Skip to content

[LoopFusion] Forget loop and block dispositions after latch merge #166233

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
merged 1 commit into from
Nov 4, 2025
+75 −6
Merged

[LoopFusion] Forget loop and block dispositions after latch merge #166233

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 10 additions & 6 deletions llvm/lib/Transforms/Scalar/LoopFuse.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1796,14 +1796,16 @@ struct LoopFuser {
// mergeLatch may remove the only block in FC1.
SE.forgetLoop(FC1.L);
SE.forgetLoop(FC0.L);
// Forget block dispositions as well, so that there are no dangling
// pointers to erased/free'ed blocks.
SE.forgetBlockAndLoopDispositions();

// Move instructions from FC0.Latch to FC1.Latch.
// Note: mergeLatch requires an updated DT.
mergeLatch(FC0, FC1);

// Forget block dispositions as well, so that there are no dangling
// pointers to erased/free'ed blocks. It should be done after mergeLatch()
// since merging the latches may affect the dispositions.
SE.forgetBlockAndLoopDispositions();

// Merge the loops.
SmallVector<BasicBlock *, 8> Blocks(FC1.L->blocks());
for (BasicBlock *BB : Blocks) {
Expand Down Expand Up @@ -2092,14 +2094,16 @@ struct LoopFuser {
// mergeLatch may remove the only block in FC1.
SE.forgetLoop(FC1.L);
SE.forgetLoop(FC0.L);
// Forget block dispositions as well, so that there are no dangling
// pointers to erased/free'ed blocks.
SE.forgetBlockAndLoopDispositions();

// Move instructions from FC0.Latch to FC1.Latch.
// Note: mergeLatch requires an updated DT.
mergeLatch(FC0, FC1);

// Forget block dispositions as well, so that there are no dangling
// pointers to erased/free'ed blocks. It should be done after mergeLatch()
// since merging the latches may affect the dispositions.
SE.forgetBlockAndLoopDispositions();

// Merge the loops.
SmallVector<BasicBlock *, 8> Blocks(FC1.L->blocks());
for (BasicBlock *BB : Blocks) {
Expand Down
65 changes: 65 additions & 0 deletions llvm/test/Transforms/LoopFusion/pr164082.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,65 @@
; REQUIRES: asserts
; RUN: opt -passes=loop-fusion -disable-output -stats < %s 2>&1 | FileCheck -check-prefix=STAT %s
Copy link
Contributor

@CongzheUalberta CongzheUalberta Nov 4, 2025

Choose a reason for hiding this comment

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

Add ; REQUIRES: asserts

-stats requires asserts.

Copy link
Contributor Author

@1997alireza 1997alireza Nov 4, 2025

Choose a reason for hiding this comment

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

Added.

; STAT: 1 loop-fusion - Loops fused

; C Code
;
;; for (int i = 0; i < 100; ++i)
;; Array[i][i] = -i;
;; for (int row = 0; row < 100; ++row)
;; for (int col = 0; col < 100; ++col)
;; if (col != row)
;; Array[row][col] = row + col;
;
; Loop fusion should not crash anymore as now forgetBlockAndLoopDispositions()
; is trigerred after mergeLatch() during the fusion.

define i32 @forget_dispositions() nounwind {
entry:
%Array = alloca [100 x [100 x i32]], align 4
br label %for.body

for.body: ; preds = %for.body, %entry
%indvars.iv33 = phi i64 [ 0, %entry ], [ %indvars.iv.next34, %for.body ]
%0 = trunc i64 %indvars.iv33 to i32
%sub = sub i32 0, %0
%arrayidx2 = getelementptr inbounds [100 x [100 x i32]], ptr %Array, i64 0, i64 %indvars.iv33, i64 %indvars.iv33
store i32 %sub, ptr %arrayidx2, align 4
%indvars.iv.next34 = add i64 %indvars.iv33, 1
%lftr.wideiv35 = trunc i64 %indvars.iv.next34 to i32
%exitcond36 = icmp eq i32 %lftr.wideiv35, 100
br i1 %exitcond36, label %for.cond6.preheader, label %for.body

for.cond6.preheader: ; preds = %for.body, %for.inc17
%indvars.iv29 = phi i64 [ %indvars.iv.next30, %for.inc17 ], [ 0, %for.body ]
br label %for.body8

for.body8: ; preds = %for.inc14, %for.cond6.preheader
%indvars.iv = phi i64 [ 0, %for.cond6.preheader ], [ %indvars.iv.next, %for.inc14 ]
%1 = trunc i64 %indvars.iv to i32
%2 = trunc i64 %indvars.iv29 to i32
Copy link
Contributor

@CongzheUalberta CongzheUalberta Nov 4, 2025

Choose a reason for hiding this comment

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

Could you simplify the test IR? For example you may get rid of all trunc instructions.

Copy link
Contributor

@amehsan amehsan Nov 4, 2025

Choose a reason for hiding this comment

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

if the IR is generated by the compiler let's leave it as is?

%cmp9 = icmp eq i32 %1, %2
br i1 %cmp9, label %for.inc14, label %if.then

if.then: ; preds = %for.body8
%3 = add i64 %indvars.iv, %indvars.iv29
%arrayidx13 = getelementptr inbounds [100 x [100 x i32]], ptr %Array, i64 0, i64 %indvars.iv29, i64 %indvars.iv
%4 = trunc i64 %3 to i32
store i32 %4, ptr %arrayidx13, align 4
br label %for.inc14

for.inc14: ; preds = %for.body8, %if.then
%indvars.iv.next = add i64 %indvars.iv, 1
%lftr.wideiv27 = trunc i64 %indvars.iv.next to i32
%exitcond28 = icmp eq i32 %lftr.wideiv27, 100
br i1 %exitcond28, label %for.inc17, label %for.body8

for.inc17: ; preds = %for.inc14
%indvars.iv.next30 = add i64 %indvars.iv29, 1
%lftr.wideiv31 = trunc i64 %indvars.iv.next30 to i32
%exitcond32 = icmp eq i32 %lftr.wideiv31, 100
br i1 %exitcond32, label %for.exit, label %for.cond6.preheader

for.exit: ; preds = %for.inc17
ret i32 0
}