This is an archive of the discontinued LLVM Phabricator instance.

Differential D153643

[GlobalIsel][X86] G_STORE extension
ClosedPublic

Authored by tschuett on Jun 23 2023, 9:37 AM.

Details

Reviewers
RKSimon
arsenm
Commits
rG68ed9d94720c: [GlobalIsel][X86] G_STORE extension

Diff Detail

Event Timeline

tschuett created this revision.Jun 23 2023, 9:37 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 23 2023, 9:37 AM
Herald added subscribers: pengfei, hiraditya. · View Herald Transcript
tschuett requested review of this revision.Jun 23 2023, 9:37 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 23 2023, 9:37 AM
Herald added a subscriber: llvm-commits. · View Herald Transcript
tschuett added reviewers: RKSimon, arsenm.Jun 23 2023, 9:38 AM
Herald added a subscriber: wdng. · View Herald TranscriptJun 23 2023, 9:38 AM
tschuett updated this revision to Diff 533990.Jun 23 2023, 9:45 AM

update check lines

Harbormaster completed remote builds in B240787: Diff 533990.Jun 23 2023, 10:45 AM
arsenm accepted this revision.Jun 23 2023, 1:09 PM

LGTM but drop the filler undef value references

llvm/test/CodeGen/X86/GlobalISel/legalize-store-avx.mir
19

don't need into ptr undef, can just not have a value reference

llvm/test/CodeGen/X86/GlobalISel/legalize-store-sse.mir
44

bb.0

This revision is now accepted and ready to land.Jun 23 2023, 1:09 PM
This revision was landed with ongoing or failed builds.Jun 23 2023, 8:51 PM
Closed by commit rG68ed9d94720c: [GlobalIsel][X86] G_STORE extension (authored by tschuett). · Explain Why
This revision was automatically updated to reflect the committed changes.
tschuett added a commit: rG68ed9d94720c: [GlobalIsel][X86] G_STORE extension.

Revision Contents

PathSize
llvm/
lib/
Target/
X86/
19 lines
test/
CodeGen/
X86/
GlobalISel/
89 lines
74 lines

