This is an archive of the discontinued LLVM Phabricator instance.

Differential D105912

[AArch64][GlobalISel] Legalize store <2 x i16>
ClosedPublic

Authored by jroelofs on Jul 13 2021, 10:33 AM.

Details

Reviewers
aemerson
paquette
Commits
rG43c7ca8e4963: [AArch64][GlobalISel] Legalize store <2 x i16>

Diff Detail

Unit TestsFailed

TimeTest
120 msx64 debian > LLVM.CodeGen/AArch64::arm64-rev.ll
2,650 msx64 debian > libarcher.critical::critical.c
2,530 msx64 debian > libarcher.races::critical-unrelated.c
2,660 msx64 debian > libarcher.races::lock-nested-unrelated.c
2,590 msx64 debian > libarcher.races::lock-unrelated.c
View Full Test Results (17 Failed)

Event Timeline

jroelofs created this revision.Jul 13 2021, 10:33 AM
Herald added subscribers: danielkiss, hiraditya, kristof.beyls, rovka. · View Herald TranscriptJul 13 2021, 10:33 AM
jroelofs requested review of this revision.Jul 13 2021, 10:33 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 13 2021, 10:33 AM
Herald added a subscriber: llvm-commits. · View Herald Transcript
paquette accepted this revision.Jul 13 2021, 10:37 AM

LGTM with test nit

llvm/test/CodeGen/AArch64/GlobalISel/legalize-load-store.mir
549

You can drop the alignment, registers, liveins, frameInfo, and machineFunctionInfo bits here.

This revision is now accepted and ready to land.Jul 13 2021, 10:37 AM
paquette mentioned this in D105913: [AArch64][GlobalISel] Legalize load <2 x i16>.Jul 13 2021, 10:38 AM
This revision was landed with ongoing or failed builds.Jul 13 2021, 11:13 AM
Closed by commit rG43c7ca8e4963: [AArch64][GlobalISel] Legalize store <2 x i16> (authored by jroelofs). · Explain Why
This revision was automatically updated to reflect the committed changes.
jroelofs added a commit: rG43c7ca8e4963: [AArch64][GlobalISel] Legalize store <2 x i16>.
Harbormaster completed remote builds in B113786: Diff 358337.Jul 13 2021, 12:54 PM

Revision Contents

PathSize
llvm/
lib/
Target/
AArch64/
GISel/
3 lines
test/
CodeGen/
AArch64/
GlobalISel/
33 lines

Diff 358337

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

Show First 20 Lines • Show All 308 Lines • ▼ Show 20 Lines getActionDefinitionsBuilder(G_LOAD)
}) })
.clampMaxNumElements(0, s8, 16) .clampMaxNumElements(0, s8, 16)
.clampMaxNumElements(0, s16, 8) .clampMaxNumElements(0, s16, 8)
.clampMaxNumElements(0, s32, 4) .clampMaxNumElements(0, s32, 4)
.clampMaxNumElements(0, s64, 2) .clampMaxNumElements(0, s64, 2)
.customIf(IsPtrVecPred); .customIf(IsPtrVecPred);
getActionDefinitionsBuilder(G_STORE) getActionDefinitionsBuilder(G_STORE)
.legalForTypesWithMemDesc({{s8, p0, s8, 8}, .legalForTypesWithMemDesc({{s8, p0, s8, 8},
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-      .legalForTypesWithMemDesc({{s8, p0, s8, 8},
-                                 {s16, p0, s8, 8}, // truncstorei8 from s16
-                                 {s32, p0, s8, 8}, // truncstorei8 from s32
-                                 {s64, p0, s8, 8}, // truncstorei8 from s64
-                                 {s16, p0, s16, 8},
-                                 {s32, p0, s16, 8}, // truncstorei16 from s32
-                                 {s64, p0, s16, 8}, // truncstorei16 from s64
-                                 {s32, p0, s8, 8},
-                                 {s32, p0, s16, 8},
-                                 {s32, p0, s32, 8},

22 diff lines are omitted. See full path.

