymm8, %ymm6
|
|
; AVX2-NEXT: vpxor %ymm6, %ymm7, %ymm6
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm3, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpand %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpandn %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpmaddubsw %ymm7, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpsllw $8, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm4, %ymm8, %ymm9
|
|
; AVX2-NEXT: vpand %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpandn %ymm8, %ymm5, %ymm8
|
|
; AVX2-NEXT: vpmaddubsw %ymm8, %ymm4, %ymm8
|
|
; AVX2-NEXT: vpsllw $8, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpor %ymm8, %ymm9, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm4, %ymm7
|
|
; AVX2-NEXT: vpand %ymm5, %ymm7, %ymm7
|
|
; AVX2-NEXT: vpandn %ymm1, %ymm5, %ymm1
|
|
; AVX2-NEXT: vpmaddubsw %ymm1, %ymm4, %ymm1
|
|
; AVX2-NEXT: vpsllw $8, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpor %ymm1, %ymm7, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm6, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm4
|
|
; AVX2-NEXT: vpshufb %ymm4, %ymm3, %ymm3
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpor %ymm0, %ymm3, %ymm0
|
|
; AVX2-NEXT: vpsrlw $1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulh_v32i8:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm3 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
|
|
; AVX512-NEXT: vpand %ymm3, %ymm2, %ymm4
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm5
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm5 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero,xmm5[8],zero,xmm5[9],zero,xmm5[10],zero,xmm5[11],zero,xmm5[12],zero,xmm5[13],zero,xmm5[14],zero,xmm5[15],zero
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm5, %ymm4
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm6 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
|
|
; AVX512-NEXT: vpand %ymm6, %ymm2, %ymm7
|
|
; AVX512-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX512-NEXT: vpxor %ymm4, %ymm7, %ymm4
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm7 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
|
|
; AVX512-NEXT: vpand %ymm7, %ymm2, %ymm8
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm9 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm9, %ymm10
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm10, %ymm10
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm10 = ymm10 ^ ymm4 ^ ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
|
|
; AVX512-NEXT: vpand %ymm4, %ymm2, %ymm8
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm11 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm11, %ymm12
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm12, %ymm12
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm12 = ymm12 ^ ymm10 ^ ymm8
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm8 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm8, %ymm10
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm10, %ymm10
|
|
; AVX512-NEXT: vpbroadcastd {{.*#+}} ymm13 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
|
|
; AVX512-NEXT: vpand %ymm2, %ymm13, %ymm2
|
|
; AVX512-NEXT: vpmullw %ymm2, %ymm5, %ymm2
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm2 = ymm2 ^ ymm12 ^ ymm10
|
|
; AVX512-NEXT: vpsrlw $8, %ymm2, %ymm2
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
|
|
; AVX512-NEXT: vpand %ymm3, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm6, %ymm1, %ymm5
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpxor %ymm3, %ymm5, %ymm3
|
|
; AVX512-NEXT: vpand %ymm7, %ymm1, %ymm5
|
|
; AVX512-NEXT: vpmullw %ymm5, %ymm0, %ymm5
|
|
; AVX512-NEXT: vpand %ymm1, %ymm9, %ymm6
|
|
; AVX512-NEXT: vpmullw %ymm6, %ymm0, %ymm6
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm6 = ymm6 ^ ymm3 ^ ymm5
|
|
; AVX512-NEXT: vpand %ymm4, %ymm1, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm11, %ymm4
|
|
; AVX512-NEXT: vpmullw %ymm4, %ymm0, %ymm4
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm4 = ymm4 ^ ymm6 ^ ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm8, %ymm3
|
|
; AVX512-NEXT: vpmullw %ymm3, %ymm0, %ymm3
|
|
; AVX512-NEXT: vpand %ymm1, %ymm13, %ymm1
|
|
; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpternlogq {{.*#+}} ymm0 = ymm0 ^ ymm4 ^ ymm3
|
|
; AVX512-NEXT: vpsrlw $8, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpackuswb %ymm2, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <32 x i8> %a to <32 x i16>
|
|
%b.ext = zext <32 x i8> %b to <32 x i16>
|
|
%clmul = call <32 x i16> @llvm.clmul.v32i16(<32 x i16> %a.ext, <32 x i16> %b.ext)
|
|
%res.ext = lshr <32 x i16> %clmul, splat (i16 8)
|
|
%res = trunc <32 x i16> %res.ext to <32 x i8>
|
|
ret <32 x i8> %res
|
|
}
|
|
|
|
define <16 x i16> @clmulh_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
|
|
; AVX1-LABEL: clmulh_v16i16:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovdqa %ymm1, %ymm5
|
|
; AVX1-NEXT: vmovdqa %ymm0, %ymm4
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm12 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm4, %xmm0
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm3
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm8 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX1-NEXT: vpshufb %xmm3, %xmm8, %xmm3
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm7
|
|
; AVX1-NEXT: vpand %xmm2, %xmm7, %xmm7
|
|
