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
3470 lines
116 KiB
LLVM
3470 lines
116 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE2,X86-SSE2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE2,X64-SSE2
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE4,X86-SSE4
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE4,X64-SSE4
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE4,X86-SSE4
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE4,X64-SSE4
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1,X86-AVX1
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1,X64-AVX1
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2,X86-AVX2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2,X64-AVX2
|
|
; RUN: llc < %s -mtriple=x86_64- -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512bw,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512VL
|
|
|
|
;
|
|
; Truncate
|
|
;
|
|
|
|
define i1 @trunc_v2i64_v2i1(<2 x i64>) nounwind {
|
|
; SSE-LABEL: trunc_v2i64_v2i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: psllq $63, %xmm0
|
|
; SSE-NEXT: movmskpd %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: trunc_v2i64_v2i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpsllq $63, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v2i64_v2i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpsllq $63, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $3, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v2i64_v2i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllq $63, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $3, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v2i64_v2i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllq $63, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovmskpd %xmm0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <2 x i64> %0 to <2 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v2i1(<2 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v4i32_v4i1(<4 x i32>) nounwind {
|
|
; SSE-LABEL: trunc_v4i32_v4i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pslld $31, %xmm0
|
|
; SSE-NEXT: movmskps %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: trunc_v4i32_v4i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v4i32_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v4i32_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v4i32_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovmskps %xmm0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <4 x i32> %0 to <4 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v8i16_v8i1(<8 x i16>) nounwind {
|
|
; SSE-LABEL: trunc_v8i16_v8i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: psllw $15, %xmm0
|
|
; SSE-NEXT: packsswb %xmm0, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: trunc_v8i16_v8i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpsllw $15, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v8i16_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0
|
|
; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v8i16_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllw $15, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpmovw2m %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v8i16_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllw $15, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpmovw2m %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <8 x i16> %0 to <8 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v16i8_v16i1(<16 x i8>) nounwind {
|
|
; SSE-LABEL: trunc_v16i8_v16i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: psllw $7, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: trunc_v16i8_v16i1:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; AVX-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX-NEXT: xorb %ah, %al
|
|
; AVX-NEXT: setnp %al
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%a = trunc <16 x i8> %0 to <16 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v4i64_v4i1(<4 x i64>) nounwind {
|
|
; SSE-LABEL: trunc_v4i64_v4i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; SSE-NEXT: pslld $31, %xmm0
|
|
; SSE-NEXT: movmskps %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: trunc_v4i64_v4i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; AVX1-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovmskps %xmm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v4i64_v4i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpsllq $63, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v4i64_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpsllq $63, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v4i64_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllq $63, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v4i64_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllq $63, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <4 x i64> %0 to <4 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v8i32_v8i1(<8 x i32>) nounwind {
|
|
; SSE2-LABEL: trunc_v8i32_v8i1:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pslld $16, %xmm1
|
|
; SSE2-NEXT: psrad $16, %xmm1
|
|
; SSE2-NEXT: pslld $16, %xmm0
|
|
; SSE2-NEXT: psrad $16, %xmm0
|
|
; SSE2-NEXT: packssdw %xmm1, %xmm0
|
|
; SSE2-NEXT: psllw $15, %xmm0
|
|
; SSE2-NEXT: packsswb %xmm0, %xmm0
|
|
; SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE2-NEXT: testb %al, %al
|
|
; SSE2-NEXT: setnp %al
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: trunc_v8i32_v8i1:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pxor %xmm2, %xmm2
|
|
; SSE4-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
|
|
; SSE4-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
|
|
; SSE4-NEXT: packusdw %xmm1, %xmm0
|
|
; SSE4-NEXT: psllw $15, %xmm0
|
|
; SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE4-NEXT: testb %al, %al
|
|
; SSE4-NEXT: setnp %al
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: trunc_v8i32_v8i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpslld $31, %xmm0, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; AVX1-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
|
|
; AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v8i32_v8i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpslld $31, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v8i32_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpslld $31, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v8i32_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpslld $31, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v8i32_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <8 x i32> %0 to <8 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v16i16_v16i1(<16 x i16>) nounwind {
|
|
; SSE2-LABEL: trunc_v16i16_v16i1:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
|
|
; SSE2-NEXT: pand %xmm2, %xmm1
|
|
; SSE2-NEXT: pand %xmm2, %xmm0
|
|
; SSE2-NEXT: packuswb %xmm1, %xmm0
|
|
; SSE2-NEXT: psllw $7, %xmm0
|
|
; SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE2-NEXT: xorb %ah, %al
|
|
; SSE2-NEXT: setnp %al
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: trunc_v16i16_v16i1:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pmovzxbw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
|
|
; SSE4-NEXT: pand %xmm2, %xmm1
|
|
; SSE4-NEXT: pand %xmm2, %xmm0
|
|
; SSE4-NEXT: packuswb %xmm1, %xmm0
|
|
; SSE4-NEXT: psllw $7, %xmm0
|
|
; SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE4-NEXT: xorb %ah, %al
|
|
; SSE4-NEXT: setnp %al
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; X86-AVX1-LABEL: trunc_v16i16_v16i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb %ah, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: trunc_v16i16_v16i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb %ah, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: trunc_v16i16_v16i1:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorb %ah, %al
|
|
; X86-AVX2-NEXT: setnp %al
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: trunc_v16i16_v16i1:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorb %ah, %al
|
|
; X64-AVX2-NEXT: setnp %al
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: trunc_v16i16_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v16i16_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllw $15, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vpmovw2m %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v16i16_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllw $15, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vpmovw2m %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <16 x i16> %0 to <16 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v32i8_v32i1(<32 x i8>) nounwind {
