Skip to content

[LoopPeel] Use loop guards when checking if last iter can be peeled. #142605

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
fhahn merged 2 commits into from
Jun 10, 2025
Merged

[LoopPeel] Use loop guards when checking if last iter can be peeled. #142605

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9306a03
[LoopPeel] Use loop guards when checking if last iter can be peeled.
fhahn Jun 3, 2025
c7deb11
!fixup remove assert.
fhahn Jun 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions llvm/lib/Transforms/Utils/LoopPeel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -374,6 +374,9 @@ static bool shouldPeelLastIteration(Loop &L, CmpPredicate Pred,
L.getLoopPredecessor()->getTerminator()))
return false;

auto Guards = ScalarEvolution::LoopGuards::collect(&L, SE);
BTC = SE.applyLoopGuards(BTC, Guards);
Copy link
Contributor

@nikic nikic Jun 9, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

FWIW, I do think using loop guards for the isKnownNonZero(BTC) check makes sense, I was just saying it shouldn't be used for the high cost expansion check.

Copy link
Contributor Author

@fhahn fhahn Jun 10, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah yes, makes sense. I'll adjust it separately, with a separate test case.

RightSCEV = SE.applyLoopGuards(RightSCEV, Guards);
const SCEV *ValAtLastIter = LeftAR->evaluateAtIteration(BTC, SE);
const SCEV *ValAtSecondToLastIter = LeftAR->evaluateAtIteration(
SE.getMinusSCEV(BTC, SE.getOne(BTC->getType())), SE);
Expand Down
28 changes: 24 additions & 4 deletions llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,14 +13,33 @@ define void @peel_with_guard_known_nonnegative_1(i32 %n) {
; CHECK-NEXT: [[N_EXT:%.*]] = zext i32 [[N]] to i64
; CHECK-NEXT: [[N_1:%.*]] = add i32 [[N]], 1
; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N_1]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = add nsw i64 [[WIDE_TRIP_COUNT]], -1
; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i64 [[TMP0]], 0
; CHECK-NEXT: br i1 [[TMP1]], label %[[PH_SPLIT:.*]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN:.*]]
; CHECK: [[PH_SPLIT]]:
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
; CHECK-NEXT: [[IV1:%.*]] = phi i64 [ 0, %[[PH_SPLIT]] ], [ [[IV_NEXT1:%.*]], %[[LOOP]] ]
; CHECK-NEXT: [[IV_NEXT1]] = add nuw nsw i64 [[IV1]], 1
; CHECK-NEXT: [[TMP2:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], 1
; CHECK-NEXT: [[EC1:%.*]] = icmp eq i64 [[IV_NEXT1]], [[TMP2]]
; CHECK-NEXT: br i1 [[EC1]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]]:
; CHECK-NEXT: [[DOTPH:%.*]] = phi i64 [ [[IV_NEXT1]], %[[LOOP]] ]
; CHECK-NEXT: br label %[[EXIT_LOOPEXIT_PEEL_BEGIN]]
; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[PH]] ], [ [[DOTPH]], %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]] ]
; CHECK-NEXT: br label %[[LOOP_PEEL:.*]]
; CHECK: [[LOOP_PEEL]]:
; CHECK-NEXT: [[C:%.*]] = icmp eq i64 [[IV]], [[N_EXT]]
; CHECK-NEXT: [[SPEC_SELECT:%.*]] = select i1 [[C]], i32 10, i32 20
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
; CHECK-NEXT: [[IV_NEXT:%.*]] = add i64 [[IV]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[WIDE_TRIP_COUNT]]
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP]]
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_NEXT:.*]], label %[[EXIT_LOOPEXIT_PEEL_NEXT]]
; CHECK: [[EXIT_LOOPEXIT_PEEL_NEXT]]:
; CHECK-NEXT: br label %[[LOOP_PEEL_NEXT:.*]]
; CHECK: [[LOOP_PEEL_NEXT]]:
; CHECK-NEXT: br label %[[EXIT_LOOPEXIT:.*]]
; CHECK: [[EXIT_LOOPEXIT]]:
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT]]:
Expand Down Expand Up @@ -137,7 +156,7 @@ define void @peel_with_guard2(i32 %n) {
; CHECK-NEXT: [[IV_NEXT]] = add nuw i32 [[IV]], 1
; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[N]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], [[TMP2]]
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP2:![0-9]+]]
; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]]:
; CHECK-NEXT: [[DOTPH:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP_LATCH]] ]
; CHECK-NEXT: br label %[[EXIT_LOOPEXIT_PEEL_BEGIN]]
Expand Down Expand Up @@ -188,4 +207,5 @@ exit:
;.
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.peeled.count", i32 1}
; CHECK: [[LOOP2]] = distinct !{[[LOOP2]], [[META1]]}
;.