[AMDGPU][UnifyLoopExits] Fix duplicate successor handling #170759
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Robert Imschweiler (ro-i) ChangesFixes #170730 Maybe I can also come up with a more elegant solution. But I think it wouldn't make a lot of sense trying to avoid multiple "fake" target blocks. Would require a mapping between original and fake target blocks... Full diff: https://github.com/llvm/llvm-project/pull/170759.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/BasicBlockUtils.cpp b/llvm/lib/Transforms/Utils/BasicBlockUtils.cpp
index da3e60da9df74..b0c04086f5e84 100644
--- a/llvm/lib/Transforms/Utils/BasicBlockUtils.cpp
+++ b/llvm/lib/Transforms/Utils/BasicBlockUtils.cpp
@@ -755,9 +755,11 @@ BasicBlock *llvm::SplitCallBrEdge(BasicBlock *CallBrBlock, BasicBlock *Succ,
updateCycleLoopInfo<CycleInfo, Cycle>(CI, CallBrBlock, CallBrTarget, Succ);
if (DTU) {
DTU->applyUpdates({{DominatorTree::Insert, CallBrBlock, CallBrTarget}});
- if (DTU->getDomTree().dominates(CallBrBlock, Succ))
- DTU->applyUpdates({{DominatorTree::Delete, CallBrBlock, Succ},
- {DominatorTree::Insert, CallBrTarget, Succ}});
+ if (DTU->getDomTree().dominates(CallBrBlock, Succ)) {
+ if (!is_contained(successors(CallBrBlock), Succ))
+ DTU->applyUpdates({{DominatorTree::Delete, CallBrBlock, Succ}});
+ DTU->applyUpdates({{DominatorTree::Insert, CallBrTarget, Succ}});
+ }
}
return CallBrTarget;
diff --git a/llvm/test/Transforms/UnifyLoopExits/basic.ll b/llvm/test/Transforms/UnifyLoopExits/basic.ll
index d04d142f196d3..a1c4814883c93 100644
--- a/llvm/test/Transforms/UnifyLoopExits/basic.ll
+++ b/llvm/test/Transforms/UnifyLoopExits/basic.ll
@@ -61,13 +61,13 @@ define void @loop_1_callbr(i1 %PredEntry, i1 %PredB, i1 %PredC, i1 %PredD) {
; CHECK-NEXT: br label [[B:%.*]]
; CHECK: B:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDB:%.*]])
-; CHECK-NEXT: to label [[C:%.*]] [label %B.target.E]
+; CHECK-NEXT: to label [[C:%.*]] [label [[B_TARGET_E:%.*]]]
; CHECK: C:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDC:%.*]])
-; CHECK-NEXT: to label [[D:%.*]] [label %C.target.F]
+; CHECK-NEXT: to label [[D:%.*]] [label [[C_TARGET_F:%.*]]]
; CHECK: D:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDD:%.*]])
-; CHECK-NEXT: to label [[A]] [label %D.target.F]
+; CHECK-NEXT: to label [[A]] [label [[D_TARGET_F:%.*]]]
; CHECK: E:
; CHECK-NEXT: br label [[EXIT:%.*]]
; CHECK: F:
@@ -83,7 +83,7 @@ define void @loop_1_callbr(i1 %PredEntry, i1 %PredB, i1 %PredC, i1 %PredD) {
; CHECK: D.target.F:
; CHECK-NEXT: br label [[LOOP_EXIT_GUARD]]
; CHECK: loop.exit.guard:
-; CHECK-NEXT: [[GUARD_X:%.*]] = phi i1 [ true, [[B_TARGET_E:%.*]] ], [ false, [[C_TARGET_F:%.*]] ], [ false, [[D_TARGET_F:%.*]] ]