|
|
; SSE-LABEL: trunc_v32i8_v32i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm1, %xmm0
|
|
; SSE-NEXT: psllw $7, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: trunc_v32i8_v32i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v32i8_v32i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpsllw $7, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v32i8_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v32i8_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllw $7, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v32i8_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllw $7, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <32 x i8> %0 to <32 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v8i64_v8i1(<8 x i64>) nounwind {
|
|
; X86-SSE2-LABEL: trunc_v8i64_v8i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; X86-SSE2-NEXT: pslld $16, %xmm0
|
|
; X86-SSE2-NEXT: psrad $16, %xmm0
|
|
; X86-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,2],mem[0,2]
|
|
; X86-SSE2-NEXT: pslld $16, %xmm2
|
|
; X86-SSE2-NEXT: psrad $16, %xmm2
|
|
; X86-SSE2-NEXT: packssdw %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: psllw $15, %xmm0
|
|
; X86-SSE2-NEXT: packsswb %xmm0, %xmm0
|
|
; X86-SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE2-NEXT: testb %al, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: trunc_v8i64_v8i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
|
|
; X64-SSE2-NEXT: pslld $16, %xmm2
|
|
; X64-SSE2-NEXT: psrad $16, %xmm2
|
|
; X64-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; X64-SSE2-NEXT: pslld $16, %xmm0
|
|
; X64-SSE2-NEXT: psrad $16, %xmm0
|
|
; X64-SSE2-NEXT: packssdw %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psllw $15, %xmm0
|
|
; X64-SSE2-NEXT: packsswb %xmm0, %xmm0
|
|
; X64-SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE2-NEXT: testb %al, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: trunc_v8i64_v8i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE4-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm3[1,2,3],xmm1[4],xmm3[5,6,7]
|
|
; X86-SSE4-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm3[1,2,3],xmm0[4],xmm3[5,6,7]
|
|
; X86-SSE4-NEXT: packusdw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3],xmm2[4],xmm3[5,6,7]
|
|
; X86-SSE4-NEXT: pblendw {{.*#+}} xmm3 = mem[0],xmm3[1,2,3],mem[4],xmm3[5,6,7]
|
|
; X86-SSE4-NEXT: packusdw %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: packusdw %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: psllw $15, %xmm0
|
|
; X86-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE4-NEXT: testb %al, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: trunc_v8i64_v8i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE4-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1,2,3],xmm3[4],xmm4[5,6,7]
|
|
; X64-SSE4-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1,2,3],xmm2[4],xmm4[5,6,7]
|
|
; X64-SSE4-NEXT: packusdw %xmm3, %xmm2
|
|
; X64-SSE4-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm4[1,2,3],xmm1[4],xmm4[5,6,7]
|
|
; X64-SSE4-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1,2,3],xmm0[4],xmm4[5,6,7]
|
|
; X64-SSE4-NEXT: packusdw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: packusdw %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: psllw $15, %xmm0
|
|
; X64-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE4-NEXT: testb %al, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: trunc_v8i64_v8i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
|
|
; AVX1-NEXT: vpslld $31, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpslld $31, %xmm1, %xmm1
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v8i64_v8i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm1[0,2],ymm0[4,6],ymm1[4,6]
|
|
; AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; AVX2-NEXT: vpslld $31, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v8i64_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v8i64_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllq $63, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v8i64_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllq $63, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vpmovq2m %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <8 x i64> %0 to <8 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v16i32_v16i1(<16 x i32>) nounwind {
|
|
; X86-SSE2-LABEL: trunc_v16i32_v16i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: packuswb %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pand 8(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: packuswb %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: packuswb %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: psllw $7, %xmm0
|
|
; X86-SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE2-NEXT: xorb %ah, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: trunc_v16i32_v16i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: packuswb %xmm3, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: packuswb %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: packuswb %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psllw $7, %xmm0
|
|
; X64-SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE2-NEXT: xorb %ah, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: trunc_v16i32_v16i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: pmovzxbd {{.*#+}} xmm3 = [255,255,255,255]
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE4-NEXT: packusdw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: pand 8(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: packusdw %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: packuswb %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: psllw $7, %xmm0
|
|
; X86-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorb %ah, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: trunc_v16i32_v16i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmovzxbd {{.*#+}} xmm4 = [255,255,255,255]
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm3
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm2
|
|
; X64-SSE4-NEXT: packusdw %xmm3, %xmm2
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: packusdw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: packuswb %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: psllw $7, %xmm0
|
|
; X64-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb %ah, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: trunc_v16i32_v16i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vbroadcastss {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255]
|
|
; AVX1-NEXT: vandps %ymm2, %ymm1, %ymm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3
|
|
; AVX1-NEXT: vpackusdw %xmm3, %xmm1, %xmm1
|
|
; AVX1-NEXT: vandps %ymm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vpackusdw %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v16i32_v16i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpackusdw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
|
|
; AVX2-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX2-NEXT: xorb %ah, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v16i32_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v16i32_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v16i32_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vpmovd2m %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <16 x i32> %0 to <16 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v32i16_v32i1(<32 x i16>) nounwind {
|
|
; X86-SSE2-LABEL: trunc_v32i16_v32i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: packuswb %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pand 8(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: packuswb %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pxor %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: psllw $7, %xmm2
|
|
; X86-SSE2-NEXT: pmovmskb %xmm2, %eax
|
|
; X86-SSE2-NEXT: xorb %ah, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: trunc_v32i16_v32i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: packuswb %xmm3, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: packuswb %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psllw $7, %xmm0
|
|
