This is an archive of the discontinued LLVM Phabricator instance.

Differential D17953

AVX512: Add extract_subvector patterns.
ClosedPublic

Authored by igorb on Mar 8 2016, 12:06 AM.

Details

Reviewers
delena
Commits
rG999ac754f247: AVX512: Add extract_subvector patterns v8i1->v4i1 , v4i1->v2i1.
rL262929: AVX512: Add extract_subvector patterns v8i1->v4i1 , v4i1->v2i1.
Summary

AVX512: Add extract_subvector patterns v8i1->v4i1 , v4i1->v2i1.

Diff Detail

Repository
rL LLVM

Event Timeline

igorb updated this revision to Diff 50026.Mar 8 2016, 12:06 AM
igorb retitled this revision from to AVX512: Add extract_subvector patterns..
igorb updated this object.
igorb added a reviewer: delena.
igorb set the repository for this revision to rL LLVM.
igorb added a subscriber: llvm-commits.
delena accepted this revision.Mar 8 2016, 6:55 AM
delena edited edge metadata.

LGTM

This revision is now accepted and ready to land.Mar 8 2016, 6:55 AM
igorb added a comment.Mar 8 2016, 6:57 AM

Thanks for review!

Closed by commit rL262929: AVX512: Add extract_subvector patterns v8i1->v4i1 , v4i1->v2i1. (authored by ibreger). · Explain WhyMar 8 2016, 7:26 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Target/
X86/
8 lines
test/
CodeGen/
X86/
23 lines

Diff 50026

lib/Target/X86/X86InstrAVX512.td

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 2,506 Lines • ▼ Show 20 Lines
defm : operation_subvector_mask_lowering<VK8, v8i1, VK32, v32i1>; defm : operation_subvector_mask_lowering<VK8, v8i1, VK32, v32i1>;
defm : operation_subvector_mask_lowering<VK8, v8i1, VK64, v64i1>; defm : operation_subvector_mask_lowering<VK8, v8i1, VK64, v64i1>;
defm : operation_subvector_mask_lowering<VK16, v16i1, VK32, v32i1>; defm : operation_subvector_mask_lowering<VK16, v16i1, VK32, v32i1>;
defm : operation_subvector_mask_lowering<VK16, v16i1, VK64, v64i1>; defm : operation_subvector_mask_lowering<VK16, v16i1, VK64, v64i1>;
defm : operation_subvector_mask_lowering<VK32, v32i1, VK64, v64i1>; defm : operation_subvector_mask_lowering<VK32, v32i1, VK64, v64i1>;
def : Pat<(v2i1 (extract_subvector (v4i1 VK4:$src), (iPTR 2))),
(v2i1 (COPY_TO_REGCLASS
(KSHIFTRWri (COPY_TO_REGCLASS VK4:$src, VK16), (i8 2)),
VK2))>;
def : Pat<(v4i1 (extract_subvector (v8i1 VK8:$src), (iPTR 4))),
(v4i1 (COPY_TO_REGCLASS
(KSHIFTRWri (COPY_TO_REGCLASS VK8:$src, VK16), (i8 4)),
VK4))>;
def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))), def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))),
(v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>; (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>;
def : Pat<(v16i1 (extract_subvector (v32i1 VK32:$src), (iPTR 16))), def : Pat<(v16i1 (extract_subvector (v32i1 VK32:$src), (iPTR 16))),
(v16i1 (COPY_TO_REGCLASS (KSHIFTRDri VK32:$src, (i8 16)), VK16))>; (v16i1 (COPY_TO_REGCLASS (KSHIFTRDri VK32:$src, (i8 16)), VK16))>;
def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 32))), def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 32))),
(v32i1 (COPY_TO_REGCLASS (KSHIFTRQri VK64:$src, (i8 32)), VK32))>; (v32i1 (COPY_TO_REGCLASS (KSHIFTRQri VK64:$src, (i8 32)), VK32))>;
def : Pat<(v8i1 (X86vshli VK8:$src, (i8 imm:$imm))), def : Pat<(v8i1 (X86vshli VK8:$src, (i8 imm:$imm))),
▲ Show 20 Lines • Show All 5,147 Lines • Show Last 20 Lines

test/CodeGen/X86/avx512-skx-insert-subvec.ll

Show First 20 Lines • Show All 139 Lines • ▼ Show 20 Lines
; CHECK-NEXT: kunpckdq %k1, %k0, %k0 ; CHECK-NEXT: kunpckdq %k1, %k0, %k0
; CHECK-NEXT: vpmovm2b %k0, %zmm0 ; CHECK-NEXT: vpmovm2b %k0, %zmm0
; CHECK-NEXT: retq ; CHECK-NEXT: retq
%res = shufflevector <8 x i1> %a, <8 x i1> %b, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> %res = shufflevector <8 x i1> %a, <8 x i1> %b, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
ret <64 x i1> %res ret <64 x i1> %res
} }
define <4 x i1> @test9(<8 x i1> %a, <8 x i1> %b) {
; CHECK-LABEL: test9:
; CHECK: # BB#0:
; CHECK-NEXT: vpsllw $15, %xmm0, %xmm0
; CHECK-NEXT: vpmovw2m %xmm0, %k0
; CHECK-NEXT: kshiftrw $4, %k0, %k0
; CHECK-NEXT: vpmovm2d %k0, %xmm0
; CHECK-NEXT: retq
%res = shufflevector <8 x i1> %a, <8 x i1> %b, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
ret <4 x i1> %res
}
define <2 x i1> @test10(<4 x i1> %a, <4 x i1> %b) {
; CHECK-LABEL: test10:
; CHECK: # BB#0:
; CHECK-NEXT: vpslld $31, %xmm0, %xmm0
; CHECK-NEXT: vptestmd %xmm0, %xmm0, %k0
; CHECK-NEXT: kshiftrw $2, %k0, %k0
; CHECK-NEXT: vpmovm2q %k0, %xmm0
; CHECK-NEXT: retq
%res = shufflevector <4 x i1> %a, <4 x i1> %b, <2 x i32> <i32 2, i32 3>
ret <2 x i1> %res
}