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
3921 lines
151 KiB
LLVM
3921 lines
151 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,SSE2,X86-SSE2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2,X64-SSE2
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE4,X86-SSE4,X86-SSE41
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE4,X64-SSE4,X64-SSE41
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE4,X86-SSE4,X86-SSE42
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE4,X64-SSE4,X64-SSE42
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1,X86-AVX,X86-AVX1
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1,X64-AVX,X64-AVX1
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,X86-AVX,X86-AVX2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,X64-AVX,X64-AVX2
|
|
; 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
|
|
|
|
;
|
|
; vXi64
|
|
;
|
|
|
|
define i64 @test_v2i64(<2 x i64> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v2i64:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,0,2147483648,0]
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm3
|
|
; X86-SSE2-NEXT: pxor %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: pxor %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm3
|
|
; X86-SSE2-NEXT: movd %xmm3, %eax
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %edx
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v2i64:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movq %xmm0, %rcx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movq %xmm0, %rax
|
|
; X64-SSE2-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE2-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE41-LABEL: test_v2i64:
|
|
; X86-SSE41: # %bb.0:
|
|
; X86-SSE41-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = [2147483648,2147483648]
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm3
|
|
; X86-SSE41-NEXT: pxor %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
|
|
; X86-SSE41-NEXT: pxor %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm3, %xmm4
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm4
|
|
; X86-SSE41-NEXT: pmovsxdq %xmm4, %xmm5
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm5, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm2
|
|
; X86-SSE41-NEXT: movd %xmm2, %eax
|
|
; X86-SSE41-NEXT: pextrd $1, %xmm2, %edx
|
|
; X86-SSE41-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v2i64:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pextrq $1, %xmm0, %rax
|
|
; X64-SSE4-NEXT: movq %xmm0, %rcx
|
|
; X64-SSE4-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE4-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-SSE42-LABEL: test_v2i64:
|
|
; X86-SSE42: # %bb.0:
|
|
; X86-SSE42-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE42-NEXT: pshufd {{.*#+}} xmm2 = xmm0[2,3,2,3]
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm2
|
|
; X86-SSE42-NEXT: movd %xmm2, %eax
|
|
; X86-SSE42-NEXT: pextrd $1, %xmm2, %edx
|
|
; X86-SSE42-NEXT: retl
|
|
;
|
|
; X86-AVX-LABEL: test_v2i64:
|
|
; X86-AVX: # %bb.0:
|
|
; X86-AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX-NEXT: retl
|
|
;
|
|
; X64-AVX-LABEL: test_v2i64:
|
|
; X64-AVX: # %bb.0:
|
|
; X64-AVX-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v2i64:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vmovq %xmm0, %rax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v2i64:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovq %xmm0, %rax
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i64 @llvm.vector.reduce.smax.v2i64(<2 x i64> %a0)
|
|
ret i64 %1
|
|
}
|
|
|
|
define i64 @test_v4i64(<4 x i64> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v4i64:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,0,2147483648,0]
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: pxor %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm4
|
|
; X86-SSE2-NEXT: pxor %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm6, %xmm4
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
|
|
; X86-SSE2-NEXT: pxor %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm4
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %eax
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %edx
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v4i64:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pxor %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pxor %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm4
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm3, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm3
|
|
; X64-SSE2-NEXT: movq %xmm3, %rcx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
|
|
; X64-SSE2-NEXT: movq %xmm0, %rax
|
|
; X64-SSE2-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE2-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE41-LABEL: test_v4i64:
|
|
; X86-SSE41: # %bb.0:
|
|
; X86-SSE41-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pmovzxdq {{.*#+}} xmm3 = [2147483648,2147483648]
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm4
|
|
; X86-SSE41-NEXT: pxor %xmm3, %xmm4
|
|
; X86-SSE41-NEXT: movdqa %xmm4, %xmm5
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm5
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm5[0,0,2,2]
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm4
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm6, %xmm4
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm4, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X86-SSE41-NEXT: movapd %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
|
|
; X86-SSE41-NEXT: pxor %xmm2, %xmm3
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm4
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X86-SSE41-NEXT: pmovsxdq %xmm4, %xmm5
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm5, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm2
|
|
; X86-SSE41-NEXT: movd %xmm2, %eax
|
|
; X86-SSE41-NEXT: pextrd $1, %xmm2, %edx
|
|
; X86-SSE41-NEXT: retl
|
|
;
|
|
; X64-SSE41-LABEL: test_v4i64:
|
|
; X64-SSE41: # %bb.0:
|
|
; X64-SSE41-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = [2147483648,2147483648]
|
|
; X64-SSE41-NEXT: movdqa %xmm1, %xmm3
|
|
; X64-SSE41-NEXT: pxor %xmm0, %xmm3
|
|
; X64-SSE41-NEXT: pxor %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm0, %xmm4
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm5, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X64-SSE41-NEXT: pextrq $1, %xmm1, %rax
|
|
; X64-SSE41-NEXT: movq %xmm1, %rcx
|
|
; X64-SSE41-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE41-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE41-NEXT: retq
|
|
;
|
|
; X86-SSE42-LABEL: test_v4i64:
|
|
; X86-SSE42: # %bb.0:
|
|
; X86-SSE42-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X86-SSE42-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
|
|
; X86-SSE42-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm2
|
|
; X86-SSE42-NEXT: movd %xmm2, %eax
|
|
; X86-SSE42-NEXT: pextrd $1, %xmm2, %edx
|
|
; X86-SSE42-NEXT: retl
|
|
;
|
|
; X64-SSE42-LABEL: test_v4i64:
|
|
; X64-SSE42: # %bb.0:
|
|
; X64-SSE42-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X64-SSE42-NEXT: pextrq $1, %xmm1, %rax
|
|
; X64-SSE42-NEXT: movq %xmm1, %rcx
|
|
; X64-SSE42-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE42-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE42-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v4i64:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v4i64:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX1-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX1-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX1-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX1-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v4i64:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v4i64:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX2-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX2-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX2-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v4i64:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vmovq %xmm0, %rax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v4i64:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovq %xmm0, %rax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i64 @llvm.vector.reduce.smax.v4i64(<4 x i64> %a0)
|
|
ret i64 %1
|
|
}
|
|
|
|
define i64 @test_v8i64(<8 x i64> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v8i64:
|
|
; 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), %xmm5
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,0,2147483648,0]
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm4
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm6
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm6
|
|
; X86-SSE2-NEXT: movdqa %xmm6, %xmm7
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm7
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm4, %xmm6
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm7[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm4, %xmm6
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm7[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm6, %xmm4
|
|
; X86-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm5, %xmm4
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm6
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm1, %xmm6
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm5, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm2[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm0[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm6
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm6, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: pxor %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %eax
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %edx
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v8i64:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [2147483648,2147483648]
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm5
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm6
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm6
|
|
; X64-SSE2-NEXT: movdqa %xmm6, %xmm7
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm5, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm5, %xmm6
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm8, %xmm6
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm6, %xmm5
|
|
; X64-SSE2-NEXT: pand %xmm5, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm5
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm5, %xmm1
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm6
|
|
; X64-SSE2-NEXT: pxor %xmm4, %xmm6
|
|
; X64-SSE2-NEXT: movdqa %xmm6, %xmm7
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm3, %xmm6
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm6[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm8, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm7[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm6
|
|
; X64-SSE2-NEXT: pand %xmm6, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm2, %xmm6
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm6
|
|
; X64-SSE2-NEXT: pxor %xmm6, %xmm4
|
|
; X64-SSE2-NEXT: movdqa %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm6
|
|
; X64-SSE2-NEXT: pandn %xmm5, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm6, %xmm0
|
|
; X64-SSE2-NEXT: movq %xmm0, %rcx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movq %xmm0, %rax
|
|
; X64-SSE2-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE2-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE41-LABEL: test_v8i64:
|
|
; X86-SSE41: # %bb.0:
|
|
; X86-SSE41-NEXT: pushl %ebp
|
|
; X86-SSE41-NEXT: movl %esp, %ebp
|
|
; X86-SSE41-NEXT: andl $-16, %esp
|
|
; X86-SSE41-NEXT: subl $16, %esp
|
|
; X86-SSE41-NEXT: movdqa %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: movdqa 8(%ebp), %xmm4
|
|
; X86-SSE41-NEXT: pmovzxdq {{.*#+}} xmm5 = [2147483648,2147483648]
|
|
; X86-SSE41-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm6
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm6
|
|
; X86-SSE41-NEXT: movdqa %xmm6, %xmm7
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm7
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm4
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm3, %xmm1
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm1
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm6
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm6
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm6[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm6[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm4, %xmm1
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm1
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm2
|
|
; X86-SSE41-NEXT: movapd %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm1, %xmm3
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm3[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm0[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm1, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm4
|
|
; X86-SSE41-NEXT: movapd %xmm4, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm4[2,3,2,3]
|
|
; X86-SSE41-NEXT: pxor %xmm1, %xmm5
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm5
|
|
; X86-SSE41-NEXT: pmovsxdq %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm4, %xmm1
|
|
; X86-SSE41-NEXT: movd %xmm1, %eax
|
|
; X86-SSE41-NEXT: pextrd $1, %xmm1, %edx
|
|
; X86-SSE41-NEXT: movl %ebp, %esp
|
|
; X86-SSE41-NEXT: popl %ebp
|
|
; X86-SSE41-NEXT: retl
|
|
;
|
|
; X64-SSE41-LABEL: test_v8i64:
|
|
; X64-SSE41: # %bb.0:
|
|
; X64-SSE41-NEXT: movdqa %xmm0, %xmm4
|
|
; X64-SSE41-NEXT: pmovzxdq {{.*#+}} xmm5 = [2147483648,2147483648]
|
|
; X64-SSE41-NEXT: movdqa %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm1, %xmm6
|
|
; X64-SSE41-NEXT: pxor %xmm5, %xmm6
|
|
; X64-SSE41-NEXT: movdqa %xmm6, %xmm7
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm7
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm6
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm8, %xmm6
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm3
|
|
; X64-SSE41-NEXT: movapd %xmm3, %xmm1
|
|
; X64-SSE41-NEXT: xorpd %xmm5, %xmm1
|
|
; X64-SSE41-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm4, %xmm6
|
|
; X64-SSE41-NEXT: pxor %xmm5, %xmm6
|
|
; X64-SSE41-NEXT: movdqa %xmm6, %xmm7
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm7
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm8 = xmm7[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm6
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm8, %xmm6
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm4, %xmm2
|
|
; X64-SSE41-NEXT: xorpd %xmm2, %xmm5
|
|
; X64-SSE41-NEXT: movapd %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm1, %xmm0
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm0[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm1, %xmm5
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm1, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm3
|
|
; X64-SSE41-NEXT: pextrq $1, %xmm3, %rax
|
|
; X64-SSE41-NEXT: movq %xmm3, %rcx
|
|
; X64-SSE41-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE41-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE41-NEXT: retq
|
|
;
|
|
; X86-SSE42-LABEL: test_v8i64:
|
|
; X86-SSE42: # %bb.0:
|
|
; X86-SSE42-NEXT: pushl %ebp
|
|
; X86-SSE42-NEXT: movl %esp, %ebp
|
|
; X86-SSE42-NEXT: andl $-16, %esp
|
|
; X86-SSE42-NEXT: subl $16, %esp
|
|
; X86-SSE42-NEXT: movdqa %xmm0, %xmm3
|
|
; X86-SSE42-NEXT: movdqa 8(%ebp), %xmm4
|
|
; X86-SSE42-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm4
|
|
; X86-SSE42-NEXT: movdqa %xmm3, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm3, %xmm2
|
|
; X86-SSE42-NEXT: movapd %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm4
|
|
; X86-SSE42-NEXT: pshufd {{.*#+}} xmm1 = xmm4[2,3,2,3]
|
|
; X86-SSE42-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm4, %xmm1
|
|
; X86-SSE42-NEXT: movd %xmm1, %eax
|
|
; X86-SSE42-NEXT: pextrd $1, %xmm1, %edx
|
|
; X86-SSE42-NEXT: movl %ebp, %esp
|
|
; X86-SSE42-NEXT: popl %ebp
|
|
; X86-SSE42-NEXT: retl
|
|
;
|
|
; X64-SSE42-LABEL: test_v8i64:
|
|
; X64-SSE42: # %bb.0:
|
|
; X64-SSE42-NEXT: movdqa %xmm0, %xmm4
|
|
; X64-SSE42-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm3, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm3
|
|
; X64-SSE42-NEXT: movdqa %xmm4, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm4, %xmm2
|
|
; X64-SSE42-NEXT: movapd %xmm2, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm3, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm3
|
|
; X64-SSE42-NEXT: pextrq $1, %xmm3, %rax
|
|
; X64-SSE42-NEXT: movq %xmm3, %rcx
|
|
; X64-SSE42-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE42-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE42-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v8i64:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm4
|
|
; X86-AVX1-NEXT: vblendvpd %xmm4, %xmm3, %xmm2, %xmm2
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm3
|
|
; X86-AVX1-NEXT: vblendvpd %xmm3, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm2, %xmm0, %xmm1
|
|
; X86-AVX1-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0
|
|
; X86-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v8i64:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm4
|
|
; X64-AVX1-NEXT: vblendvpd %xmm4, %xmm3, %xmm2, %xmm2
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm3
|
|
; X64-AVX1-NEXT: vblendvpd %xmm3, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm2, %xmm0, %xmm1
|
|
; X64-AVX1-NEXT: vblendvpd %xmm1, %xmm0, %xmm2, %xmm0
|
|
; X64-AVX1-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX1-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX1-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX1-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v8i64:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm2
|
|
; X86-AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
|
|
; X86-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v8i64:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm2
|
|
; X64-AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
|
|
; X64-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX2-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX2-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX2-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v8i64:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vmovq %xmm0, %rax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v8i64:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsq %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovq %xmm0, %rax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i64 @llvm.vector.reduce.smax.v8i64(<8 x i64> %a0)
|
|
ret i64 %1
|
|
}
|
|
|
|
define i64 @test_v16i64(<16 x i64> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v16i64:
|
|
; 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 $32, %esp
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm7
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: movaps %xmm0, (%esp) # 16-byte Spill
|
|
; X86-SSE2-NEXT: movdqa 8(%ebp), %xmm5
|
|
; X86-SSE2-NEXT: movdqa 72(%ebp), %xmm1
|
|
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2147483648,0,2147483648,0]
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm4
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm6
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm6
|
|
; X86-SSE2-NEXT: movdqa %xmm6, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm0
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm4, %xmm6
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[0,0,2,2]
|
|
; X86-SSE2-NEXT: pand %xmm4, %xmm6
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm6, %xmm4
|
|
; X86-SSE2-NEXT: movdqa 40(%ebp), %xmm6
|
|
; X86-SSE2-NEXT: pand %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm4
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm6, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm5
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm1[0,0,2,2]
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm6, %xmm5
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm1[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm6
|
|
; X86-SSE2-NEXT: movdqa 56(%ebp), %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm6, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm7, %xmm4
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm4, %xmm6
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm2, %xmm6
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm6[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movdqa 24(%ebp), %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm4
|
|
; X86-SSE2-NEXT: pand %xmm4, %xmm7
|
|
; X86-SSE2-NEXT: pandn %xmm5, %xmm4
|
|
; X86-SSE2-NEXT: por %xmm7, %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: movdqa (%esp), %xmm5 # 16-byte Reload
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm6
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm6
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm6[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm7
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm7, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm6
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm6
|
|
; X86-SSE2-NEXT: movdqa 24(%ebp), %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm6, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm5
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm6
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm6
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm6[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm7
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm6[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm7, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm2
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm4
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pxor %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: pxor %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm4
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X86-SSE2-NEXT: pcmpeqd %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,0,2,2]
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
|
|
; X86-SSE2-NEXT: por %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %eax
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %edx
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v16i64:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm8 = [2147483648,2147483648]
|
|
; X64-SSE2-NEXT: movdqa %xmm7, %xmm9
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm9
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm10
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm10
|
|
; X64-SSE2-NEXT: movdqa %xmm10, %xmm11
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm9, %xmm11
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm9, %xmm10
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm12, %xmm10
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm11[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm10, %xmm9