; AVX1-NEXT: vpshufb %xmm7, %xmm6, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm14 = [2,2,2,2,2,2,2,2]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm14, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm9, %xmm10, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm8, %xmm0
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm7, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm11, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm10, %xmm9, %xmm1
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm13 = [2048,2048,2048,2048,2048,2048,2048,2048]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm13, %xmm7
|
|
; AVX1-NEXT: vpmullw %xmm7, %xmm3, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm11 = [4096,4096,4096,4096,4096,4096,4096,4096]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm11, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm9, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm10 = [8192,8192,8192,8192,8192,8192,8192,8192]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm10, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm9, %xmm7
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm9 = [16384,16384,16384,16384,16384,16384,16384,16384]
|
|
; AVX1-NEXT: vpand %xmm0, %xmm9, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm3, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm7, %xmm15, %xmm15
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} xmm7 = [32768,32768,32768,32768,32768,32768,32768,32768]
|
|
; AVX1-NEXT: vpand %xmm7, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm3, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm15, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm3
|
|
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm1, %xmm4
|
|
; AVX1-NEXT: vpsrlw $4, %xmm4, %xmm1
|
|
; AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
|
|
; AVX1-NEXT: vpand %xmm1, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpmullw %xmm5, %xmm0, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm15, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm14, %xmm15, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand %xmm2, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpshufb %xmm4, %xmm8, %xmm4
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4, %xmm14
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm15
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm15, %xmm15
|
|
; AVX1-NEXT: vpxor %xmm15, %xmm14, %xmm14
|
|
; AVX1-NEXT: vpxor %xmm5, %xmm14, %xmm5
|
|
; AVX1-NEXT: vpand %xmm1, %xmm13, %xmm1
|
|
; AVX1-NEXT: vpmullw %xmm1, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm11, %xmm11
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm11, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm10, %xmm10
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm10, %xmm1
|
|
; AVX1-NEXT: vpand %xmm4, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpmullw %xmm0, %xmm9, %xmm9
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm9, %xmm1
|
|
; AVX1-NEXT: vpand %xmm7, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpmullw %xmm4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm5, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm12, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm1
|
|
; AVX1-NEXT: vpshufb %xmm1, %xmm8, %xmm1
|
|
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
|
|
; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vpsrlw $1, %xmm3, %xmm1
|
|
; AVX1-NEXT: vpsrlw $1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulh_v16i16:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm2 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
|
|
; AVX2-NEXT: # ymm2 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm0, %ymm4
|
|
; AVX2-NEXT: vpbroadcastb {{.*#+}} ymm0 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
|
|
; AVX2-NEXT: vpand %ymm0, %ymm4, %ymm5
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240,0,128,64,192,32,160,96,224,16,144,80,208,48,176,112,240]
|
|
; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm5, %ymm3, %ymm5
|
|
; AVX2-NEXT: vpsrlw $4, %ymm4, %ymm4
|
|
; AVX2-NEXT: vpand %ymm0, %ymm4, %ymm6
|
|
; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15,0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15]
|
|
; AVX2-NEXT: # ymm4 = mem[0,1,0,1]
|
|
; AVX2-NEXT: vpshufb %ymm6, %ymm4, %ymm6
|
|
; AVX2-NEXT: vpor %ymm6, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm1, %ymm6
|
|
; AVX2-NEXT: vpsrlw $4, %ymm6, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm1, %ymm4, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand %ymm0, %ymm6, %ymm6
|
|
; AVX2-NEXT: vpshufb %ymm6, %ymm3, %ymm6
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm9
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm9, %ymm9
|
|
; AVX2-NEXT: vpxor %ymm9, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; AVX2-NEXT: vpmullw %ymm1, %ymm5, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm8, %ymm1
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm7, %ymm1
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm7
|
|
; AVX2-NEXT: vpmullw %ymm7, %ymm5, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm8
|
|
; AVX2-NEXT: vpmullw %ymm5, %ymm8, %ymm8
|
|
; AVX2-NEXT: vpxor %ymm7, %ymm8, %ymm7
|
|
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
|
|
; AVX2-NEXT: vpmullw %ymm6, %ymm5, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm7, %ymm5
|
|
; AVX2-NEXT: vpxor %ymm5, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm2
|
|
; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
|
|
; AVX2-NEXT: vpsrlw $4, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm0, %ymm1, %ymm0
