This is an archive of the discontinued LLVM Phabricator instance.

Differential D106413

[AArch64][GlobalISel] Widen s2 and s4 G_IMPLICIT_DEF + G_FREEZE
ClosedPublic

Authored by paquette on Jul 20 2021, 5:32 PM.

Details

Reviewers
aemerson
jroelofs
Commits
rGd0af732bd00c: [AArch64][GlobalISel] Widen s2 and s4 G_IMPLICIT_DEF + G_FREEZE
Summary

These had

.clampScalar(0, s1, 64)
.widenScalarToNextPow2(0, 8)

If you have s2 or s4, then widenScalarToNextPow2 does nothing.

This changes the widenScalarToNextPow2 rule to use s8 as the minimum type instead, allowing us to correctly widen s2 and s4.

This does not impact s1, since it's marked as legal already.

Diff Detail

Event Timeline

paquette created this revision.Jul 20 2021, 5:32 PM
Herald added subscribers: danielkiss, hiraditya, kristof.beyls, rovka. · View Herald TranscriptJul 20 2021, 5:32 PM
paquette requested review of this revision.Jul 20 2021, 5:32 PM
Herald added a project: Restricted Project. · View Herald TranscriptJul 20 2021, 5:32 PM
Harbormaster completed remote builds in B115228: Diff 360326.Jul 20 2021, 7:06 PM
aemerson accepted this revision.Jul 20 2021, 11:33 PM
This revision is now accepted and ready to land.Jul 20 2021, 11:33 PM
This revision was landed with ongoing or failed builds.Jul 21 2021, 12:59 PM
Closed by commit rGd0af732bd00c: [AArch64][GlobalISel] Widen s2 and s4 G_IMPLICIT_DEF + G_FREEZE (authored by paquette). · Explain Why
This revision was automatically updated to reflect the committed changes.
paquette added a commit: rGd0af732bd00c: [AArch64][GlobalISel] Widen s2 and s4 G_IMPLICIT_DEF + G_FREEZE.

Revision Contents

PathSize
llvm/
lib/
Target/
AArch64/
GISel/
2 lines
test/
CodeGen/
AArch64/
GlobalISel/
32 lines

Diff 360572

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

Show First 20 Lines • Show All 74 Lines • ▼ Show 20 LinesAArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
// Some instructions only support s16 if the subtarget has full 16-bit FP // Some instructions only support s16 if the subtarget has full 16-bit FP
// support. // support.
const bool HasFP16 = ST.hasFullFP16(); const bool HasFP16 = ST.hasFullFP16();
const LLT &MinFPScalar = HasFP16 ? s16 : s32; const LLT &MinFPScalar = HasFP16 ? s16 : s32;
getActionDefinitionsBuilder({G_IMPLICIT_DEF, G_FREEZE}) getActionDefinitionsBuilder({G_IMPLICIT_DEF, G_FREEZE})
.legalFor({p0, s1, s8, s16, s32, s64}) .legalFor({p0, s1, s8, s16, s32, s64})
.legalFor(PackedVectorAllTypeList) .legalFor(PackedVectorAllTypeList)
.clampScalar(0, s1, s64) .clampScalar(0, s8, s64)
.widenScalarToNextPow2(0, 8) .widenScalarToNextPow2(0, 8)
.fewerElementsIf( .fewerElementsIf(
[=](const LegalityQuery &Query) { [=](const LegalityQuery &Query) {
return Query.Types[0].isVector() && return Query.Types[0].isVector() &&
(Query.Types[0].getElementType() != s64 || (Query.Types[0].getElementType() != s64 ||
Query.Types[0].getNumElements() != 2); Query.Types[0].getNumElements() != 2);
}, },
[=](const LegalityQuery &Query) { [=](const LegalityQuery &Query) {
▲ Show 20 Lines • Show All 1,224 Lines • Show Last 20 Lines

llvm/test/CodeGen/AArch64/GlobalISel/legalize-freeze.mir

Show First 20 Lines • Show All 79 Lines • ▼ Show 20 Lines bb.0:
; CHECK: $w0 = COPY [[UV]](<4 x s8>) ; CHECK: $w0 = COPY [[UV]](<4 x s8>)
; CHECK: $w1 = COPY [[UV1]](<4 x s8>) ; CHECK: $w1 = COPY [[UV1]](<4 x s8>)
%d0:_(<8 x s8>) = COPY $d0 %d0:_(<8 x s8>) = COPY $d0
%0:_(<8 x s8>) = G_FREEZE %d0 %0:_(<8 x s8>) = G_FREEZE %d0
%1:_(<4 x s8>), %2:_(<4 x s8>) = G_UNMERGE_VALUES %0 %1:_(<4 x s8>), %2:_(<4 x s8>) = G_UNMERGE_VALUES %0
$w0 = COPY %1 $w0 = COPY %1
$w1 = COPY %2 $w1 = COPY %2
... ...
---
name: test_freeze_s1
body: |
bb.0.entry:
liveins: $x0
; CHECK-LABEL: name: test_freeze_s1
; CHECK: %x:_(s1) = G_IMPLICIT_DEF
; CHECK: %freeze:_(s1) = G_FREEZE %x
; CHECK: %ext:_(s64) = G_ZEXT %freeze(s1)
; CHECK: $x0 = COPY %ext(s64)
%x:_(s1) = G_IMPLICIT_DEF
%freeze:_(s1) = G_FREEZE %x
%ext:_(s64) = G_ZEXT %freeze
$x0 = COPY %ext(s64)
...
---
name: test_freeze_s2
body: |
bb.0.entry:
liveins: $x0
; CHECK-LABEL: name: test_freeze_s2
; CHECK: [[DEF:%[0-9]+]]:_(s8) = G_IMPLICIT_DEF
; CHECK: [[COPY:%[0-9]+]]:_(s8) = COPY [[DEF]](s8)
; CHECK: [[FREEZE:%[0-9]+]]:_(s8) = G_FREEZE [[COPY]]
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 3
; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[FREEZE]](s8)
; CHECK: %ext:_(s64) = G_AND [[ANYEXT]], [[C]]
; CHECK: $x0 = COPY %ext(s64)
%x:_(s2) = G_IMPLICIT_DEF
%freeze:_(s2) = G_FREEZE %x
%ext:_(s64) = G_ZEXT %freeze
$x0 = COPY %ext(s64)