; X64-SSE2-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE2-NEXT: xorb %ah, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: trunc_v32i16_v32i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: pmovzxbw {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE4-NEXT: packuswb %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: pand 8(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: packuswb %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: pxor %xmm0, %xmm2
|
|
; X86-SSE4-NEXT: psllw $7, %xmm2
|
|
; X86-SSE4-NEXT: pmovmskb %xmm2, %eax
|
|
; X86-SSE4-NEXT: xorb %ah, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: trunc_v32i16_v32i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmovzxbw {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm3
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm2
|
|
; X64-SSE4-NEXT: packuswb %xmm3, %xmm2
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE4-NEXT: pand %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: packuswb %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: psllw $7, %xmm0
|
|
; X64-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb %ah, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: trunc_v32i16_v32i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vxorps %ymm1, %ymm0, %ymm0
|
|
; X86-AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb %ah, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: trunc_v32i16_v32i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vxorps %ymm1, %ymm0, %ymm0
|
|
; X64-AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb %ah, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: trunc_v32i16_v32i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpbroadcastw {{.*#+}} ymm2 = [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]
|
|
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpackuswb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; AVX2-NEXT: vpsllw $7, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v32i16_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512F-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v32i16_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllw $15, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpmovw2m %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v32i16_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllw $15, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vpmovw2m %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <32 x i16> %0 to <32 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @trunc_v64i8_v64i1(<64 x i8>) nounwind {
|
|
; X86-SSE-LABEL: trunc_v64i8_v64i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pxor 8(%ebp), %xmm1
|
|
; X86-SSE-NEXT: pxor %xmm0, %xmm1
|
|
; X86-SSE-NEXT: psllw $7, %xmm1
|
|
; X86-SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: trunc_v64i8_v64i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm1, %xmm0
|
|
; X64-SSE-NEXT: psllw $7, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: trunc_v64i8_v64i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vxorps %ymm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vxorps %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpsllw $7, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: trunc_v64i8_v64i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpsllw $7, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: trunc_v64i8_v64i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm3, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: trunc_v64i8_v64i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpsllw $7, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpmovb2m %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovq %k0, %rax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrq $32, %rax
|
|
; AVX512BW-NEXT: xorl %ecx, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: trunc_v64i8_v64i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpsllw $7, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vpmovb2m %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovq %k0, %rax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrq $32, %rax
|
|
; AVX512VL-NEXT: xorl %ecx, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = trunc <64 x i8> %0 to <64 x i1>
|
|
%b = call i1 @llvm.vector.reduce.xor.v64i1(<64 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
;
|
|
; Comparison With Zero
|
|
;
|
|
|
|
define i1 @icmp0_v2i64_v2i1(<2 x i64>) nounwind {
|
|
; SSE2-LABEL: icmp0_v2i64_v2i1:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pxor %xmm1, %xmm1
|
|
; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
|
|
; SSE2-NEXT: pand %xmm1, %xmm0
|
|
; SSE2-NEXT: movmskpd %xmm0, %eax
|
|
; SSE2-NEXT: testb %al, %al
|
|
; SSE2-NEXT: setnp %al
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: icmp0_v2i64_v2i1:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pxor %xmm1, %xmm1
|
|
; SSE4-NEXT: pcmpeqq %xmm0, %xmm1
|
|
; SSE4-NEXT: movmskpd %xmm1, %eax
|
|
; SSE4-NEXT: testb %al, %al
|
|
; SSE4-NEXT: setnp %al
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp0_v2i64_v2i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX1OR2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v2i64_v2i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512F-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $3, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v2i64_v2i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $3, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v2i64_v2i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmq %xmm0, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <2 x i64> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v2i1(<2 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v4i32_v4i1(<4 x i32>) nounwind {
|
|
; SSE-LABEL: icmp0_v4i32_v4i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm1, %xmm1
|
|
; SSE-NEXT: pcmpeqd %xmm0, %xmm1
|
|
; SSE-NEXT: movmskps %xmm1, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp0_v4i32_v4i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX1OR2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v4i32_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v4i32_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v4i32_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmd %xmm0, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <4 x i32> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v8i16_v8i1(<8 x i16>) nounwind {
|
|
; SSE-LABEL: icmp0_v8i16_v8i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm1, %xmm1
|
|
; SSE-NEXT: pcmpeqw %xmm0, %xmm1
|
|
; SSE-NEXT: packsswb %xmm1, %xmm1
|
|
; SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp0_v8i16_v8i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX1OR2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v8i16_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX512F-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v8i16_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmw %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v8i16_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmw %xmm0, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i16> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v16i8_v16i1(<16 x i8>) nounwind {
|
|
; SSE-LABEL: icmp0_v16i8_v16i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm1, %xmm1
|
|
; SSE-NEXT: pcmpeqb %xmm0, %xmm1
|
|
; SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp0_v16i8_v16i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX1OR2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1OR2-NEXT: xorb %ah, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v16i8_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX512F-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX512F-NEXT: xorb %ah, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v16i8_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v16i8_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmb %xmm0, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i8> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v4i64_v4i1(<4 x i64>) nounwind {
|
|
; SSE2-LABEL: icmp0_v4i64_v4i1:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pxor %xmm2, %xmm2
|
|
; SSE2-NEXT: pcmpeqd %xmm2, %xmm1
|
|
; SSE2-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,3],xmm1[1,3]
|
|
; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; SSE2-NEXT: andps %xmm2, %xmm0
|
|
; SSE2-NEXT: movmskps %xmm0, %eax
|
|
; SSE2-NEXT: testb %al, %al