|
|
; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
|
|
; AVX2-NEXT: vpor %ymm0, %ymm2, %ymm0
|
|
; AVX2-NEXT: vpsrlw $1, %ymm0, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulh_v16i16:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
|
|
; AVX512-NEXT: vextracti32x4 $3, %zmm1, %xmm2
|
|
; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
|
|
; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm3
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX512-NEXT: vmovq %xmm2, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm2
|
|
; AVX512-NEXT: vextracti32x4 $2, %zmm1, %xmm3
|
|
; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm4
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm4[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm3, %xmm4, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm3
|
|
; AVX512-NEXT: vmovq %xmm3, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm3
|
|
; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm3, %ymm2
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm3
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm3[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm4[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm3, %xmm4, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm5, %xmm5
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm4, %xmm3
|
|
; AVX512-NEXT: vmovq %xmm3, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm5, %xmm3
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[1,1,1,1]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX512-NEXT: vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
|
|
; AVX512-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX512-NEXT: vmovq %xmm5, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[3,3,3,3]
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovq %xmm0, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX512-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
|
|
; AVX512-NEXT: vpsrld $16, %zmm0, %zmm0
|
|
; AVX512-NEXT: vpmovdw %zmm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <16 x i16> %a to <16 x i32>
|
|
%b.ext = zext <16 x i16> %b to <16 x i32>
|
|
%clmul = call <16 x i32> @llvm.clmul.v16i32(<16 x i32> %a.ext, <16 x i32> %b.ext)
|
|
%res.ext = lshr <16 x i32> %clmul, splat (i32 16)
|
|
%res = trunc <16 x i32> %res.ext to <16 x i16>
|
|
ret <16 x i16> %res
|
|
}
|
|
|
|
define <8 x i32> @clmulh_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
|
|
; AVX1-LABEL: clmulh_v8i32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm3[0,1],xmm2[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm5
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm6 = xmm5[0,1],xmm2[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm4, %xmm6, %xmm4
|
|
; AVX1-NEXT: vmovq %xmm4, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vmovd %eax, %xmm4
|
|
; AVX1-NEXT: vpsrlq $32, %xmm3, %xmm6
|
|
; AVX1-NEXT: vpsrlq $32, %xmm5, %xmm7
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vmovq %xmm6, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm5[2],xmm2[2],xmm5[3],xmm2[3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX1-NEXT: vmovq %xmm6, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm5 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX1-NEXT: vmovq %xmm3, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm4, %xmm3
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm1[0,1],xmm2[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpblendw {{.*#+}} xmm5 = xmm0[0,1],xmm2[2,3,4,5,6,7]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX1-NEXT: vmovq %xmm4, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vmovd %eax, %xmm4
|
|
; AVX1-NEXT: vpsrlq $32, %xmm1, %xmm5
|
|
; AVX1-NEXT: vpsrlq $32, %xmm0, %xmm6
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX1-NEXT: vmovq %xmm5, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
|
|
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm5, %xmm2, %xmm2
|
|
; AVX1-NEXT: vmovq %xmm2, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $2, %eax, %xmm4, %xmm2
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm1 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpsrldq {{.*#+}} xmm0 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovq %xmm0, %rax
|
|
; AVX1-NEXT: shrq $32, %rax
|
|
; AVX1-NEXT: vpinsrd $3, %eax, %xmm2, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: clmulh_v8i32:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
|
|
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm4 = xmm3[0],xmm2[1,2,3]
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm5
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm6 = xmm5[0],xmm2[1,2,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm4, %xmm6, %xmm4
|
|
; AVX2-NEXT: vmovq %xmm4, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vmovd %eax, %xmm4
|
|
; AVX2-NEXT: vpsrlq $32, %xmm3, %xmm6
|
|
; AVX2-NEXT: vpsrlq $32, %xmm5, %xmm7
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm5[2],xmm2[2],xmm5[3],xmm2[3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX2-NEXT: vmovq %xmm6, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm5 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX2-NEXT: vmovq %xmm3, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm4, %xmm3
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm4 = xmm1[0],xmm2[1,2,3]
|
|
; AVX2-NEXT: vpblendd {{.*#+}} xmm5 = xmm0[0],xmm2[1,2,3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX2-NEXT: vmovq %xmm4, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vmovd %eax, %xmm4