|
|
; X64-SSE2-NEXT: pand %xmm9, %xmm3
|
|
; X64-SSE2-NEXT: pandn %xmm7, %xmm9
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm9
|
|
; X64-SSE2-NEXT: movdqa %xmm9, %xmm3
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm5, %xmm7
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm7
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm10
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm10
|
|
; X64-SSE2-NEXT: movdqa %xmm10, %xmm11
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm7, %xmm11
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm7, %xmm10
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm10[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm12, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm11[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm7, %xmm10
|
|
; X64-SSE2-NEXT: pand %xmm10, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm5, %xmm10
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm10
|
|
; X64-SSE2-NEXT: movdqa %xmm10, %xmm1
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm5
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm5
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm5[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm7, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm10
|
|
; X64-SSE2-NEXT: pandn %xmm9, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm10, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm6, %xmm5
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm7
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm7
|
|
; X64-SSE2-NEXT: movdqa %xmm7, %xmm9
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm5, %xmm9
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm9[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm5, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm7[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm10, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm9[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm7, %xmm5
|
|
; X64-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm6, %xmm5
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm4, %xmm6
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm6
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm7
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm7
|
|
; X64-SSE2-NEXT: movdqa %xmm7, %xmm9
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm6, %xmm9
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm9[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm6, %xmm7
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm7[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm10, %xmm6
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm9[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm6, %xmm7
|
|
; X64-SSE2-NEXT: pand %xmm7, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm4, %xmm7
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm7
|
|
; X64-SSE2-NEXT: movdqa %xmm7, %xmm0
|
|
; X64-SSE2-NEXT: pxor %xmm8, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm4
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm2, %xmm4
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm4[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm6, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm7
|
|
; X64-SSE2-NEXT: pandn %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm7, %xmm2
|
|
; X64-SSE2-NEXT: pxor %xmm2, %xmm8
|
|
; X64-SSE2-NEXT: movdqa %xmm8, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm3, %xmm8
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: movq %xmm0, %rcx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movq %xmm0, %rax
|
|
; X64-SSE2-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE2-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE41-LABEL: test_v16i64:
|
|
; X86-SSE41: # %bb.0:
|
|
; X86-SSE41-NEXT: pushl %ebp
|
|
; X86-SSE41-NEXT: movl %esp, %ebp
|
|
; X86-SSE41-NEXT: andl $-16, %esp
|
|
; X86-SSE41-NEXT: subl $32, %esp
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm3
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm2
|
|
; X86-SSE41-NEXT: movaps %xmm0, (%esp) # 16-byte Spill
|
|
; X86-SSE41-NEXT: movdqa 8(%ebp), %xmm7
|
|
; X86-SSE41-NEXT: movdqa 72(%ebp), %xmm4
|
|
; X86-SSE41-NEXT: pmovzxdq {{.*#+}} xmm5 = [2147483648,2147483648]
|
|
; X86-SSE41-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm7, %xmm6
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm6
|
|
; X86-SSE41-NEXT: movdqa %xmm6, %xmm1
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm6[1,1,3,3]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[0,0,2,2]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: movdqa 40(%ebp), %xmm6
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm7, %xmm4
|
|
; X86-SSE41-NEXT: movdqa %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm1
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm1
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm7
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm7
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[0,0,2,2]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm4, %xmm1
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm1
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm6
|
|
; X86-SSE41-NEXT: movapd %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm1, %xmm7
|
|
; X86-SSE41-NEXT: movdqa 56(%ebp), %xmm1
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm7, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm6, %xmm4
|
|
; X86-SSE41-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa %xmm3, %xmm2
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm7
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm7
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: movdqa 24(%ebp), %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm7[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm1
|
|
; X86-SSE41-NEXT: movdqa %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movdqa (%esp), %xmm7 # 16-byte Reload
|
|
; X86-SSE41-NEXT: movdqa %xmm7, %xmm2
|
|
; X86-SSE41-NEXT: pxor %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: movdqa %xmm2, %xmm3
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm1, %xmm2
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm7, %xmm6
|
|
; X86-SSE41-NEXT: movapd %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm2, %xmm3
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm7 = xmm0[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm2, %xmm7
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm7, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm4, %xmm2
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm6, %xmm1
|
|
; X86-SSE41-NEXT: movapd %xmm1, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm2, %xmm3
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm6 = xmm0[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm2, %xmm6
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm6, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm4
|
|
; X86-SSE41-NEXT: movapd %xmm4, %xmm0
|
|
; X86-SSE41-NEXT: xorpd %xmm5, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm4[2,3,2,3]
|
|
; X86-SSE41-NEXT: pxor %xmm1, %xmm5
|
|
; X86-SSE41-NEXT: movapd %xmm0, %xmm2
|
|
; X86-SSE41-NEXT: pcmpgtd %xmm5, %xmm2
|
|
; X86-SSE41-NEXT: pcmpeqd %xmm0, %xmm5
|
|
; X86-SSE41-NEXT: pmovsxdq %xmm2, %xmm0
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3]
|
|
; X86-SSE41-NEXT: pand %xmm0, %xmm3
|
|
; X86-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
|
|
; X86-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X86-SSE41-NEXT: blendvpd %xmm0, %xmm4, %xmm1
|
|
; X86-SSE41-NEXT: movd %xmm1, %eax
|
|
; X86-SSE41-NEXT: pextrd $1, %xmm1, %edx
|
|
; X86-SSE41-NEXT: movl %ebp, %esp
|
|
; X86-SSE41-NEXT: popl %ebp
|
|
; X86-SSE41-NEXT: retl
|
|
;
|
|
; X64-SSE41-LABEL: test_v16i64:
|
|
; X64-SSE41: # %bb.0:
|
|
; X64-SSE41-NEXT: movdqa %xmm0, %xmm8
|
|
; X64-SSE41-NEXT: pmovzxdq {{.*#+}} xmm9 = [2147483648,2147483648]
|
|
; X64-SSE41-NEXT: movdqa %xmm7, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm3, %xmm10
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm10
|
|
; X64-SSE41-NEXT: movdqa %xmm10, %xmm11
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm11
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm10
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm12, %xmm10
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm11[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm10, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm7
|
|
; X64-SSE41-NEXT: movapd %xmm7, %xmm3
|
|
; X64-SSE41-NEXT: xorpd %xmm9, %xmm3
|
|
; X64-SSE41-NEXT: movdqa %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm1, %xmm10
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm10
|
|
; X64-SSE41-NEXT: movdqa %xmm10, %xmm11
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm11
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm10
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm12, %xmm10
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm11[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm10, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm1, %xmm5
|
|
; X64-SSE41-NEXT: movapd %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: xorpd %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movapd %xmm0, %xmm1
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm3, %xmm1
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm1[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm10, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm5, %xmm7
|
|
; X64-SSE41-NEXT: movapd %xmm7, %xmm1
|
|
; X64-SSE41-NEXT: xorpd %xmm9, %xmm1
|
|
; X64-SSE41-NEXT: movdqa %xmm6, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm2, %xmm3
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm3
|
|
; X64-SSE41-NEXT: movdqa %xmm3, %xmm5
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm5
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm5[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm10, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm6
|
|
; X64-SSE41-NEXT: movapd %xmm6, %xmm2
|
|
; X64-SSE41-NEXT: xorpd %xmm9, %xmm2
|
|
; X64-SSE41-NEXT: movdqa %xmm4, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm8, %xmm3
|
|
; X64-SSE41-NEXT: pxor %xmm9, %xmm3
|
|
; X64-SSE41-NEXT: movdqa %xmm3, %xmm5
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm5
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm10 = xmm5[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm10, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm5[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm3, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm8, %xmm4
|
|
; X64-SSE41-NEXT: movapd %xmm4, %xmm0
|
|
; X64-SSE41-NEXT: xorpd %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: movapd %xmm0, %xmm3
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm2, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm3[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm5, %xmm2
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm4, %xmm6
|
|
; X64-SSE41-NEXT: xorpd %xmm6, %xmm9
|
|
; X64-SSE41-NEXT: movapd %xmm9, %xmm0
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm1, %xmm0
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm1, %xmm9
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm9[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm1, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm6, %xmm7
|
|
; X64-SSE41-NEXT: pextrq $1, %xmm7, %rax