|
|
; SSE2-NEXT: setnp %al
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: icmp0_v4i64_v4i1:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pxor %xmm2, %xmm2
|
|
; SSE4-NEXT: pcmpeqq %xmm2, %xmm1
|
|
; SSE4-NEXT: pcmpeqq %xmm2, %xmm0
|
|
; SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; SSE4-NEXT: movmskps %xmm0, %eax
|
|
; SSE4-NEXT: testb %al, %al
|
|
; SSE4-NEXT: setnp %al
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: icmp0_v4i64_v4i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqq %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v4i64_v4i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX2-NEXT: vpcmpeqq %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v4i64_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512F-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v4i64_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v4i64_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmq %ymm0, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <4 x i64> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v8i32_v8i1(<8 x i32>) nounwind {
|
|
; SSE-LABEL: icmp0_v8i32_v8i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm2, %xmm2
|
|
; SSE-NEXT: pcmpeqd %xmm2, %xmm1
|
|
; SSE-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; SSE-NEXT: packsswb %xmm0, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: icmp0_v8i32_v8i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v8i32_v8i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v8i32_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v8i32_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v8i32_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmd %ymm0, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i32> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v16i16_v16i1(<16 x i16>) nounwind {
|
|
; SSE-LABEL: icmp0_v16i16_v16i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm2, %xmm2
|
|
; SSE-NEXT: pcmpeqw %xmm2, %xmm1
|
|
; SSE-NEXT: pcmpeqw %xmm2, %xmm0
|
|
; SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: icmp0_v16i16_v16i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v16i16_v16i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX2-NEXT: xorb %ah, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v16i16_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX512F-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v16i16_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmw %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v16i16_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmw %ymm0, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i16> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v32i8_v32i1(<32 x i8>) nounwind {
|
|
; SSE-LABEL: icmp0_v32i8_v32i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pxor %xmm2, %xmm2
|
|
; SSE-NEXT: pcmpeqb %xmm2, %xmm1
|
|
; SSE-NEXT: pcmpeqb %xmm2, %xmm0
|
|
; SSE-NEXT: pxor %xmm1, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: icmp0_v32i8_v32i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v32i8_v32i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX2-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v32i8_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
|
|
; AVX512F-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v32i8_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v32i8_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmb %ymm0, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <32 x i8> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v8i64_v8i1(<8 x i64>) nounwind {
|
|
; X86-SSE2-LABEL: icmp0_v8i64_v8i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm1[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm4
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: packssdw %xmm4, %xmm1
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpeqd 8(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: packssdw %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: packssdw %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: packsswb %xmm1, %xmm1
|
|
; X86-SSE2-NEXT: pmovmskb %xmm1, %eax
|
|
; X86-SSE2-NEXT: testb %al, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: icmp0_v8i64_v8i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm5
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: packssdw %xmm5, %xmm3
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: packssdw %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: packssdw %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: packsswb %xmm1, %xmm1
|
|
; X64-SSE2-NEXT: pmovmskb %xmm1, %eax
|
|
; X64-SSE2-NEXT: testb %al, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: icmp0_v8i64_v8i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE4-NEXT: pcmpeqq %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pcmpeqq %xmm3, %xmm0
|
|
; X86-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pcmpeqq %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: pcmpeqq 8(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: packssdw %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: packssdw %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE4-NEXT: testb %al, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: icmp0_v8i64_v8i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm4, %xmm3
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm4, %xmm2
|
|
; X64-SSE4-NEXT: packssdw %xmm3, %xmm2
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm4, %xmm1
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: packssdw %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE4-NEXT: testb %al, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp0_v8i64_v8i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpcmpeqq %xmm3, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqq %xmm3, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vpcmpeqq %xmm3, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqq %xmm3, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v8i64_v8i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX2-NEXT: vpcmpeqq %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpcmpeqq %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v8i64_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v8i64_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v8i64_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmq %zmm0, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i64> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v16i32_v16i1(<16 x i32>) nounwind {
|
|
; X86-SSE-LABEL: icmp0_v16i32_v16i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE-NEXT: pcmpeqd %xmm3, %xmm1
|
|
; X86-SSE-NEXT: pcmpeqd %xmm3, %xmm0
|
|
; X86-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqd %xmm3, %xmm2
|
|
; X86-SSE-NEXT: pcmpeqd 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: packssdw %xmm3, %xmm2
|
|
; X86-SSE-NEXT: packsswb %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp0_v16i32_v16i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE-NEXT: pcmpeqd %xmm4, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqd %xmm4, %xmm2
|
|
; X64-SSE-NEXT: packssdw %xmm3, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqd %xmm4, %xmm1
|
|
; X64-SSE-NEXT: pcmpeqd %xmm4, %xmm0
|
|
; X64-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE-NEXT: packsswb %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp0_v16i32_v16i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX1-NEXT: vpcmpeqd %xmm3, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqd %xmm3, %xmm1, %xmm1
|
|
; AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
|
|
; AVX1-NEXT: vpcmpeqd %xmm3, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v16i32_v16i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX2-NEXT: vpcmpeqd %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpcmpeqd %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
|
|
; AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX2-NEXT: xorb %ah, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v16i32_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v16i32_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v16i32_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmd %zmm0, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i32> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v32i16_v32i1(<32 x i16>) nounwind {
|
|
; X86-SSE-LABEL: icmp0_v32i16_v32i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE-NEXT: pcmpeqw %xmm3, %xmm1
