This is an archive of the discontinued LLVM Phabricator instance.

Differential D148170

[LoopIdiomRecognize] Remove NUW flag from SCEV in getTripCount.
ClosedPublic

Authored by craig.topper on Apr 12 2023, 1:55 PM.

Details

Reviewers
reames
nikic
Commits
rG8bba57b1f12e: [LoopIdiomRecognize] Remove NUW flag from SCEV in getTripCount.
Summary

Based on the conversation in D147355.

Diff Detail

Event Timeline

craig.topper created this revision.Apr 12 2023, 1:55 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 12 2023, 1:55 PM
Herald added subscribers: StephenFan, hiraditya. · View Herald Transcript
craig.topper requested review of this revision.Apr 12 2023, 1:55 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 12 2023, 1:55 PM
craig.topper added inline comments.Apr 12 2023, 1:55 PM
llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
1010

This one also looks suspect if we aren't zero extending?

Harbormaster completed remote builds in B225175: Diff 512964.Apr 12 2023, 2:42 PM
nikic added inline comments.Apr 13 2023, 1:27 AM
llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
1010

Yes, this one should be dropped as well.

craig.topper updated this revision to Diff 513297.Apr 13 2023, 10:26 AM

Remove other NUW.

nikic accepted this revision.Apr 13 2023, 11:49 AM

LGTM

This revision is now accepted and ready to land.Apr 13 2023, 11:49 AM
This revision was landed with ongoing or failed builds.Apr 13 2023, 11:58 AM
Closed by commit rG8bba57b1f12e: [LoopIdiomRecognize] Remove NUW flag from SCEV in getTripCount. (authored by craig.topper). · Explain Why
This revision was automatically updated to reflect the committed changes.
craig.topper added a commit: rG8bba57b1f12e: [LoopIdiomRecognize] Remove NUW flag from SCEV in getTripCount..
Harbormaster completed remote builds in B225398: Diff 513297.Apr 13 2023, 12:27 PM

Revision Contents

PathSize
llvm/
lib/
Transforms/
Scalar/
4 lines
test/
Transforms/
LoopIdiom/
6 lines

Diff 513324

llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp

Show First 20 Lines • Show All 997 Lines • ▼ Show 20 Linesstatic const SCEV *getTripCount(const SCEV *BECount, Type *IntPtr,
// one to it prior to zero extending without overflow. Provided this is safe, // one to it prior to zero extending without overflow. Provided this is safe,
// it allows better simplification of the +1. // it allows better simplification of the +1.
if (DL->getTypeSizeInBits(BECount->getType()) < if (DL->getTypeSizeInBits(BECount->getType()) <
DL->getTypeSizeInBits(IntPtr) && DL->getTypeSizeInBits(IntPtr) &&
SE->isLoopEntryGuardedByCond( SE->isLoopEntryGuardedByCond(
CurLoop, ICmpInst::ICMP_NE, BECount, CurLoop, ICmpInst::ICMP_NE, BECount,
SE->getMinusOne(BECount->getType()))) { SE->getMinusOne(BECount->getType()))) {
TripCountS = SE->getZeroExtendExpr( TripCountS = SE->getZeroExtendExpr(
SE->getAddExpr(BECount, SE->getOne(BECount->getType()), SCEV::FlagNUW), SE->getAddExpr(BECount, SE->getOne(BECount->getType())),
IntPtr); IntPtr);
} else { } else {
TripCountS = SE->getAddExpr(SE->getTruncateOrZeroExtend(BECount, IntPtr), TripCountS = SE->getAddExpr(SE->getTruncateOrZeroExtend(BECount, IntPtr),
SE->getOne(IntPtr), SCEV::FlagNUW); SE->getOne(IntPtr));
craig.topperAuthorUnsubmitted
Done
ReplyInline Actions

This one also looks suspect if we aren't zero extending?

craig.topper: This one also looks suspect if we aren't zero extending?
nikicUnsubmitted
Not Done
ReplyInline Actions

Yes, this one should be dropped as well.

nikic: Yes, this one should be dropped as well.
} }
return TripCountS; return TripCountS;
} }
/// Compute the number of bytes as a SCEV from the backedge taken count. /// Compute the number of bytes as a SCEV from the backedge taken count.
/// ///
/// This also maps the SCEV into the provided type and tries to handle the /// This also maps the SCEV into the provided type and tries to handle the
▲ Show 20 Lines • Show All 1,851 Lines • Show Last 20 Lines

llvm/test/Transforms/LoopIdiom/memcpy-intrinsic.ll

Show First 20 Lines • Show All 190 Lines • ▼ Show 20 Lines
; Function Attrs: nofree nosync nounwind uwtable mustprogress ; Function Attrs: nofree nosync nounwind uwtable mustprogress
define dso_local i32 @copy_noalias_negative_stride(ptr noalias nocapture %arg, ptr nocapture readonly %arg1, i32 %arg2) local_unnamed_addr { define dso_local i32 @copy_noalias_negative_stride(ptr noalias nocapture %arg, ptr nocapture readonly %arg1, i32 %arg2) local_unnamed_addr {
; CHECK-LABEL: @copy_noalias_negative_stride( ; CHECK-LABEL: @copy_noalias_negative_stride(
; CHECK-NEXT: bb: ; CHECK-NEXT: bb:
; CHECK-NEXT: [[I:%.*]] = icmp sgt i32 [[ARG2:%.*]], -1 ; CHECK-NEXT: [[I:%.*]] = icmp sgt i32 [[ARG2:%.*]], -1
; CHECK-NEXT: br i1 [[I]], label [[BB3:%.*]], label [[BB5:%.*]] ; CHECK-NEXT: br i1 [[I]], label [[BB3:%.*]], label [[BB5:%.*]]
; CHECK: bb3: ; CHECK: bb3:
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[ARG2]] to i64 ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[ARG2]], 1
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw nsw i64 [[TMP0]], 12 ; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 12 ; CHECK-NEXT: [[TMP2:%.*]] = mul nuw nsw i64 [[TMP1]], 12
; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[ARG:%.*]], ptr align 4 [[ARG1:%.*]], i64 [[TMP2]], i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[ARG:%.*]], ptr align 4 [[ARG1:%.*]], i64 [[TMP2]], i1 false)
; CHECK-NEXT: br label [[BB6:%.*]] ; CHECK-NEXT: br label [[BB6:%.*]]
; CHECK: bb4: ; CHECK: bb4:
; CHECK-NEXT: br label [[BB5]] ; CHECK-NEXT: br label [[BB5]]
; CHECK: bb5: ; CHECK: bb5:
; CHECK-NEXT: ret i32 12 ; CHECK-NEXT: ret i32 12
; CHECK: bb6: ; CHECK: bb6:
; CHECK-NEXT: [[I7:%.*]] = phi i32 [ [[I13:%.*]], [[BB6]] ], [ [[ARG2]], [[BB3]] ] ; CHECK-NEXT: [[I7:%.*]] = phi i32 [ [[I13:%.*]], [[BB6]] ], [ [[ARG2]], [[BB3]] ]
▲ Show 20 Lines • Show All 195 Lines • Show Last 20 Lines