|
|
; X64-SSE41-NEXT: movq %xmm7, %rcx
|
|
; X64-SSE41-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE41-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE41-NEXT: retq
|
|
;
|
|
; X86-SSE42-LABEL: test_v16i64:
|
|
; X86-SSE42: # %bb.0:
|
|
; X86-SSE42-NEXT: pushl %ebp
|
|
; X86-SSE42-NEXT: movl %esp, %ebp
|
|
; X86-SSE42-NEXT: andl $-16, %esp
|
|
; X86-SSE42-NEXT: subl $16, %esp
|
|
; X86-SSE42-NEXT: movdqa %xmm0, %xmm3
|
|
; X86-SSE42-NEXT: movdqa 72(%ebp), %xmm4
|
|
; X86-SSE42-NEXT: movdqa 8(%ebp), %xmm5
|
|
; X86-SSE42-NEXT: movdqa %xmm5, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm5, %xmm4
|
|
; X86-SSE42-NEXT: movdqa 40(%ebp), %xmm5
|
|
; X86-SSE42-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm5, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm5
|
|
; X86-SSE42-NEXT: movapd %xmm5, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm5, %xmm4
|
|
; X86-SSE42-NEXT: movdqa 56(%ebp), %xmm1
|
|
; X86-SSE42-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X86-SSE42-NEXT: movdqa 24(%ebp), %xmm2
|
|
; X86-SSE42-NEXT: movdqa %xmm3, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm3, %xmm2
|
|
; X86-SSE42-NEXT: movapd %xmm2, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X86-SSE42-NEXT: movapd %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm4
|
|
; X86-SSE42-NEXT: pshufd {{.*#+}} xmm1 = xmm4[2,3,2,3]
|
|
; X86-SSE42-NEXT: movdqa %xmm4, %xmm0
|
|
; X86-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X86-SSE42-NEXT: blendvpd %xmm0, %xmm4, %xmm1
|
|
; X86-SSE42-NEXT: movd %xmm1, %eax
|
|
; X86-SSE42-NEXT: pextrd $1, %xmm1, %edx
|
|
; X86-SSE42-NEXT: movl %ebp, %esp
|
|
; X86-SSE42-NEXT: popl %ebp
|
|
; X86-SSE42-NEXT: retl
|
|
;
|
|
; X64-SSE42-LABEL: test_v16i64:
|
|
; X64-SSE42: # %bb.0:
|
|
; X64-SSE42-NEXT: movdqa %xmm0, %xmm8
|
|
; X64-SSE42-NEXT: movdqa %xmm3, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm7, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm3, %xmm7
|
|
; X64-SSE42-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm5, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm1, %xmm5
|
|
; X64-SSE42-NEXT: movapd %xmm5, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm7, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm5, %xmm7
|
|
; X64-SSE42-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm6, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm6
|
|
; X64-SSE42-NEXT: movdqa %xmm8, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm4, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm8, %xmm4
|
|
; X64-SSE42-NEXT: movapd %xmm4, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm6, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm4, %xmm6
|
|
; X64-SSE42-NEXT: movapd %xmm6, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm7, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm6, %xmm7
|
|
; X64-SSE42-NEXT: pextrq $1, %xmm7, %rax
|
|
; X64-SSE42-NEXT: movq %xmm7, %rcx
|
|
; X64-SSE42-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE42-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE42-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v16i64:
|
|
; 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, %ymm1, %xmm3
|
|
; X86-AVX1-NEXT: vmovdqa 8(%ebp), %xmm4
|
|
; X86-AVX1-NEXT: vmovdqa 24(%ebp), %xmm5
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm5, %xmm3, %xmm6
|
|
; X86-AVX1-NEXT: vblendvpd %xmm6, %xmm3, %xmm5, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm5, %xmm6, %xmm7
|
|
; X86-AVX1-NEXT: vblendvpd %xmm7, %xmm6, %xmm5, %xmm5
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm3, %xmm5, %xmm6
|
|
; X86-AVX1-NEXT: vblendvpd %xmm6, %xmm5, %xmm3, %xmm3
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm4, %xmm1, %xmm5
|
|
; X86-AVX1-NEXT: vblendvpd %xmm5, %xmm1, %xmm4, %xmm1
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm2, %xmm0, %xmm4
|
|
; X86-AVX1-NEXT: vblendvpd %xmm4, %xmm0, %xmm2, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm3, %xmm0, %xmm1
|
|
; X86-AVX1-NEXT: vblendvpd %xmm1, %xmm0, %xmm3, %xmm0
|
|
; X86-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v16i64:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm4, %xmm5, %xmm6
|
|
; X64-AVX1-NEXT: vblendvpd %xmm6, %xmm5, %xmm4, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm5, %xmm6, %xmm7
|
|
; X64-AVX1-NEXT: vblendvpd %xmm7, %xmm6, %xmm5, %xmm5
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm4, %xmm5, %xmm6
|
|
; X64-AVX1-NEXT: vblendvpd %xmm6, %xmm5, %xmm4, %xmm4
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm3, %xmm1, %xmm5
|
|
; X64-AVX1-NEXT: vblendvpd %xmm5, %xmm1, %xmm3, %xmm1
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm2, %xmm0, %xmm3
|
|
; X64-AVX1-NEXT: vblendvpd %xmm3, %xmm0, %xmm2, %xmm0
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm4, %xmm0, %xmm1
|
|
; X64-AVX1-NEXT: vblendvpd %xmm1, %xmm0, %xmm4, %xmm0
|
|
; X64-AVX1-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX1-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX1-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX1-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v16i64:
|
|
; 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: vmovdqa 8(%ebp), %ymm3
|
|
; X86-AVX2-NEXT: vpcmpgtq %ymm3, %ymm1, %ymm4
|
|
; X86-AVX2-NEXT: vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
|
|
; X86-AVX2-NEXT: vpcmpgtq %ymm2, %ymm0, %ymm3
|
|
; X86-AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
|
|
; X86-AVX2-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm2
|
|
; X86-AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
|
|
; X86-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X86-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: vpextrd $1, %xmm0, %edx
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v16i64:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpcmpgtq %ymm3, %ymm1, %ymm4
|
|
; X64-AVX2-NEXT: vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %ymm2, %ymm0, %ymm3
|
|
; X64-AVX2-NEXT: vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
|
|
; X64-AVX2-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm2
|
|
; X64-AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
|
|
; X64-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX2-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX2-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX2-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v16i64:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vmovq %xmm0, %rax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v16i64:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsq %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovq %xmm0, %rax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i64 @llvm.vector.reduce.smax.v16i64(<16 x i64> %a0)
|
|
ret i64 %1
|
|
}
|
|
|
|
;
|
|
; vXi32
|
|
;
|
|
|
|
define i32 @test_v2i32(<2 x i32> %a0) nounwind {
|
|
; SSE2-LABEL: test_v2i32:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: movd %xmm0, %ecx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
|
|
; SSE2-NEXT: movd %xmm0, %eax
|
|
; SSE2-NEXT: cmpl %eax, %ecx
|
|
; SSE2-NEXT: cmovgl %ecx, %eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: test_v2i32:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; SSE4-NEXT: movd %xmm1, %eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v2i32:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i32 @llvm.vector.reduce.smax.v2i32(<2 x i32> %a0)
|
|
ret i32 %1
|
|
}
|
|
|
|
define i32 @test_v4i32(<4 x i32> %a0) nounwind {
|
|
; SSE2-LABEL: test_v4i32:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; SSE2-NEXT: pand %xmm2, %xmm0
|
|
; SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; SSE2-NEXT: por %xmm0, %xmm2
|
|
; SSE2-NEXT: movd %xmm2, %ecx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; SSE2-NEXT: movd %xmm0, %eax
|
|
; SSE2-NEXT: cmpl %eax, %ecx
|
|
; SSE2-NEXT: cmovgl %ecx, %eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: test_v4i32:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; SSE4-NEXT: movd %xmm0, %eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v4i32:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> %a0)
|
|
ret i32 %1
|
|
}
|
|
|
|
define i32 @test_v8i32(<8 x i32> %a0) nounwind {
|
|
; SSE2-LABEL: test_v8i32:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; SSE2-NEXT: pand %xmm2, %xmm0
|
|
; SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; SSE2-NEXT: por %xmm0, %xmm2
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
|
|
; SSE2-NEXT: pand %xmm1, %xmm2
|
|
; SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; SSE2-NEXT: por %xmm2, %xmm1
|
|
; SSE2-NEXT: movd %xmm1, %ecx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; SSE2-NEXT: movd %xmm0, %eax
|
|
; SSE2-NEXT: cmpl %eax, %ecx
|
|
; SSE2-NEXT: cmovgl %ecx, %eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: test_v8i32:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; SSE4-NEXT: movd %xmm0, %eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX1-LABEL: test_v8i32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovd %xmm0, %eax
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: test_v8i32:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vmovd %xmm0, %eax
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512-LABEL: test_v8i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovd %xmm0, %eax
|
|
; AVX512-NEXT: vzeroupper
|
|
; AVX512-NEXT: retq
|
|
%1 = call i32 @llvm.vector.reduce.smax.v8i32(<8 x i32> %a0)
|
|
ret i32 %1
|
|
}
|
|
|
|
define i32 @test_v16i32(<16 x i32> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v16i32:
|
|
; 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), %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v16i32:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X64-SSE2-NEXT: movd %xmm0, %eax
|
|
; X64-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v16i32:
|
|
; 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: pmaxsd %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsd 8(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
|
|
; X86-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X86-SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: movd %xmm1, %eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v16i32:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsd %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsd %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: test_v16i32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; AVX1-NEXT: vpmaxsd %xmm2, %xmm3, %xmm2
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpmaxsd %xmm2, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmovd %xmm0, %eax
|
|
; AVX1-NEXT: vzeroupper
|
|
; AVX1-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX2-LABEL: test_v16i32:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: vpmaxsd %ymm1, %ymm0, %ymm0
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX2-NEXT: vmovd %xmm0, %eax
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX512-LABEL: test_v16i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512-NEXT: vpmaxsd %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovd %xmm0, %eax
|
|
; AVX512-NEXT: vzeroupper
|
|
; AVX512-NEXT: retq
|
|
%1 = call i32 @llvm.vector.reduce.smax.v16i32(<16 x i32> %a0)
|
|
ret i32 %1
|
|
}
|
|
|
|
define i32 @test_v32i32(<32 x i32> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v32i32:
|
|