|
|
; X86-SSE-NEXT: pcmpeqw %xmm3, %xmm2
|
|
; X86-SSE-NEXT: pcmpeqw %xmm3, %xmm0
|
|
; X86-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqw 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pxor %xmm1, %xmm3
|
|
; X86-SSE-NEXT: packsswb %xmm3, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp0_v32i16_v32i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE-NEXT: pcmpeqw %xmm4, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqw %xmm4, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pcmpeqw %xmm4, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqw %xmm4, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X64-SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp0_v32i16_v32i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm3
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm4
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm4, %xmm3
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpacksswb %xmm0, %xmm3, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v32i16_v32i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX2-NEXT: vpcmpeqw %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpacksswb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v32i16_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpcmpeqw %ymm2, %ymm1, %ymm1
|
|
; AVX512F-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v32i16_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vptestnmw %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v32i16_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmw %zmm0, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <32 x i16> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp0_v64i8_v64i1(<64 x i8>) nounwind {
|
|
; X86-SSE-LABEL: icmp0_v64i8_v64i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pxor %xmm3, %xmm3
|
|
; X86-SSE-NEXT: pcmpeqb %xmm3, %xmm1
|
|
; X86-SSE-NEXT: pcmpeqb %xmm3, %xmm2
|
|
; X86-SSE-NEXT: pcmpeqb %xmm3, %xmm0
|
|
; X86-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqb 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pxor %xmm1, %xmm3
|
|
; X86-SSE-NEXT: pxor %xmm0, %xmm3
|
|
; X86-SSE-NEXT: pmovmskb %xmm3, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp0_v64i8_v64i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pxor %xmm4, %xmm4
|
|
; X64-SSE-NEXT: pcmpeqb %xmm4, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqb %xmm4, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pcmpeqb %xmm4, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqb %xmm4, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm0, %xmm1
|
|
; X64-SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp0_v64i8_v64i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm3
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm4
|
|
; AVX1-NEXT: vpxor %xmm3, %xmm4, %xmm3
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm0, %xmm3, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp0_v64i8_v64i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX2-NEXT: vpcmpeqb %ymm2, %ymm1, %ymm1
|
|
; AVX2-NEXT: vpcmpeqb %ymm2, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp0_v64i8_v64i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpcmpeqb %ymm2, %ymm1, %ymm1
|
|
; AVX512F-NEXT: vpcmpeqb %ymm2, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm3, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp0_v64i8_v64i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovq %k0, %rax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrq $32, %rax
|
|
; AVX512BW-NEXT: xorl %ecx, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp0_v64i8_v64i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vptestnmb %zmm0, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovq %k0, %rax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrq $32, %rax
|
|
; AVX512VL-NEXT: xorl %ecx, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <64 x i8> %0, zeroinitializer
|
|
%b = call i1 @llvm.vector.reduce.xor.v64i1(<64 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
; Comparison
|
|
;
|
|
|
|
define i1 @icmp_v2i64_v2i1(<2 x i64>, <2 x i64>) nounwind {
|
|
; SSE2-LABEL: icmp_v2i64_v2i1:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
|
|
; SSE2-NEXT: pand %xmm0, %xmm1
|
|
; SSE2-NEXT: movmskpd %xmm1, %eax
|
|
; SSE2-NEXT: testb %al, %al
|
|
; SSE2-NEXT: setnp %al
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: icmp_v2i64_v2i1:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pcmpeqq %xmm1, %xmm0
|
|
; SSE4-NEXT: movmskpd %xmm0, %eax
|
|
; SSE4-NEXT: testb %al, %al
|
|
; SSE4-NEXT: setnp %al
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp_v2i64_v2i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskpd %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v2i64_v2i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512F-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $3, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v2i64_v2i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $3, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v2i64_v2i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqq %xmm1, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <2 x i64> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v2i1(<2 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v4i32_v4i1(<4 x i32>, <4 x i32>) nounwind {
|
|
; SSE-LABEL: icmp_v4i32_v4i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pcmpeqd %xmm1, %xmm0
|
|
; SSE-NEXT: movmskps %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp_v4i32_v4i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vmovmskps %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v4i32_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512F-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v4i32_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v4i32_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <4 x i32> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v8i16_v8i1(<8 x i16>, <8 x i16>) nounwind {
|
|
; SSE-LABEL: icmp_v8i16_v8i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pcmpeqw %xmm1, %xmm0
|
|
; SSE-NEXT: packsswb %xmm0, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: testb %al, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp_v8i16_v8i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1OR2-NEXT: testb %al, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v8i16_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v8i16_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqw %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v8i16_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqw %xmm1, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i16> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v16i8_v16i1(<16 x i8>, <16 x i8>) nounwind {
|
|
; SSE-LABEL: icmp_v16i8_v16i1:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
|
|
; SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; SSE-NEXT: xorb %ah, %al
|
|
; SSE-NEXT: setnp %al
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1OR2-LABEL: icmp_v16i8_v16i1:
|
|
; AVX1OR2: # %bb.0:
|
|
; AVX1OR2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
|
|
; AVX1OR2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1OR2-NEXT: xorb %ah, %al
|
|
; AVX1OR2-NEXT: setnp %al
|
|
; AVX1OR2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v16i8_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX512F-NEXT: xorb %ah, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v16i8_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqb %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v16i8_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqb %xmm1, %xmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i8> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v4i64_v4i1(<4 x i64>, <4 x i64>) nounwind {
|
|
; X86-SSE2-LABEL: icmp_v4i64_v4i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpeqd 8(%ebp), %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,3],xmm1[1,3]
|
|
; X86-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; X86-SSE2-NEXT: andps %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movmskps %xmm0, %eax
|
|
; X86-SSE2-NEXT: testb %al, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: icmp_v4i64_v4i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,3],xmm1[1,3]
|
|
; X64-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
|
|
; X64-SSE2-NEXT: andps %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: movmskps %xmm0, %eax
|
|