|
|
; AVX2-NEXT: vpsrlq $32, %xmm1, %xmm5
|
|
; AVX2-NEXT: vpsrlq $32, %xmm0, %xmm6
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm5, %xmm6, %xmm5
|
|
; AVX2-NEXT: vmovq %xmm5, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
|
|
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm5, %xmm2, %xmm2
|
|
; AVX2-NEXT: vmovq %xmm2, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $2, %eax, %xmm4, %xmm2
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm1 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpsrldq {{.*#+}} xmm0 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX2-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %rax
|
|
; AVX2-NEXT: shrq $32, %rax
|
|
; AVX2-NEXT: vpinsrd $3, %eax, %xmm2, %xmm0
|
|
; AVX2-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulh_v8i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm3
|
|
; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm4 = xmm3[0],xmm2[1,2,3]
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm5
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm6 = xmm5[0],xmm2[1,2,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm4, %xmm6, %xmm4
|
|
; AVX512-NEXT: vmovq %xmm4, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vmovd %eax, %xmm4
|
|
; AVX512-NEXT: vpsrlq $32, %xmm3, %xmm6
|
|
; AVX512-NEXT: vpsrlq $32, %xmm5, %xmm7
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm6, %xmm7, %xmm6
|
|
; AVX512-NEXT: vmovq %xmm6, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $1, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm7 = xmm3[2],xmm6[2],xmm3[3],xmm6[3]
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm8 = xmm5[2],xmm6[2],xmm5[3],xmm6[3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm7, %xmm8, %xmm7
|
|
; AVX512-NEXT: vmovq %xmm7, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm4, %xmm4
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm5 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm3, %xmm5, %xmm3
|
|
; AVX512-NEXT: vmovq %xmm3, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm4, %xmm3
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm4 = xmm1[0],xmm2[1,2,3]
|
|
; AVX512-NEXT: vpblendd {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm4, %xmm2, %xmm2
|
|
; AVX512-NEXT: vmovq %xmm2, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vmovd %eax, %xmm2
|
|
; AVX512-NEXT: vpsrlq $32, %xmm1, %xmm4
|
|
; AVX512-NEXT: vpsrlq $32, %xmm0, %xmm5
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX512-NEXT: vmovq %xmm4, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $1, %eax, %xmm2, %xmm2
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm4 = xmm1[2],xmm6[2],xmm1[3],xmm6[3]
|
|
; AVX512-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm6[2],xmm0[3],xmm6[3]
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm4, %xmm5, %xmm4
|
|
; AVX512-NEXT: vmovq %xmm4, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $2, %eax, %xmm2, %xmm2
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm1 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpsrldq {{.*#+}} xmm0 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; AVX512-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovq %xmm0, %rax
|
|
; AVX512-NEXT: shrq $32, %rax
|
|
; AVX512-NEXT: vpinsrd $3, %eax, %xmm2, %xmm0
|
|
; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <8 x i32> %a to <8 x i64>
|
|
%b.ext = zext <8 x i32> %b to <8 x i64>
|
|
%clmul = call <8 x i64> @llvm.clmul.v8i64(<8 x i64> %a.ext, <8 x i64> %b.ext)
|
|
%res.ext = lshr <8 x i64> %clmul, splat (i64 32)
|
|
%res = trunc <8 x i64> %res.ext to <8 x i32>
|
|
ret <8 x i32> %res
|
|
}
|
|
|
|
define <4 x i64> @clmulh_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
|
|
; AVX1-LABEL: clmulh_v4i64:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm4
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm4[1]
|
|
; AVX1-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm3
|
|
; AVX1-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm3[1]
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; AVX2-PCLMUL-LABEL: clmulh_v4i64:
|
|
; AVX2-PCLMUL: # %bb.0:
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm2, %xmm3, %xmm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm5
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm4, %ymm5, %ymm4
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm2, %xmm3, %xmm2
|
|
; AVX2-PCLMUL-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0
|
|
; AVX2-PCLMUL-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX2-PCLMUL-NEXT: vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm4[1],ymm0[3],ymm4[3]
|
|
; AVX2-PCLMUL-NEXT: retq
|
|
;
|
|
; AVX2-VPCLMULQDQ-LABEL: clmulh_v4i64:
|
|
; AVX2-VPCLMULQDQ: # %bb.0:
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX2-VPCLMULQDQ-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX2-VPCLMULQDQ-NEXT: vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
|
|
; AVX2-VPCLMULQDQ-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: clmulh_v4i64:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm2
|
|
; AVX512-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
|
|
; AVX512-NEXT: retq
|
|
%a.ext = zext <4 x i64> %a to <4 x i128>
|
|
%b.ext = zext <4 x i64> %b to <4 x i128>
|
|
%clmul = call <4 x i128> @llvm.clmul.v4i128(<4 x i128> %a.ext, <4 x i128> %b.ext)
|
|
%res.ext = lshr <4 x i128> %clmul, splat (i128 64)
|
|
%res = trunc <4 x i128> %res.ext to <4 x i64>
|
|
ret <4 x i64> %res
|
|
}
|