; 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 72(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: movdqa 8(%ebp), %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: movdqa 40(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: movdqa 56(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: movdqa 24(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm4, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtd %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: movd %xmm1, %ecx
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X86-SSE2-NEXT: movd %xmm0, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v32i32:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm8
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm7, %xmm8
|
|
; X64-SSE2-NEXT: pand %xmm8, %xmm3
|
|
; X64-SSE2-NEXT: pandn %xmm7, %xmm8
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm8
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm5, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm5, %xmm3
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm8, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pandn %xmm8, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm6, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm6, %xmm3
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %ecx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE2-NEXT: movd %xmm0, %eax
|
|
; X64-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v32i32:
|
|
; 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: pmaxsd 72(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: pmaxsd 40(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsd %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pmaxsd 56(%ebp), %xmm2
|
|
; X86-SSE4-NEXT: pmaxsd 24(%ebp), %xmm0
|
|
; X86-SSE4-NEXT: pmaxsd %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X86-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v32i32:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsd %xmm7, %xmm3
|
|
; X64-SSE4-NEXT: pmaxsd %xmm5, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsd %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsd %xmm6, %xmm2
|
|
; X64-SSE4-NEXT: pmaxsd %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsd %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE4-NEXT: pmaxsd %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE4-NEXT: pmaxsd %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v32i32:
|
|
; 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: vpmaxsd %xmm3, %xmm4, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsd 24(%ebp), %xmm4, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm4, %xmm3, %xmm3
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsd 8(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm3, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X86-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v32i32:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm5, %xmm6, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm4, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X64-AVX1-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v32i32:
|
|
; 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: vpmaxsd %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpmaxsd 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpmaxsd %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X86-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v32i32:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpmaxsd %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpmaxsd %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpmaxsd %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X64-AVX2-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512-LABEL: test_v32i32:
|
|
; AVX512: # %bb.0:
|
|
; AVX512-NEXT: vpmaxsd %zmm1, %zmm0, %zmm0
|
|
; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512-NEXT: vpmaxsd %ymm1, %ymm0, %ymm0
|
|
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX512-NEXT: vpmaxsd %xmm1, %xmm0, %xmm0
|
|
; AVX512-NEXT: vmovd %xmm0, %eax
|
|
; AVX512-NEXT: vzeroupper
|
|
; AVX512-NEXT: retq
|
|
%1 = call i32 @llvm.vector.reduce.smax.v32i32(<32 x i32> %a0)
|
|
ret i32 %1
|
|
}
|
|
|
|
;
|
|
; vXi16
|
|
;
|
|
|
|
define i16 @test_v2i16(<2 x i16> %a0) nounwind {
|
|
; SSE-LABEL: test_v2i16:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE-NEXT: psrld $16, %xmm1
|
|
; SSE-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE-NEXT: movd %xmm1, %eax
|
|
; SSE-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v2i16:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i16 @llvm.vector.reduce.smax.v2i16(<2 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
define i16 @test_v4i16(<4 x i16> %a0) nounwind {
|
|
; SSE-LABEL: test_v4i16:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; SSE-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE-NEXT: movdqa %xmm1, %xmm0
|
|
; SSE-NEXT: psrld $16, %xmm0
|
|
; SSE-NEXT: pmaxsw %xmm1, %xmm0
|
|
; SSE-NEXT: movd %xmm0, %eax
|
|
; SSE-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; SSE-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v4i16:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i16 @llvm.vector.reduce.smax.v4i16(<4 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
define i16 @test_v8i16(<8 x i16> %a0) nounwind {
|
|
; SSE2-LABEL: test_v8i16:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE2-NEXT: psrld $16, %xmm1
|
|
; SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE2-NEXT: movd %xmm1, %eax
|
|
; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; X86-SSE4-LABEL: test_v8i16:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v8i16:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX-LABEL: test_v8i16:
|
|
; X86-AVX: # %bb.0:
|
|
; X86-AVX-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX-NEXT: retl
|
|
;
|
|
; X64-AVX-LABEL: test_v8i16:
|
|
; X64-AVX: # %bb.0:
|
|
; X64-AVX-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v8i16:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v8i16:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512VL-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
define i16 @test_v16i16(<16 x i16> %a0) nounwind {
|
|
; SSE2-LABEL: test_v16i16:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE2-NEXT: psrld $16, %xmm1
|
|
; SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; SSE2-NEXT: movd %xmm1, %eax
|
|
; SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; X86-SSE4-LABEL: test_v16i16:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v16i16:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v16i16:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v16i16:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v16i16:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v16i16:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v16i16:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v16i16:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512VL-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i16 @llvm.vector.reduce.smax.v16i16(<16 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
define i16 @test_v32i16(<32 x i16> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v32i16:
|
|
; 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: pmaxsw %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pmaxsw 8(%ebp), %xmm1
|
|
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
|
|
; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: psrld $16, %xmm0
|
|
; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: movd %xmm0, %eax
|
|
; X86-SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v32i16:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: psrld $16, %xmm1
|
|
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %eax
|
|
; X64-SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v32i16:
|
|
; 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: pmaxsw %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsw 8(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
|
|
; X86-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v32i16:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v32i16:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm2, %xmm3, %xmm2
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v32i16:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm2, %xmm3, %xmm2
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v32i16:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v32i16:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v32i16:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v32i16:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512VL-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i16 @llvm.vector.reduce.smax.v32i16(<32 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
define i16 @test_v64i16(<64 x i16> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v64i16:
|
|
; 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: pmaxsw 72(%ebp), %xmm3
|
|
; X86-SSE2-NEXT: pmaxsw 40(%ebp), %xmm1
|
|
; X86-SSE2-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pmaxsw 56(%ebp), %xmm2
|
|
; X86-SSE2-NEXT: pmaxsw 24(%ebp), %xmm0
|
|
; X86-SSE2-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: psrld $16, %xmm1
|
|
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movd %xmm1, %eax
|
|
; X86-SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v64i16:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pmaxsw %xmm7, %xmm3
|
|
; X64-SSE2-NEXT: pmaxsw %xmm5, %xmm1
|
|
; X64-SSE2-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pmaxsw %xmm6, %xmm2
|
|
; X64-SSE2-NEXT: pmaxsw %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: psrld $16, %xmm1
|
|
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %eax
|
|
; X64-SSE2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v64i16:
|
|
; 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: pmaxsw 72(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: pmaxsw 40(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pmaxsw 56(%ebp), %xmm2
|
|
; X86-SSE4-NEXT: pmaxsw 24(%ebp), %xmm0
|
|
; X86-SSE4-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v64i16:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsw %xmm7, %xmm3
|
|
; X64-SSE4-NEXT: pmaxsw %xmm5, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsw %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsw %xmm6, %xmm2
|
|
; X64-SSE4-NEXT: pmaxsw %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsw %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-SSE4-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v64i16:
|
|
; 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: vpmaxsw %xmm3, %xmm4, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsw 24(%ebp), %xmm4, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm4, %xmm3, %xmm3
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsw 8(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsw %xmm3, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v64i16:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm5, %xmm6, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsw %xmm4, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX1-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v64i16:
|
|