Diff 533988

llvm/lib/Target/X86/X86LegalizerInfo.cpp

Show First 20 Lines • Show All 42 Lines • ▼ Show 20 LinesX86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
const LLT s1 = LLT::scalar(1); const LLT s1 = LLT::scalar(1);
const LLT s8 = LLT::scalar(8); const LLT s8 = LLT::scalar(8);
const LLT s16 = LLT::scalar(16); const LLT s16 = LLT::scalar(16);
const LLT s32 = LLT::scalar(32); const LLT s32 = LLT::scalar(32);
const LLT s64 = LLT::scalar(64); const LLT s64 = LLT::scalar(64);
const LLT s80 = LLT::scalar(80); const LLT s80 = LLT::scalar(80);
const LLT s128 = LLT::scalar(128); const LLT s128 = LLT::scalar(128);
const LLT sMaxScalar = Subtarget.is64Bit() ? s64 : s32; const LLT sMaxScalar = Subtarget.is64Bit() ? s64 : s32;
const LLT v2s32 = LLT::fixed_vector(2, 32);
const LLT v4s8 = LLT::fixed_vector(4, 8);
const LLT v16s8 = LLT::fixed_vector(16, 8); const LLT v16s8 = LLT::fixed_vector(16, 8);
const LLT v8s16 = LLT::fixed_vector(8, 16); const LLT v8s16 = LLT::fixed_vector(8, 16);
const LLT v4s32 = LLT::fixed_vector(4, 32); const LLT v4s32 = LLT::fixed_vector(4, 32);
const LLT v2s64 = LLT::fixed_vector(2, 64); const LLT v2s64 = LLT::fixed_vector(2, 64);
const LLT v2p0 = LLT::fixed_vector(2, p0);
const LLT v32s8 = LLT::fixed_vector(32, 8); const LLT v32s8 = LLT::fixed_vector(32, 8);
const LLT v16s16 = LLT::fixed_vector(16, 16); const LLT v16s16 = LLT::fixed_vector(16, 16);
const LLT v8s32 = LLT::fixed_vector(8, 32); const LLT v8s32 = LLT::fixed_vector(8, 32);
const LLT v4s64 = LLT::fixed_vector(4, 64); const LLT v4s64 = LLT::fixed_vector(4, 64);
const LLT v4p0 = LLT::fixed_vector(4, p0);
const LLT v64s8 = LLT::fixed_vector(64, 8); const LLT v64s8 = LLT::fixed_vector(64, 8);
const LLT v32s16 = LLT::fixed_vector(32, 16); const LLT v32s16 = LLT::fixed_vector(32, 16);
const LLT v16s32 = LLT::fixed_vector(16, 32); const LLT v16s32 = LLT::fixed_vector(16, 32);
const LLT v8s64 = LLT::fixed_vector(8, 64); const LLT v8s64 = LLT::fixed_vector(8, 64);
// todo: AVX512 bool vector predicate types // todo: AVX512 bool vector predicate types
▲ Show 20 Lines • Show All 256 Lines • ▼ Show 20 Lines getActionDefinitionsBuilder(G_PTR_ADD)
return typePairInSet(0, 1, {{p0, s32}})(Query) || return typePairInSet(0, 1, {{p0, s32}})(Query) ||
(Is64Bit && typePairInSet(0, 1, {{p0, s64}})(Query)); (Is64Bit && typePairInSet(0, 1, {{p0, s64}})(Query));
}) })
.widenScalarToNextPow2(1, /*Min*/ 32) .widenScalarToNextPow2(1, /*Min*/ 32)
.clampScalar(1, s32, sMaxScalar); .clampScalar(1, s32, sMaxScalar);
getActionDefinitionsBuilder({G_FRAME_INDEX, G_GLOBAL_VALUE}).legalFor({p0}); getActionDefinitionsBuilder({G_FRAME_INDEX, G_GLOBAL_VALUE}).legalFor({p0});
// load/store // load/store: add more corner cases
for (unsigned Op : {G_LOAD, G_STORE}) { for (unsigned Op : {G_LOAD, G_STORE}) {
auto &Action = getActionDefinitionsBuilder(Op); auto &Action = getActionDefinitionsBuilder(Op);
Action.legalForTypesWithMemDesc({{s8, p0, s1, 1}, Action.legalForTypesWithMemDesc({{s8, p0, s1, 1},
{s8, p0, s8, 1}, {s8, p0, s8, 1},
{s16, p0, s8, 1}, {s16, p0, s8, 1},
{s16, p0, s16, 1}, {s16, p0, s16, 1},
{s32, p0, s8, 1}, {s32, p0, s8, 1},
{s32, p0, s16, 1}, {s32, p0, s16, 1},
{s32, p0, s32, 1}, {s32, p0, s32, 1},
{s80, p0, s80, 1}, {s80, p0, s80, 1},
{p0, p0, p0, 1}}); {p0, p0, p0, 1},
{v4s8, p0, v4s8, 1}});
if (Is64Bit) if (Is64Bit)
Action.legalForTypesWithMemDesc({{s64, p0, s8, 1}, Action.legalForTypesWithMemDesc({{s64, p0, s8, 1},
{s64, p0, s16, 1}, {s64, p0, s16, 1},
{s64, p0, s32, 1}, {s64, p0, s32, 1},
{s64, p0, s64, 1}}); {s64, p0, s64, 1},
{v2s32, p0, v2s32, 1}});
if (HasSSE1) if (HasSSE1)
Action.legalForTypesWithMemDesc({{v16s8, p0, v16s8, 1}, Action.legalForTypesWithMemDesc({{v16s8, p0, v16s8, 1},
{v8s16, p0, v8s16, 1}, {v8s16, p0, v8s16, 1},
{v4s32, p0, v4s32, 1}, {v4s32, p0, v4s32, 1},
{v2s64, p0, v2s64, 1}}); {v2s64, p0, v2s64, 1},
{v2p0, p0, v2p0, 1}});
if (HasAVX) if (HasAVX)
Action.legalForTypesWithMemDesc({{v32s8, p0, v32s8, 1}, Action.legalForTypesWithMemDesc({{v32s8, p0, v32s8, 1},
{v16s16, p0, v16s16, 1}, {v16s16, p0, v16s16, 1},
{v8s32, p0, v8s32, 1}, {v8s32, p0, v8s32, 1},
{v4s64, p0, v4s64, 1}}); {v4s64, p0, v4s64, 1},
{v4p0, p0, v4p0, 1}});
if (HasAVX512) if (HasAVX512)
Action.legalForTypesWithMemDesc({{v64s8, p0, v64s8, 1}, Action.legalForTypesWithMemDesc({{v64s8, p0, v64s8, 1},
{v32s16, p0, v32s16, 1}, {v32s16, p0, v32s16, 1},
{v16s32, p0, v16s32, 1}, {v16s32, p0, v16s32, 1},
{v8s64, p0, v8s64, 1}}); {v8s64, p0, v8s64, 1}});
Action.widenScalarToNextPow2(0, /*Min=*/8).clampScalar(0, s8, sMaxScalar); Action.widenScalarToNextPow2(0, /*Min=*/8).clampScalar(0, s8, sMaxScalar);
} }
▲ Show 20 Lines • Show All 164 Lines • Show Last 20 Lines