; X64-SSE2-NEXT: testb %al, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: icmp_v4i64_v4i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: pcmpeqq %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pcmpeqq 8(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movmskps %xmm0, %eax
|
|
; X86-SSE4-NEXT: testb %al, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: icmp_v4i64_v4i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movmskps %xmm0, %eax
|
|
; X64-SSE4-NEXT: testb %al, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp_v4i64_v4i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpcmpeqq %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp_v4i64_v4i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpcmpeqq %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskpd %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v4i64_v4i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512F-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb $15, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v4i64_v4i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb $15, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v4i64_v4i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqq %ymm1, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <4 x i64> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v4i1(<4 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v8i32_v8i1(<8 x i32>, <8 x i32>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v8i32_v8i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqd 8(%ebp), %xmm1
|
|
; X86-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE-NEXT: packsswb %xmm0, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: testb %al, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v8i32_v8i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqd %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X64-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE-NEXT: packsswb %xmm0, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: testb %al, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp_v8i32_v8i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpcmpeqd %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
|
|
; AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX1-NEXT: testb %al, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp_v8i32_v8i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; AVX2-NEXT: testb %al, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v8i32_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512F-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v8i32_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v8i32_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqd %ymm1, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i32> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v16i16_v16i1(<16 x i16>, <16 x i16>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v16i16_v16i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pcmpeqw %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqw 8(%ebp), %xmm1
|
|
; X86-SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v16i16_v16i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqw %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pcmpeqw %xmm2, %xmm0
|
|
; X64-SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp_v16i16_v16i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpcmpeqw %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpacksswb %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp_v16i16_v16i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX2-NEXT: xorb %ah, %al
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v16i16_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v16i16_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqw %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v16i16_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqw %ymm1, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i16> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v32i8_v32i1(<32 x i8>, <32 x i8>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v32i8_v32i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: pcmpeqb %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqb 8(%ebp), %xmm1
|
|
; X86-SSE-NEXT: pxor %xmm0, %xmm1
|
|
; X86-SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v32i8_v32i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqb %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pcmpeqb %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm1, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: icmp_v32i8_v32i1:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpcmpeqb %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; AVX1-NEXT: xorb %ah, %al
|
|
; AVX1-NEXT: setnp %al
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: icmp_v32i8_v32i1:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; AVX2-NEXT: movl %eax, %ecx
|
|
; AVX2-NEXT: shrl $16, %ecx
|
|
; AVX2-NEXT: xorl %eax, %ecx
|
|
; AVX2-NEXT: xorb %ch, %cl
|
|
; AVX2-NEXT: setnp %al
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512F-LABEL: icmp_v32i8_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v32i8_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vpcmpeqb %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v32i8_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqb %ymm1, %ymm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <32 x i8> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v8i64_v8i1(<8 x i64>, <8 x i64>) nounwind {
|
|
; X86-SSE2-LABEL: icmp_v8i64_v8i1:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %ebp
|
|
; X86-SSE2-NEXT: movl %esp, %ebp
|
|
; X86-SSE2-NEXT: andl $-16, %esp
|
|
; X86-SSE2-NEXT: subl $16, %esp
|
|
; X86-SSE2-NEXT: movdqa 8(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: pcmpeqd 72(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: pcmpeqd 56(%ebp), %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: packssdw %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pcmpeqd 40(%ebp), %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pcmpeqd 24(%ebp), %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: packssdw %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: packssdw %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: packsswb %xmm1, %xmm1
|
|
; X86-SSE2-NEXT: pmovmskb %xmm1, %eax
|
|
; X86-SSE2-NEXT: testb %al, %al
|
|
; X86-SSE2-NEXT: setnp %al
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: icmp_v8i64_v8i1:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm7, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm3[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm7
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm6, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: packssdw %xmm7, %xmm3
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm5, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: packssdw %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: packssdw %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: packsswb %xmm1, %xmm1
|
|
; X64-SSE2-NEXT: pmovmskb %xmm1, %eax
|
|
; X64-SSE2-NEXT: testb %al, %al
|
|
; X64-SSE2-NEXT: setnp %al
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: icmp_v8i64_v8i1:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %ebp
|
|
; X86-SSE4-NEXT: movl %esp, %ebp
|
|
; X86-SSE4-NEXT: andl $-16, %esp
|
|
; X86-SSE4-NEXT: subl $16, %esp
|
|
; X86-SSE4-NEXT: movdqa 8(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: pcmpeqq 72(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: pcmpeqq 56(%ebp), %xmm2
|
|
; X86-SSE4-NEXT: packssdw %xmm3, %xmm2
|
|
; X86-SSE4-NEXT: pcmpeqq 40(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pcmpeqq 24(%ebp), %xmm0
|
|
; X86-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: packssdw %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE4-NEXT: testb %al, %al
|
|
; X86-SSE4-NEXT: setnp %al
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: icmp_v8i64_v8i1:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm7, %xmm3
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm6, %xmm2
|
|
; X64-SSE4-NEXT: packssdw %xmm3, %xmm2
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm5, %xmm1
|
|
; X64-SSE4-NEXT: pcmpeqq %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: packssdw %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: packsswb %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE4-NEXT: testb %al, %al
|
|