; 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: vpmaxsw %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpmaxsw 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X86-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v64i16:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpmaxsw %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpmaxsw %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; X64-AVX2-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v64i16:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpmaxsw %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v64i16:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpmaxsw %zmm1, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsw %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsw %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorl $32767, %eax # imm = 0x7FFF
|
|
; AVX512VL-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i16 @llvm.vector.reduce.smax.v64i16(<64 x i16> %a0)
|
|
ret i16 %1
|
|
}
|
|
|
|
;
|
|
; vXi8
|
|
;
|
|
|
|
define i8 @test_v2i8(<2 x i8> %a0) nounwind {
|
|
; SSE2-LABEL: test_v2i8:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: movd %xmm0, %ecx
|
|
; SSE2-NEXT: movl %ecx, %eax
|
|
; SSE2-NEXT: shrl $8, %eax
|
|
; SSE2-NEXT: cmpb %al, %cl
|
|
; SSE2-NEXT: cmovgl %ecx, %eax
|
|
; SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: test_v2i8:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE4-NEXT: psrlw $8, %xmm1
|
|
; SSE4-NEXT: pmaxsb %xmm0, %xmm1
|
|
; SSE4-NEXT: movd %xmm1, %eax
|
|
; SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v2i8:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i8 @llvm.vector.reduce.smax.v2i8(<2 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v4i8(<4 x i8> %a0) nounwind {
|
|
; SSE2-LABEL: test_v4i8:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE2-NEXT: psrld $16, %xmm1
|
|
; SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; SSE2-NEXT: pand %xmm2, %xmm0
|
|
; SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; SSE2-NEXT: por %xmm0, %xmm2
|
|
; SSE2-NEXT: movd %xmm2, %ecx
|
|
; SSE2-NEXT: movl %ecx, %eax
|
|
; SSE2-NEXT: shrl $8, %eax
|
|
; SSE2-NEXT: cmpb %al, %cl
|
|
; SSE2-NEXT: cmovgl %ecx, %eax
|
|
; SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; SSE2-NEXT: ret{{[l|q]}}
|
|
;
|
|
; SSE4-LABEL: test_v4i8:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE4-NEXT: psrld $16, %xmm1
|
|
; SSE4-NEXT: pmaxsb %xmm0, %xmm1
|
|
; SSE4-NEXT: movdqa %xmm1, %xmm0
|
|
; SSE4-NEXT: psrlw $8, %xmm0
|
|
; SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; SSE4-NEXT: movd %xmm0, %eax
|
|
; SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v4i8:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i8 @llvm.vector.reduce.smax.v4i8(<4 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v8i8(<8 x i8> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v8i8:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,1,1]
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: psrld $16, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: movl %ecx, %eax
|
|
; X86-SSE2-NEXT: shrl $8, %eax
|
|
; X86-SSE2-NEXT: cmpb %al, %cl
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v8i8:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psrld $16, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %ecx
|
|
; X64-SSE2-NEXT: movl %ecx, %eax
|
|
; X64-SSE2-NEXT: shrl $8, %eax
|
|
; X64-SSE2-NEXT: cmpb %al, %cl
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; SSE4-LABEL: test_v8i8:
|
|
; SSE4: # %bb.0:
|
|
; SSE4-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; SSE4-NEXT: pmaxsb %xmm0, %xmm1
|
|
; SSE4-NEXT: movdqa %xmm1, %xmm0
|
|
; SSE4-NEXT: psrld $16, %xmm0
|
|
; SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; SSE4-NEXT: psrlw $8, %xmm1
|
|
; SSE4-NEXT: pmaxsb %xmm0, %xmm1
|
|
; SSE4-NEXT: movd %xmm1, %eax
|
|
; SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; SSE4-NEXT: ret{{[l|q]}}
|
|
;
|
|
; AVX-LABEL: test_v8i8:
|
|
; AVX: # %bb.0:
|
|
; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX-NEXT: vmovd %xmm0, %eax
|
|
; AVX-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX-NEXT: ret{{[l|q]}}
|
|
%1 = call i8 @llvm.vector.reduce.smax.v8i8(<8 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v16i8(<16 x i8> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v16i8:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: psrld $16, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: movl %ecx, %eax
|
|
; X86-SSE2-NEXT: shrl $8, %eax
|
|
; X86-SSE2-NEXT: cmpb %al, %cl
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v16i8:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: psrld $16, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X64-SSE2-NEXT: movl %ecx, %eax
|
|
; X64-SSE2-NEXT: shrl $8, %eax
|
|
; X64-SSE2-NEXT: cmpb %al, %cl
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v16i8:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X86-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorb $127, %al
|
|
; X86-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v16i8:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X64-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb $127, %al
|
|
; X64-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX-LABEL: test_v16i8:
|
|
; X86-AVX: # %bb.0:
|
|
; X86-AVX-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX-NEXT: xorb $127, %al
|
|
; X86-AVX-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX-NEXT: retl
|
|
;
|
|
; X64-AVX-LABEL: test_v16i8:
|
|
; X64-AVX: # %bb.0:
|
|
; X64-AVX-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX-NEXT: xorb $127, %al
|
|
; X64-AVX-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v16i8:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512BW-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorb $127, %al
|
|
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v16i8:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512VL-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorb $127, %al
|
|
; AVX512VL-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i8 @llvm.vector.reduce.smax.v16i8(<16 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v32i8(<32 x i8> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v32i8:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: psrld $16, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: movl %ecx, %eax
|
|
; X86-SSE2-NEXT: shrl $8, %eax
|
|
; X86-SSE2-NEXT: cmpb %al, %cl
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v32i8:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psrld $16, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %ecx
|
|
; X64-SSE2-NEXT: movl %ecx, %eax
|
|
; X64-SSE2-NEXT: shrl $8, %eax
|
|
; X64-SSE2-NEXT: cmpb %al, %cl
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v32i8:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X86-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorb $127, %al
|
|
; X86-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v32i8:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X64-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb $127, %al
|
|
; X64-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v32i8:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb $127, %al
|
|
; X86-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v32i8:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb $127, %al
|
|
; X64-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v32i8:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorb $127, %al
|
|
; X86-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v32i8:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorb $127, %al
|
|
; X64-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v32i8:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512BW-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorb $127, %al
|
|
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v32i8:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512VL-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorb $127, %al
|
|
; AVX512VL-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i8 @llvm.vector.reduce.smax.v32i8(<32 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v64i8(<64 x i8> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v64i8:
|
|
; 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), %xmm4
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,1,1]
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: psrld $16, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: movl %ecx, %eax
|
|
; X86-SSE2-NEXT: shrl $8, %eax
|
|
; X86-SSE2-NEXT: cmpb %al, %cl
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v64i8:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm4, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X64-SSE2-NEXT: psrld $16, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X64-SSE2-NEXT: movl %ecx, %eax
|
|
; X64-SSE2-NEXT: shrl $8, %eax
|
|
; X64-SSE2-NEXT: cmpb %al, %cl
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v64i8:
|
|
; 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: pmaxsb %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsb 8(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsb %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
|
|
; X86-SSE4-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: psrlw $8, %xmm0
|
|
; X86-SSE4-NEXT: pminub %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: phminposuw %xmm0, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorb $127, %al
|
|
; X86-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v64i8:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsb %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsb %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X64-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb $127, %al
|
|
; X64-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v64i8:
|
|
; X86-AVX1: # %bb.0:
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm2, %xmm3, %xmm2
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb $127, %al
|
|
; X86-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v64i8:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm2, %xmm3, %xmm2
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb $127, %al
|
|
; X64-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v64i8:
|
|
; X86-AVX2: # %bb.0:
|
|
; X86-AVX2-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorb $127, %al
|
|
; X86-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v64i8:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorb $127, %al
|
|
; X64-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v64i8:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512BW-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorb $127, %al
|
|
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v64i8:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512VL-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorb $127, %al
|
|
; AVX512VL-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i8 @llvm.vector.reduce.smax.v64i8(<64 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
define i8 @test_v128i8(<128 x i8> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v128i8:
|
|
; 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 72(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: movdqa 8(%ebp), %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: movdqa 40(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: pand %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm5
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: movdqa %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm5
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm1
|
|
; X86-SSE2-NEXT: movdqa 56(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm5, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm3
|
|
; X86-SSE2-NEXT: movdqa 24(%ebp), %xmm4