Lint: Pre-merge checks: clang-format: please reformat the code ``` - .legalForTypesWithMemDesc({{s8, p0, s8, 8}…
{s16, p0, s8, 8}, // truncstorei8 from s16 {s16, p0, s8, 8}, // truncstorei8 from s16
{s32, p0, s8, 8}, // truncstorei8 from s32 {s32, p0, s8, 8}, // truncstorei8 from s32
{s64, p0, s8, 8}, // truncstorei8 from s64 {s64, p0, s8, 8}, // truncstorei8 from s64
{s16, p0, s16, 8}, {s16, p0, s16, 8},
{s32, p0, s16, 8}, // truncstorei16 from s32 {s32, p0, s16, 8}, // truncstorei16 from s32
{s64, p0, s16, 8}, // truncstorei16 from s64 {s64, p0, s16, 8}, // truncstorei16 from s64
{s32, p0, s8, 8}, {s32, p0, s8, 8},
{s32, p0, s16, 8}, {s32, p0, s16, 8},
Show All 15 Lines getActionDefinitionsBuilder(G_STORE)
Query.Types[0] != Query.MMODescrs[0].MemoryTy; Query.Types[0] != Query.MMODescrs[0].MemoryTy;
}) })
// Maximum: sN * k = 128 // Maximum: sN * k = 128
.clampMaxNumElements(0, s8, 16) .clampMaxNumElements(0, s8, 16)
.clampMaxNumElements(0, s16, 8) .clampMaxNumElements(0, s16, 8)
.clampMaxNumElements(0, s32, 4) .clampMaxNumElements(0, s32, 4)
.clampMaxNumElements(0, s64, 2) .clampMaxNumElements(0, s64, 2)
.lowerIfMemSizeNotPow2() .lowerIfMemSizeNotPow2()
.customIf(IsPtrVecPred); .customIf(IsPtrVecPred)
.scalarizeIf(typeIs(0, v2s16), 0);
// Constants // Constants
getActionDefinitionsBuilder(G_CONSTANT) getActionDefinitionsBuilder(G_CONSTANT)
.legalFor({p0, s8, s16, s32, s64}) .legalFor({p0, s8, s16, s32, s64})
.clampScalar(0, s8, s64) .clampScalar(0, s8, s64)
.widenScalarToNextPow2(0); .widenScalarToNextPow2(0);
getActionDefinitionsBuilder(G_FCONSTANT) getActionDefinitionsBuilder(G_FCONSTANT)
.legalIf([=](const LegalityQuery &Query) { .legalIf([=](const LegalityQuery &Query) {
▲ Show 20 Lines • Show All 825 Lines • Show Last 20 Lines

llvm/test/CodeGen/AArch64/GlobalISel/legalize-load-store.mir

Show First 20 Lines • Show All 536 Lines • ▼ Show 20 Lines bb.1:
; CHECK: [[PTR_ADD1:%[0-9]+]]:_(p0) = G_PTR_ADD %ptr, [[C1]](s64) ; CHECK: [[PTR_ADD1:%[0-9]+]]:_(p0) = G_PTR_ADD %ptr, [[C1]](s64)
; CHECK: G_STORE [[DEF]](<2 x s64>), [[PTR_ADD1]](p0) :: (store (<2 x s64>) into unknown-address + 32) ; CHECK: G_STORE [[DEF]](<2 x s64>), [[PTR_ADD1]](p0) :: (store (<2 x s64>) into unknown-address + 32)
; CHECK: RET_ReallyLR ; CHECK: RET_ReallyLR
%val:_(<6 x s64>) = G_IMPLICIT_DEF %val:_(<6 x s64>) = G_IMPLICIT_DEF
%ptr:_(p0) = COPY $x0 %ptr:_(p0) = COPY $x0
G_STORE %val(<6 x s64>), %ptr(p0) :: (store (<6 x s64>), align 16) G_STORE %val(<6 x s64>), %ptr(p0) :: (store (<6 x s64>), align 16)
RET_ReallyLR RET_ReallyLR
... ...
---
name: store_2xi16
alignment: 4
tracksRegLiveness: true
registers:
paquetteUnsubmitted
Not Done
ReplyInline Actions

You can drop the alignment, registers, liveins, frameInfo, and machineFunctionInfo bits here.

paquette: You can drop the `alignment`, `registers`, `liveins`, `frameInfo`, and `machineFunctionInfo`…
- { id: 0, class: _ }
- { id: 1, class: _ }
liveins:
- { reg: '$x0' }
- { reg: '$s0' }
frameInfo:
maxAlignment: 1
machineFunctionInfo: {}
body: |
bb.1:
liveins: $s0, $x0
; CHECK-LABEL: name: store_2xi16
; CHECK: liveins: $s0, $x0
; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $s0
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<2 x s16>)
; CHECK: G_STORE [[UV]](s16), [[COPY]](p0) :: (store (s16), align 4)
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
; CHECK: [[PTR_ADD:%[0-9]+]]:_(p0) = G_PTR_ADD [[COPY]], [[C]](s64)
; CHECK: G_STORE [[UV1]](s16), [[PTR_ADD]](p0) :: (store (s16) into unknown-address + 2)
; CHECK: RET_ReallyLR
%0:_(p0) = COPY $x0
%1:_(<2 x s16>) = COPY $s0
G_STORE %1(<2 x s16>), %0(p0) :: (store (<2 x s16>))
RET_ReallyLR
...