; X64-SSE4-NEXT: setnp %al
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: icmp_v8i64_v8i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: pushl %ebp
|
|
; X86-AVX1-NEXT: movl %esp, %ebp
|
|
; X86-AVX1-NEXT: andl $-32, %esp
|
|
; X86-AVX1-NEXT: subl $32, %esp
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
|
|
; X86-AVX1-NEXT: vpcmpeqq %xmm3, %xmm4, %xmm3
|
|
; X86-AVX1-NEXT: vpcmpeqq %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpackssdw %xmm3, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X86-AVX1-NEXT: vpcmpeqq 24(%ebp), %xmm2, %xmm2
|
|
; X86-AVX1-NEXT: vpcmpeqq 8(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; X86-AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; X86-AVX1-NEXT: testb %al, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: icmp_v8i64_v8i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpcmpeqq %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqq %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpackssdw %xmm4, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqq %xmm3, %xmm4, %xmm3
|
|
; X64-AVX1-NEXT: vpcmpeqq %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpackssdw %xmm3, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
|
|
; X64-AVX1-NEXT: vmovmskps %ymm0, %eax
|
|
; X64-AVX1-NEXT: testb %al, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: icmp_v8i64_v8i1:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: pushl %ebp
|
|
; X86-AVX2-NEXT: movl %esp, %ebp
|
|
; X86-AVX2-NEXT: andl $-32, %esp
|
|
; X86-AVX2-NEXT: subl $32, %esp
|
|
; X86-AVX2-NEXT: vpcmpeqq %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpcmpeqq 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; X86-AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; X86-AVX2-NEXT: testb %al, %al
|
|
; X86-AVX2-NEXT: setnp %al
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: icmp_v8i64_v8i1:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpeqq %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpeqq %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; X64-AVX2-NEXT: vmovmskps %ymm0, %eax
|
|
; X64-AVX2-NEXT: testb %al, %al
|
|
; X64-AVX2-NEXT: setnp %al
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: icmp_v8i64_v8i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: testb %al, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v8i64_v8i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: testb %al, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v8i64_v8i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqq %zmm1, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: testb %al, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <8 x i64> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v8i1(<8 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v16i32_v16i1(<16 x i32>, <16 x i32>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v16i32_v16i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: movdqa 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqd 72(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqd 56(%ebp), %xmm2
|
|
; X86-SSE-NEXT: packssdw %xmm3, %xmm2
|
|
; X86-SSE-NEXT: pcmpeqd 40(%ebp), %xmm1
|
|
; X86-SSE-NEXT: pcmpeqd 24(%ebp), %xmm0
|
|
; X86-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X86-SSE-NEXT: packsswb %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v16i32_v16i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqd %xmm7, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqd %xmm6, %xmm2
|
|
; X64-SSE-NEXT: packssdw %xmm3, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqd %xmm5, %xmm1
|
|
; X64-SSE-NEXT: pcmpeqd %xmm4, %xmm0
|
|
; X64-SSE-NEXT: packssdw %xmm1, %xmm0
|
|
; X64-SSE-NEXT: packsswb %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: icmp_v16i32_v16i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: pushl %ebp
|
|
; X86-AVX1-NEXT: movl %esp, %ebp
|
|
; X86-AVX1-NEXT: andl $-32, %esp
|
|
; X86-AVX1-NEXT: subl $32, %esp
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
|
|
; X86-AVX1-NEXT: vpcmpeqd %xmm3, %xmm4, %xmm3
|
|
; X86-AVX1-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpackssdw %xmm3, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X86-AVX1-NEXT: vpcmpeqd 24(%ebp), %xmm2, %xmm2
|
|
; X86-AVX1-NEXT: vpcmpeqd 8(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb %ah, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: icmp_v16i32_v16i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpcmpeqd %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqd %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpackssdw %xmm4, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqd %xmm3, %xmm4, %xmm3
|
|
; X64-AVX1-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpackssdw %xmm3, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb %ah, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: icmp_v16i32_v16i1:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: pushl %ebp
|
|
; X86-AVX2-NEXT: movl %esp, %ebp
|
|
; X86-AVX2-NEXT: andl $-32, %esp
|
|
; X86-AVX2-NEXT: subl $32, %esp
|
|
; X86-AVX2-NEXT: vpcmpeqd %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpcmpeqd 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
|
|
; X86-AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorb %ah, %al
|
|
; X86-AVX2-NEXT: setnp %al
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: icmp_v16i32_v16i1:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpeqd %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpeqd %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
|
|
; X64-AVX2-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorb %ah, %al
|
|
; X64-AVX2-NEXT: setnp %al
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: icmp_v16i32_v16i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: movl %eax, %ecx
|
|
; AVX512F-NEXT: shrl $8, %eax
|
|
; AVX512F-NEXT: xorb %cl, %al
|
|
; AVX512F-NEXT: setnp %al
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v16i32_v16i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $8, %eax
|
|
; AVX512BW-NEXT: xorb %cl, %al
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v16i32_v16i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqd %zmm1, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $8, %eax
|
|
; AVX512VL-NEXT: xorb %cl, %al
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <16 x i32> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v32i16_v32i1(<32 x i16>, <32 x i16>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v32i16_v32i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: movdqa 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqw 56(%ebp), %xmm2
|
|
; X86-SSE-NEXT: pcmpeqw 24(%ebp), %xmm0
|
|
; X86-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqw 72(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqw 40(%ebp), %xmm1
|
|
; X86-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X86-SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; X86-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v32i16_v32i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqw %xmm6, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqw %xmm4, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pcmpeqw %xmm7, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqw %xmm5, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X64-SSE-NEXT: packsswb %xmm1, %xmm0
|
|
; X64-SSE-NEXT: pmovmskb %xmm0, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: icmp_v32i16_v32i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: pushl %ebp
|
|
; X86-AVX1-NEXT: movl %esp, %ebp
|
|
; X86-AVX1-NEXT: andl $-32, %esp
|
|
; X86-AVX1-NEXT: subl $32, %esp
|
|
; X86-AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpeqw 8(%ebp), %xmm1, %xmm2
|
|
; X86-AVX1-NEXT: vpxor %xmm2, %xmm3, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; X86-AVX1-NEXT: vpcmpeqw 24(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpacksswb %xmm0, %xmm2, %xmm0
|
|
; X86-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb %ah, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: icmp_v32i16_v32i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vpcmpeqw %xmm3, %xmm1, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm5
|
|
; X64-AVX1-NEXT: vpxor %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm3
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; X64-AVX1-NEXT: vpcmpeqw %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm2