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm4, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm4, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X86-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X86-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,1,1]
|
|
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pand %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: pandn %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: por %xmm1, %xmm0
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE2-NEXT: psrld $16, %xmm1
|
|
; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X86-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X86-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X86-SSE2-NEXT: movd %xmm2, %ecx
|
|
; X86-SSE2-NEXT: movl %ecx, %eax
|
|
; X86-SSE2-NEXT: shrl $8, %eax
|
|
; X86-SSE2-NEXT: cmpb %al, %cl
|
|
; X86-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X86-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE2-NEXT: movl %ebp, %esp
|
|
; X86-SSE2-NEXT: popl %ebp
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v128i8:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm8
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm7, %xmm8
|
|
; X64-SSE2-NEXT: pand %xmm8, %xmm3
|
|
; X64-SSE2-NEXT: pandn %xmm7, %xmm8
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm8
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm5, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm5, %xmm3
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm8, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pandn %xmm8, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm3, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm6, %xmm3
|
|
; X64-SSE2-NEXT: pand %xmm3, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm6, %xmm3
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm4, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: pand %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm0
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: pandn %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
|
|
; X64-SSE2-NEXT: movdqa %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE2-NEXT: psrld $16, %xmm0
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pcmpgtb %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: pand %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: movd %xmm1, %ecx
|
|
; X64-SSE2-NEXT: movl %ecx, %eax
|
|
; X64-SSE2-NEXT: shrl $8, %eax
|
|
; X64-SSE2-NEXT: cmpb %al, %cl
|
|
; X64-SSE2-NEXT: cmovgl %ecx, %eax
|
|
; X64-SSE2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v128i8:
|
|
; 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: pmaxsb 72(%ebp), %xmm3
|
|
; X86-SSE4-NEXT: pmaxsb 40(%ebp), %xmm1
|
|
; X86-SSE4-NEXT: pmaxsb %xmm3, %xmm1
|
|
; X86-SSE4-NEXT: pmaxsb 56(%ebp), %xmm2
|
|
; X86-SSE4-NEXT: pmaxsb 24(%ebp), %xmm0
|
|
; X86-SSE4-NEXT: pmaxsb %xmm2, %xmm0
|
|
; X86-SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
|
|
; X86-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X86-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X86-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X86-SSE4-NEXT: movd %xmm0, %eax
|
|
; X86-SSE4-NEXT: xorb $127, %al
|
|
; X86-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-SSE4-NEXT: movl %ebp, %esp
|
|
; X86-SSE4-NEXT: popl %ebp
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE4-LABEL: test_v128i8:
|
|
; X64-SSE4: # %bb.0:
|
|
; X64-SSE4-NEXT: pmaxsb %xmm7, %xmm3
|
|
; X64-SSE4-NEXT: pmaxsb %xmm5, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsb %xmm3, %xmm1
|
|
; X64-SSE4-NEXT: pmaxsb %xmm6, %xmm2
|
|
; X64-SSE4-NEXT: pmaxsb %xmm4, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsb %xmm2, %xmm0
|
|
; X64-SSE4-NEXT: pmaxsb %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
|
|
; X64-SSE4-NEXT: movdqa %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: psrlw $8, %xmm1
|
|
; X64-SSE4-NEXT: pminub %xmm0, %xmm1
|
|
; X64-SSE4-NEXT: phminposuw %xmm1, %xmm0
|
|
; X64-SSE4-NEXT: movd %xmm0, %eax
|
|
; X64-SSE4-NEXT: xorb $127, %al
|
|
; X64-SSE4-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-SSE4-NEXT: retq
|
|
;
|
|
; X86-AVX1-LABEL: test_v128i8:
|
|
; 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: vpmaxsb %xmm3, %xmm4, %xmm3
|
|
; X86-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsb 24(%ebp), %xmm4, %xmm4
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm4, %xmm3, %xmm3
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm2, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsb 8(%ebp), %xmm1, %xmm1
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpmaxsb %xmm3, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX1-NEXT: xorb $127, %al
|
|
; X86-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX1-NEXT: movl %ebp, %esp
|
|
; X86-AVX1-NEXT: popl %ebp
|
|
; X86-AVX1-NEXT: vzeroupper
|
|
; X86-AVX1-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v128i8:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5
|
|
; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm5, %xmm6, %xmm5
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm4, %xmm5, %xmm4
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm3, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm2, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpmaxsb %xmm4, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX1-NEXT: xorb $127, %al
|
|
; X64-AVX1-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX1-NEXT: vzeroupper
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X86-AVX2-LABEL: test_v128i8:
|
|
; 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: vpmaxsb %ymm2, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vpmaxsb 8(%ebp), %ymm1, %ymm1
|
|
; X86-AVX2-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; X86-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X86-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X86-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X86-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X86-AVX2-NEXT: xorb $127, %al
|
|
; X86-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X86-AVX2-NEXT: movl %ebp, %esp
|
|
; X86-AVX2-NEXT: popl %ebp
|
|
; X86-AVX2-NEXT: vzeroupper
|
|
; X86-AVX2-NEXT: retl
|
|
;
|
|
; X64-AVX2-LABEL: test_v128i8:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vpmaxsb %ymm3, %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpmaxsb %ymm2, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; X64-AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; X64-AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vphminposuw %xmm0, %xmm0
|
|
; X64-AVX2-NEXT: vmovd %xmm0, %eax
|
|
; X64-AVX2-NEXT: xorb $127, %al
|
|
; X64-AVX2-NEXT: # kill: def $al killed $al killed $eax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v128i8:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpmaxsb %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpxor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512BW-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512BW-NEXT: vmovd %xmm0, %eax
|
|
; AVX512BW-NEXT: xorb $127, %al
|
|
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v128i8:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpmaxsb %zmm1, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsb %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsb %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpxord {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpsrlw $8, %xmm0, %xmm1
|
|
; AVX512VL-NEXT: vpminub %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vphminposuw %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovd %xmm0, %eax
|
|
; AVX512VL-NEXT: xorb $127, %al
|
|
; AVX512VL-NEXT: # kill: def $al killed $al killed $eax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%1 = call i8 @llvm.vector.reduce.smax.v128i8(<128 x i8> %a0)
|
|
ret i8 %1
|
|
}
|
|
|
|
; Special Cases
|
|
|
|
define i64 @test_v6i64(<6 x i64> %a0) nounwind {
|
|
; X86-SSE2-LABEL: test_v6i64:
|
|
; X86-SSE2: # %bb.0:
|
|
; X86-SSE2-NEXT: pushl %edi
|
|
; X86-SSE2-NEXT: pushl %esi
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; X86-SSE2-NEXT: cmpl %ecx, %eax
|
|
; X86-SSE2-NEXT: movl %edx, %edi
|
|
; X86-SSE2-NEXT: sbbl %esi, %edi
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-SSE2-NEXT: cmovll %esi, %edx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE2-NEXT: cmovll %ecx, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %edi
|
|
; X86-SSE2-NEXT: movl %esi, %ecx
|
|
; X86-SSE2-NEXT: sbbl %edx, %ecx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE2-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE2-NEXT: cmovgel %edi, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE2-NEXT: movl %esi, %edi
|
|
; X86-SSE2-NEXT: sbbl %edx, %edi
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-SSE2-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE2-NEXT: cmovgel %ecx, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %edi
|
|
; X86-SSE2-NEXT: movl %esi, %ecx
|
|
; X86-SSE2-NEXT: sbbl %edx, %ecx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE2-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE2-NEXT: cmovgel %edi, %eax
|
|
; X86-SSE2-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE2-NEXT: movl %esi, %edi
|
|
; X86-SSE2-NEXT: sbbl %edx, %edi
|
|
; X86-SSE2-NEXT: cmovgel %ecx, %eax
|
|
; X86-SSE2-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE2-NEXT: popl %esi
|
|
; X86-SSE2-NEXT: popl %edi
|
|
; X86-SSE2-NEXT: retl
|
|
;
|
|
; X64-SSE2-LABEL: test_v6i64:
|
|
; X64-SSE2: # %bb.0:
|
|
; X64-SSE2-NEXT: movq %rcx, %xmm1
|
|
; X64-SSE2-NEXT: movq %rdx, %xmm0
|
|
; X64-SSE2-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
|
|
; X64-SSE2-NEXT: movq %rsi, %xmm1
|
|
; X64-SSE2-NEXT: movq %rdi, %xmm2
|
|
; X64-SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm1[0]
|
|
; X64-SSE2-NEXT: movq %r9, %xmm1
|
|
; X64-SSE2-NEXT: movq %r8, %xmm3
|
|
; X64-SSE2-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
|
|
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2147483648,2147483648]
|
|
; X64-SSE2-NEXT: movdqa %xmm3, %xmm4
|
|
; X64-SSE2-NEXT: pxor %xmm1, %xmm4
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm5
|
|
; X64-SSE2-NEXT: pxor %xmm1, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm5, %xmm6
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm4, %xmm6
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm4, %xmm5
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm5[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm7, %xmm4
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm6[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm4, %xmm5
|
|
; X64-SSE2-NEXT: pand %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: pandn %xmm3, %xmm5
|
|
; X64-SSE2-NEXT: por %xmm2, %xmm5
|
|
; X64-SSE2-NEXT: movdqa %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: pxor %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pxor %xmm0, %xmm1
|
|
; X64-SSE2-NEXT: movdqa %xmm2, %xmm3
|
|
; X64-SSE2-NEXT: pcmpgtd %xmm1, %xmm3
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[0,0,2,2]
|
|
; X64-SSE2-NEXT: pcmpeqd %xmm2, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
|
|
; X64-SSE2-NEXT: pand %xmm4, %xmm1
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,1,3,3]
|
|
; X64-SSE2-NEXT: por %xmm1, %xmm2
|
|
; X64-SSE2-NEXT: pand %xmm2, %xmm5
|
|
; X64-SSE2-NEXT: pandn %xmm0, %xmm2
|
|
; X64-SSE2-NEXT: por %xmm5, %xmm2
|
|
; X64-SSE2-NEXT: movq %xmm2, %rcx
|
|
; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; X64-SSE2-NEXT: movq %xmm0, %rax
|
|