llvm/test/CodeGen/X86/GlobalISel/legalize-store-avx.mir

  • This file was added.
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=X86-64-AVX
# test G_STORE with vectors
---
name: test_gstore_v2s32
body: |
bb.1:
; X86-64-AVX-LABEL: name: test_gstore_v2s32
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<2 x s32>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<2 x s32>), [[DEF1]](p0) :: (store (<2 x s32>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<2 x s32>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<2 x s32>), %1:_(p0) :: (store (<2 x s32>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
arsenmUnsubmitted
Not Done
ReplyInline Actions

don't need into ptr undef, can just not have a value reference

arsenm: don't need into `ptr undef`, can just not have a value reference
RET 0, implicit %2:_(p0)
...
---
name: test_gstore_v4s8
body: |
bb.1:
; X86-64-AVX-LABEL: name: test_gstore_v4s8
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<4 x s8>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<4 x s8>), [[DEF1]](p0) :: (store (<4 x s8>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<4 x s8>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<4 x s8>), %1:_(p0) :: (store (<4 x s8>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...
---
name: test_gstore_v4p0
body: |
bb.1:
; X86-64-SEE-LABEL: name: test_gstore_v4p0
; X86-64-SEE: [[DEF:%[0-9]+]]:_(<4 x p0>) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[UV:%[0-9]+]]:_(<2 x s64>), [[UV1:%[0-9]+]]:_(<2 x s64>) = G_UNMERGE_VALUES [[DEF]](<4 x p0>)
; X86-64-SEE-NEXT: G_STORE [[UV]](<2 x s64>), [[DEF1]](p0) :: (store (<2 x s64>) into `ptr undef`, align 32)
; X86-64-SEE-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
; X86-64-SEE-NEXT: [[PTR_ADD:%[0-9]+]]:_(p0) = G_PTR_ADD [[DEF1]], [[C]](s64)
; X86-64-SEE-NEXT: G_STORE [[UV1]](<2 x s64>), [[PTR_ADD]](p0) :: (store (<2 x s64>) into `ptr undef` + 16, basealign 32)
; X86-64-SEE-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-SEE-NEXT: RET 0, implicit [[COPY]](p0)
; X86-64-AVX-LABEL: name: test_gstore_v4p0
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<4 x p0>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<4 x p0>), [[DEF1]](p0) :: (store (<4 x p0>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<4 x p0>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<4 x p0>), %1:_(p0) :: (store (<4 x p0>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...
---
name: test_gstore_v2p0
body: |
bb.1:
; X86-64-SEE-LABEL: name: test_gstore_v2p0
; X86-64-SEE: [[DEF:%[0-9]+]]:_(<2 x p0>) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-SEE-NEXT: G_STORE [[DEF]](<2 x p0>), [[DEF1]](p0) :: (store (<2 x p0>) into `ptr undef`)
; X86-64-SEE-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-SEE-NEXT: RET 0, implicit [[COPY]](p0)
; X86-64-AVX-LABEL: name: test_gstore_v2p0
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<2 x p0>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<2 x p0>), [[DEF1]](p0) :: (store (<2 x p0>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<2 x p0>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<2 x p0>), %1:_(p0) :: (store (<2 x p0>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...