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; X64-AVX1-NEXT: vpcmpeqw %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm0
|
|
; X64-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb %ah, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: icmp_v32i16_v32i1:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: pushl %ebp
|
|
; X86-AVX2-NEXT: movl %esp, %ebp
|
|
; X86-AVX2-NEXT: andl $-32, %esp
|
|
; X86-AVX2-NEXT: subl $32, %esp
|
|
; X86-AVX2-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpcmpeqw 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpacksswb %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; X86-AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; X86-AVX2-NEXT: movl %eax, %ecx
|
|
; X86-AVX2-NEXT: shrl $16, %ecx
|
|
; X86-AVX2-NEXT: xorl %eax, %ecx
|
|
; X86-AVX2-NEXT: xorb %ch, %cl
|
|
; X86-AVX2-NEXT: setnp %al
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: icmp_v32i16_v32i1:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpeqw %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpeqw %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpacksswb %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
|
|
; X64-AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; X64-AVX2-NEXT: movl %eax, %ecx
|
|
; X64-AVX2-NEXT: shrl $16, %ecx
|
|
; X64-AVX2-NEXT: xorl %eax, %ecx
|
|
; X64-AVX2-NEXT: xorb %ch, %cl
|
|
; X64-AVX2-NEXT: setnp %al
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: icmp_v32i16_v32i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm2
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3
|
|
; AVX512F-NEXT: vpcmpeqw %ymm2, %ymm3, %ymm2
|
|
; AVX512F-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpxor %ymm2, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v32i16_v32i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpcmpeqw %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovd %k0, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v32i16_v32i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqw %zmm1, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovd %k0, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <32 x i16> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v32i1(<32 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
define i1 @icmp_v64i8_v64i1(<64 x i8>, <64 x i8>) nounwind {
|
|
; X86-SSE-LABEL: icmp_v64i8_v64i1:
|
|
; X86-SSE: # %bb.0:
|
|
; X86-SSE-NEXT: pushl %ebp
|
|
; X86-SSE-NEXT: movl %esp, %ebp
|
|
; X86-SSE-NEXT: andl $-16, %esp
|
|
; X86-SSE-NEXT: subl $16, %esp
|
|
; X86-SSE-NEXT: movdqa 8(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqb 56(%ebp), %xmm2
|
|
; X86-SSE-NEXT: pcmpeqb 24(%ebp), %xmm0
|
|
; X86-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE-NEXT: pcmpeqb 72(%ebp), %xmm3
|
|
; X86-SSE-NEXT: pcmpeqb 40(%ebp), %xmm1
|
|
; X86-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X86-SSE-NEXT: pxor %xmm0, %xmm1
|
|
; X86-SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; X86-SSE-NEXT: xorb %ah, %al
|
|
; X86-SSE-NEXT: setnp %al
|
|
; X86-SSE-NEXT: movl %ebp, %esp
|
|
; X86-SSE-NEXT: popl %ebp
|
|
; X86-SSE-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: icmp_v64i8_v64i1:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pcmpeqb %xmm6, %xmm2
|
|
; X64-SSE-NEXT: pcmpeqb %xmm4, %xmm0
|
|
; X64-SSE-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE-NEXT: pcmpeqb %xmm7, %xmm3
|
|
; X64-SSE-NEXT: pcmpeqb %xmm5, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm3, %xmm1
|
|
; X64-SSE-NEXT: pxor %xmm0, %xmm1
|
|
; X64-SSE-NEXT: pmovmskb %xmm1, %eax
|
|
; X64-SSE-NEXT: xorb %ah, %al
|
|
; X64-SSE-NEXT: setnp %al
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: icmp_v64i8_v64i1:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: pushl %ebp
|
|
; X86-AVX1-NEXT: movl %esp, %ebp
|
|
; X86-AVX1-NEXT: andl $-32, %esp
|
|
; X86-AVX1-NEXT: subl $32, %esp
|
|
; X86-AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpeqb 8(%ebp), %xmm1, %xmm2
|
|
; X86-AVX1-NEXT: vpxor %xmm2, %xmm3, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; X86-AVX1-NEXT: vpcmpeqb 24(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor %xmm0, %xmm2, %xmm0
|
|
; X86-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb %ah, %al
|
|
; X86-AVX1-NEXT: setnp %al
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: icmp_v64i8_v64i1:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vpcmpeqb %xmm3, %xmm1, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm5
|
|
; X64-AVX1-NEXT: vpxor %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm3
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
|
|
; X64-AVX1-NEXT: vpcmpeqb %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm2
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
|
|
; X64-AVX1-NEXT: vpcmpeqb %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor %xmm0, %xmm4, %xmm0
|
|
; X64-AVX1-NEXT: vpmovmskb %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb %ah, %al
|
|
; X64-AVX1-NEXT: setnp %al
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: icmp_v64i8_v64i1:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: pushl %ebp
|
|
; X86-AVX2-NEXT: movl %esp, %ebp
|
|
; X86-AVX2-NEXT: andl $-32, %esp
|
|
; X86-AVX2-NEXT: subl $32, %esp
|
|
; X86-AVX2-NEXT: vpcmpeqb %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpcmpeqb 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; X86-AVX2-NEXT: movl %eax, %ecx
|
|
; X86-AVX2-NEXT: shrl $16, %ecx
|
|
; X86-AVX2-NEXT: xorl %eax, %ecx
|
|
; X86-AVX2-NEXT: xorb %ch, %cl
|
|
; X86-AVX2-NEXT: setnp %al
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: icmp_v64i8_v64i1:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpeqb %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpeqb %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpxor %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpmovmskb %ymm0, %eax
|
|
; X64-AVX2-NEXT: movl %eax, %ecx
|
|
; X64-AVX2-NEXT: shrl $16, %ecx
|
|
; X64-AVX2-NEXT: xorl %eax, %ecx
|
|
; X64-AVX2-NEXT: xorb %ch, %cl
|
|
; X64-AVX2-NEXT: setnp %al
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: icmp_v64i8_v64i1:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm2
|
|
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3
|
|
; AVX512F-NEXT: vpcmpeqb %ymm2, %ymm3, %ymm2
|
|
; AVX512F-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm2, %xmm1
|
|
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm3, %xmm1
|
|
; AVX512F-NEXT: vpxor %xmm2, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpxor %xmm1, %xmm0, %xmm0
|
|
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
|
|
; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
|
|
; AVX512F-NEXT: kshiftrw $8, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $4, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $2, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kshiftrw $1, %k0, %k1
|
|
; AVX512F-NEXT: kxorw %k1, %k0, %k0
|
|
; AVX512F-NEXT: kmovw %k0, %eax
|
|
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512F-NEXT: vzeroupper
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: icmp_v64i8_v64i1:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpcmpeqb %zmm1, %zmm0, %k0
|
|
; AVX512BW-NEXT: kmovq %k0, %rax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrq $32, %rax
|
|
; AVX512BW-NEXT: xorl %ecx, %eax
|
|
; AVX512BW-NEXT: movl %eax, %ecx
|
|
; AVX512BW-NEXT: shrl $16, %ecx
|
|
; AVX512BW-NEXT: xorl %eax, %ecx
|
|
; AVX512BW-NEXT: xorb %ch, %cl
|
|
; AVX512BW-NEXT: setnp %al
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: icmp_v64i8_v64i1:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpcmpeqb %zmm1, %zmm0, %k0
|
|
; AVX512VL-NEXT: kmovq %k0, %rax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrq $32, %rax
|
|
; AVX512VL-NEXT: xorl %ecx, %eax
|
|
; AVX512VL-NEXT: movl %eax, %ecx
|
|
; AVX512VL-NEXT: shrl $16, %ecx
|
|
; AVX512VL-NEXT: xorl %eax, %ecx
|
|
; AVX512VL-NEXT: xorb %ch, %cl
|
|
; AVX512VL-NEXT: setnp %al
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%a = icmp eq <64 x i8> %0, %1
|
|
%b = call i1 @llvm.vector.reduce.xor.v64i1(<64 x i1> %a)
|
|
ret i1 %b
|
|
}
|
|
|
|
declare i1 @llvm.vector.reduce.xor.v2i1(<2 x i1>)
|
|
declare i1 @llvm.vector.reduce.xor.v4i1(<4 x i1>)
|
|
declare i1 @llvm.vector.reduce.xor.v8i1(<8 x i1>)
|
|
declare i1 @llvm.vector.reduce.xor.v16i1(<16 x i1>)
|
|
declare i1 @llvm.vector.reduce.xor.v32i1(<32 x i1>)
|
|
declare i1 @llvm.vector.reduce.xor.v64i1(<64 x i1>)
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; AVX512: {{.*}}
|