; X64-SSE2-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE2-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE2-NEXT: retq
|
|
;
|
|
; X86-SSE4-LABEL: test_v6i64:
|
|
; X86-SSE4: # %bb.0:
|
|
; X86-SSE4-NEXT: pushl %edi
|
|
; X86-SSE4-NEXT: pushl %esi
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; X86-SSE4-NEXT: cmpl %ecx, %eax
|
|
; X86-SSE4-NEXT: movl %edx, %edi
|
|
; X86-SSE4-NEXT: sbbl %esi, %edi
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-SSE4-NEXT: cmovll %esi, %edx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE4-NEXT: cmovll %ecx, %eax
|
|
; X86-SSE4-NEXT: cmpl %eax, %edi
|
|
; X86-SSE4-NEXT: movl %esi, %ecx
|
|
; X86-SSE4-NEXT: sbbl %edx, %ecx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE4-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE4-NEXT: cmovgel %edi, %eax
|
|
; X86-SSE4-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE4-NEXT: movl %esi, %edi
|
|
; X86-SSE4-NEXT: sbbl %edx, %edi
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-SSE4-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE4-NEXT: cmovgel %ecx, %eax
|
|
; X86-SSE4-NEXT: cmpl %eax, %edi
|
|
; X86-SSE4-NEXT: movl %esi, %ecx
|
|
; X86-SSE4-NEXT: sbbl %edx, %ecx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-SSE4-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE4-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-SSE4-NEXT: cmovgel %edi, %eax
|
|
; X86-SSE4-NEXT: cmpl %eax, %ecx
|
|
; X86-SSE4-NEXT: movl %esi, %edi
|
|
; X86-SSE4-NEXT: sbbl %edx, %edi
|
|
; X86-SSE4-NEXT: cmovgel %ecx, %eax
|
|
; X86-SSE4-NEXT: cmovgel %esi, %edx
|
|
; X86-SSE4-NEXT: popl %esi
|
|
; X86-SSE4-NEXT: popl %edi
|
|
; X86-SSE4-NEXT: retl
|
|
;
|
|
; X64-SSE41-LABEL: test_v6i64:
|
|
; X64-SSE41: # %bb.0:
|
|
; X64-SSE41-NEXT: movq %rcx, %xmm0
|
|
; X64-SSE41-NEXT: movq %rdx, %xmm1
|
|
; X64-SSE41-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
; X64-SSE41-NEXT: movq %rsi, %xmm0
|
|
; X64-SSE41-NEXT: movq %rdi, %xmm3
|
|
; X64-SSE41-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
|
|
; X64-SSE41-NEXT: movq %r9, %xmm0
|
|
; X64-SSE41-NEXT: movq %r8, %xmm2
|
|
; X64-SSE41-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
|
|
; X64-SSE41-NEXT: pmovzxdq {{.*#+}} xmm4 = [2147483648,2147483648]
|
|
; X64-SSE41-NEXT: movdqa %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm4, %xmm0
|
|
; X64-SSE41-NEXT: movdqa %xmm3, %xmm5
|
|
; X64-SSE41-NEXT: pxor %xmm4, %xmm5
|
|
; X64-SSE41-NEXT: movdqa %xmm5, %xmm6
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm0, %xmm6
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm7 = xmm6[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm5
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm5[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm7, %xmm5
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm6[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm5, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm3, %xmm2
|
|
; X64-SSE41-NEXT: movapd %xmm2, %xmm0
|
|
; X64-SSE41-NEXT: xorpd %xmm4, %xmm0
|
|
; X64-SSE41-NEXT: pxor %xmm1, %xmm4
|
|
; X64-SSE41-NEXT: movapd %xmm0, %xmm3
|
|
; X64-SSE41-NEXT: pcmpgtd %xmm4, %xmm3
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm5 = xmm3[0,0,2,2]
|
|
; X64-SSE41-NEXT: pcmpeqd %xmm0, %xmm4
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,1,3,3]
|
|
; X64-SSE41-NEXT: pand %xmm5, %xmm4
|
|
; X64-SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,3,3]
|
|
; X64-SSE41-NEXT: por %xmm4, %xmm0
|
|
; X64-SSE41-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X64-SSE41-NEXT: pextrq $1, %xmm1, %rax
|
|
; X64-SSE41-NEXT: movq %xmm1, %rcx
|
|
; X64-SSE41-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE41-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE41-NEXT: retq
|
|
;
|
|
; X64-SSE42-LABEL: test_v6i64:
|
|
; X64-SSE42: # %bb.0:
|
|
; X64-SSE42-NEXT: movq %rcx, %xmm0
|
|
; X64-SSE42-NEXT: movq %rdx, %xmm1
|
|
; X64-SSE42-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
|
|
; X64-SSE42-NEXT: movq %r9, %xmm0
|
|
; X64-SSE42-NEXT: movq %r8, %xmm2
|
|
; X64-SSE42-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
|
|
; X64-SSE42-NEXT: movq %rsi, %xmm0
|
|
; X64-SSE42-NEXT: movq %rdi, %xmm3
|
|
; X64-SSE42-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
|
|
; X64-SSE42-NEXT: movdqa %xmm3, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm2, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm3, %xmm2
|
|
; X64-SSE42-NEXT: movapd %xmm2, %xmm0
|
|
; X64-SSE42-NEXT: pcmpgtq %xmm1, %xmm0
|
|
; X64-SSE42-NEXT: blendvpd %xmm0, %xmm2, %xmm1
|
|
; X64-SSE42-NEXT: pextrq $1, %xmm1, %rax
|
|
; X64-SSE42-NEXT: movq %xmm1, %rcx
|
|
; X64-SSE42-NEXT: cmpq %rax, %rcx
|
|
; X64-SSE42-NEXT: cmovgq %rcx, %rax
|
|
; X64-SSE42-NEXT: retq
|
|
;
|
|
; X86-AVX-LABEL: test_v6i64:
|
|
; X86-AVX: # %bb.0:
|
|
; X86-AVX-NEXT: pushl %edi
|
|
; X86-AVX-NEXT: pushl %esi
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; X86-AVX-NEXT: cmpl %ecx, %eax
|
|
; X86-AVX-NEXT: movl %edx, %edi
|
|
; X86-AVX-NEXT: sbbl %esi, %edi
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-AVX-NEXT: cmovll %esi, %edx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-AVX-NEXT: cmovll %ecx, %eax
|
|
; X86-AVX-NEXT: cmpl %eax, %edi
|
|
; X86-AVX-NEXT: movl %esi, %ecx
|
|
; X86-AVX-NEXT: sbbl %edx, %ecx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-AVX-NEXT: cmovgel %esi, %edx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-AVX-NEXT: cmovgel %edi, %eax
|
|
; X86-AVX-NEXT: cmpl %eax, %ecx
|
|
; X86-AVX-NEXT: movl %esi, %edi
|
|
; X86-AVX-NEXT: sbbl %edx, %edi
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; X86-AVX-NEXT: cmovgel %esi, %edx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-AVX-NEXT: cmovgel %ecx, %eax
|
|
; X86-AVX-NEXT: cmpl %eax, %edi
|
|
; X86-AVX-NEXT: movl %esi, %ecx
|
|
; X86-AVX-NEXT: sbbl %edx, %ecx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; X86-AVX-NEXT: cmovgel %esi, %edx
|
|
; X86-AVX-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; X86-AVX-NEXT: cmovgel %edi, %eax
|
|
; X86-AVX-NEXT: cmpl %eax, %ecx
|
|
; X86-AVX-NEXT: movl %esi, %edi
|
|
; X86-AVX-NEXT: sbbl %edx, %edi
|
|
; X86-AVX-NEXT: cmovgel %ecx, %eax
|
|
; X86-AVX-NEXT: cmovgel %esi, %edx
|
|
; X86-AVX-NEXT: popl %esi
|
|
; X86-AVX-NEXT: popl %edi
|
|
; X86-AVX-NEXT: retl
|
|
;
|
|
; X64-AVX1-LABEL: test_v6i64:
|
|
; X64-AVX1: # %bb.0:
|
|
; X64-AVX1-NEXT: vmovq %rcx, %xmm0
|
|
; X64-AVX1-NEXT: vmovq %rdx, %xmm1
|
|
; X64-AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
|
|
; X64-AVX1-NEXT: vmovddup {{.*#+}} xmm1 = [9223372036854775808,9223372036854775808]
|
|
; X64-AVX1-NEXT: # xmm1 = mem[0,0]
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX1-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX1-NEXT: vmovq %r9, %xmm1
|
|
; X64-AVX1-NEXT: vmovq %r8, %xmm2
|
|
; X64-AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
|
|
; X64-AVX1-NEXT: vmovq %rsi, %xmm2
|
|
; X64-AVX1-NEXT: vmovq %rdi, %xmm3
|
|
; X64-AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm3[0],xmm2[0]
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm1, %xmm2, %xmm3
|
|
; X64-AVX1-NEXT: vblendvpd %xmm3, %xmm2, %xmm1, %xmm1
|
|
; X64-AVX1-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm2
|
|
; X64-AVX1-NEXT: vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
|
|
; X64-AVX1-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX1-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX1-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX1-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX1-NEXT: retq
|
|
;
|
|
; X64-AVX2-LABEL: test_v6i64:
|
|
; X64-AVX2: # %bb.0:
|
|
; X64-AVX2-NEXT: vmovq %rcx, %xmm0
|
|
; X64-AVX2-NEXT: vmovq %rdx, %xmm1
|
|
; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
|
|
; X64-AVX2-NEXT: vmovq %rsi, %xmm1
|
|
; X64-AVX2-NEXT: vmovq %rdi, %xmm2
|
|
; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
|
|
; X64-AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
|
|
; X64-AVX2-NEXT: vmovq %r9, %xmm1
|
|
; X64-AVX2-NEXT: vmovq %r8, %xmm2
|
|
; X64-AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
|
|
; X64-AVX2-NEXT: vinserti128 $1, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm2
|
|
; X64-AVX2-NEXT: vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
|
|
; X64-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1
|
|
; X64-AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm2
|
|
; X64-AVX2-NEXT: vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
|
|
; X64-AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; X64-AVX2-NEXT: vmovq %xmm0, %rcx
|
|
; X64-AVX2-NEXT: cmpq %rax, %rcx
|
|
; X64-AVX2-NEXT: cmovgq %rcx, %rax
|
|
; X64-AVX2-NEXT: vzeroupper
|
|
; X64-AVX2-NEXT: retq
|
|
;
|
|
; AVX512BW-LABEL: test_v6i64:
|
|
; AVX512BW: # %bb.0:
|
|
; AVX512BW-NEXT: vpbroadcastq {{.*#+}} zmm1 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
|
|
; AVX512BW-NEXT: movb $64, %al
|
|
; AVX512BW-NEXT: kmovd %eax, %k1
|
|
; AVX512BW-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1}
|
|
; AVX512BW-NEXT: movb $-128, %al
|
|
; AVX512BW-NEXT: kmovd %eax, %k1
|
|
; AVX512BW-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1}
|
|
; AVX512BW-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512BW-NEXT: vpmaxsq %zmm1, %zmm0, %zmm0
|
|
; AVX512BW-NEXT: vmovq %xmm0, %rax
|
|
; AVX512BW-NEXT: vzeroupper
|
|
; AVX512BW-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: test_v6i64:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm1 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
|
|
; AVX512VL-NEXT: movb $64, %al
|
|
; AVX512VL-NEXT: kmovd %eax, %k1
|
|
; AVX512VL-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: movb $-128, %al
|
|
; AVX512VL-NEXT: kmovd %eax, %k1
|
|
; AVX512VL-NEXT: vmovdqa64 %zmm1, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
|
|
; AVX512VL-NEXT: vpmaxsq %ymm1, %ymm0, %ymm0
|
|
; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm1
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
|
|
; AVX512VL-NEXT: vpmaxsq %xmm1, %xmm0, %xmm0
|
|
; AVX512VL-NEXT: vmovq %xmm0, %rax
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
%r = call i64 @llvm.vector.reduce.smax.v6i64(<6 x i64> %a0)
|
|
ret i64 %r
|
|
}
|
|
|
|
declare i64 @llvm.vector.reduce.smax.v2i64(<2 x i64>)
|
|
declare i64 @llvm.vector.reduce.smax.v4i64(<4 x i64>)
|
|
declare i64 @llvm.vector.reduce.smax.v8i64(<8 x i64>)
|
|
declare i64 @llvm.vector.reduce.smax.v16i64(<16 x i64>)
|
|
|
|
declare i32 @llvm.vector.reduce.smax.v2i32(<2 x i32>)
|
|
declare i32 @llvm.vector.reduce.smax.v4i32(<4 x i32>)
|
|
declare i32 @llvm.vector.reduce.smax.v8i32(<8 x i32>)
|
|
declare i32 @llvm.vector.reduce.smax.v16i32(<16 x i32>)
|
|
declare i32 @llvm.vector.reduce.smax.v32i32(<32 x i32>)
|
|
|
|
declare i16 @llvm.vector.reduce.smax.v2i16(<2 x i16>)
|
|
declare i16 @llvm.vector.reduce.smax.v4i16(<4 x i16>)
|
|
declare i16 @llvm.vector.reduce.smax.v8i16(<8 x i16>)
|
|
declare i16 @llvm.vector.reduce.smax.v16i16(<16 x i16>)
|
|
declare i16 @llvm.vector.reduce.smax.v32i16(<32 x i16>)
|
|
declare i16 @llvm.vector.reduce.smax.v64i16(<64 x i16>)
|
|
|
|
declare i8 @llvm.vector.reduce.smax.v2i8(<2 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v4i8(<4 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v8i8(<8 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v16i8(<16 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v32i8(<32 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v64i8(<64 x i8>)
|
|
declare i8 @llvm.vector.reduce.smax.v128i8(<128 x i8>)
|