llvm/test/CodeGen/X86/GlobalISel/legalize-store-sse.mir

  • This file was added.
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+sse -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=X86-64-SEE
# test G_STORE with vectors
---
name: test_gstore_v2s32
body: |
bb.1:
; X86-64-AVX-LABEL: name: test_gstore_v2s32
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<2 x s32>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<2 x s32>), [[DEF1]](p0) :: (store (<2 x s32>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
; X86-64-SEE-LABEL: name: test_gstore_v2s32
; X86-64-SEE: [[DEF:%[0-9]+]]:_(<2 x s32>) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-SEE-NEXT: G_STORE [[DEF]](<2 x s32>), [[DEF1]](p0) :: (store (<2 x s32>) into `ptr undef`)
; X86-64-SEE-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-SEE-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<2 x s32>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<2 x s32>), %1:_(p0) :: (store (<2 x s32>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...
---
name: test_gstore_v4s8
body: |
bb.1:
; X86-64-AVX-LABEL: name: test_gstore_v4s8
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<4 x s8>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<4 x s8>), [[DEF1]](p0) :: (store (<4 x s8>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
; X86-64-SEE-LABEL: name: test_gstore_v4s8
; X86-64-SEE: [[DEF:%[0-9]+]]:_(<4 x s8>) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-SEE-NEXT: G_STORE [[DEF]](<4 x s8>), [[DEF1]](p0) :: (store (<4 x s8>) into `ptr undef`)
; X86-64-SEE-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-SEE-NEXT: RET 0, implicit [[COPY]](p0)
arsenmUnsubmitted
Not Done
ReplyInline Actions

bb.0

arsenm: bb.0
%0:_(<4 x s8>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<4 x s8>), %1:_(p0) :: (store (<4 x s8>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...
---
name: test_gstore_v2p0
body: |
bb.1:
; X86-64-AVX-LABEL: name: test_gstore_v2p0
; X86-64-AVX: [[DEF:%[0-9]+]]:_(<2 x p0>) = IMPLICIT_DEF
; X86-64-AVX-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-AVX-NEXT: G_STORE [[DEF]](<2 x p0>), [[DEF1]](p0) :: (store (<2 x p0>) into `ptr undef`)
; X86-64-AVX-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-AVX-NEXT: RET 0, implicit [[COPY]](p0)
; X86-64-SEE-LABEL: name: test_gstore_v2p0
; X86-64-SEE: [[DEF:%[0-9]+]]:_(<2 x p0>) = IMPLICIT_DEF
; X86-64-SEE-NEXT: [[DEF1:%[0-9]+]]:_(p0) = IMPLICIT_DEF
; X86-64-SEE-NEXT: G_STORE [[DEF]](<2 x p0>), [[DEF1]](p0) :: (store (<2 x p0>) into `ptr undef`)
; X86-64-SEE-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY [[DEF1]](p0)
; X86-64-SEE-NEXT: RET 0, implicit [[COPY]](p0)
%0:_(<2 x p0>) = IMPLICIT_DEF
%1:_(p0) = IMPLICIT_DEF
G_STORE %0:_(<2 x p0>), %1:_(p0) :: (store (<2 x p0>) into `ptr undef`)
%2:_(p0) = COPY %1:_(p0)
RET 0, implicit %2:_(p0)
...