+; CHECK-NEXT: [[GUARD_X:%.*]] = phi i1 [ true, [[B_TARGET_E]] ], [ false, [[C_TARGET_F]] ], [ false, [[D_TARGET_F]] ]
; CHECK-NEXT: br i1 [[GUARD_X]], label [[X:%.*]], label [[Y]]
;
entry:
@@ -175,13 +175,13 @@ define void @loop_2_callbr(i1 %PredA, i1 %PredB, i1 %PredC) {
; CHECK-NEXT: br label [[A:%.*]]
; CHECK: A:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDA:%.*]])
-; CHECK-NEXT: to label [[B:%.*]] [label %A.target.X]
+; CHECK-NEXT: to label [[B:%.*]] [label [[A_TARGET_X:%.*]]]
; CHECK: B:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDB:%.*]])
-; CHECK-NEXT: to label [[C:%.*]] [label %B.target.Y]
+; CHECK-NEXT: to label [[C:%.*]] [label [[B_TARGET_Y:%.*]]]
; CHECK: C:
; CHECK-NEXT: callbr void asm "", "r,!i"(i1 [[PREDC:%.*]])
-; CHECK-NEXT: to label [[D:%.*]] [label %C.target.Z]
+; CHECK-NEXT: to label [[D:%.*]] [label [[C_TARGET_Z:%.*]]]
; CHECK: D:
; CHECK-NEXT: br label [[A]]
; CHECK: X:
@@ -199,7 +199,7 @@ define void @loop_2_callbr(i1 %PredA, i1 %PredB, i1 %PredC) {
; CHECK: C.target.Z:
; CHECK-NEXT: br label [[LOOP_EXIT_GUARD]]
; CHECK: loop.exit.guard:
-; CHECK-NEXT: [[GUARD_X:%.*]] = phi i1 [ true, [[A_TARGET_X:%.*]] ], [ false, [[B_TARGET_Y:%.*]] ], [ false, [[C_TARGET_Z:%.*]] ]
+; CHECK-NEXT: [[GUARD_X:%.*]] = phi i1 [ true, [[A_TARGET_X]] ], [ false, [[B_TARGET_Y]] ], [ false, [[C_TARGET_Z]] ]
; CHECK-NEXT: [[GUARD_Y:%.*]] = phi i1 [ false, [[A_TARGET_X]] ], [ true, [[B_TARGET_Y]] ], [ false, [[C_TARGET_Z]] ]
; CHECK-NEXT: br i1 [[GUARD_X]], label [[X:%.*]], label [[LOOP_EXIT_GUARD1:%.*]]
; CHECK: loop.exit.guard1:
@@ -232,3 +232,30 @@ Z:
exit:
ret void
}
+
+; Test that UnifyLoopExits handles callbr with duplicate successors correctly.
+; The exit block appears twice as a successor of the callbr instruction.
+define void @callbr_duplicate_successors() {
+; CHECK-LABEL: @callbr_duplicate_successors(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: callbr void asm sideeffect "", "!i,!i"()
+; CHECK-NEXT: to label [[LOOP_TARGET_EXIT:%.*]] [label [[LOOP]], label [[LOOP_TARGET_EXIT1:%.*]]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+; CHECK: loop.target.exit:
+; CHECK-NEXT: br label [[EXIT:%.*]]
+; CHECK: loop.target.exit1:
+; CHECK-NEXT: br label [[EXIT]]
+;
+entry:
+ br label %loop
+
+loop:
+ callbr void asm sideeffect "", "!i,!i"()
+ to label %exit [label %loop, label %exit]
+
+exit:
+ ret void
+}
|
arsenm
changed the title
Member
|
Reverse ping. If you don't mind, could I merge it? |
arsenm
reviewed
Dec 8, 2025
| DTU->applyUpdates({{DominatorTree::Delete, CallBrBlock, Succ}, | ||
| {DominatorTree::Insert, CallBrTarget, Succ}}); | ||
| if (DTU->getDomTree().dominates(CallBrBlock, Succ)) { | ||
| if (!is_contained(successors(CallBrBlock), Succ)) |
Contributor
There was a problem hiding this comment.
I thought DomTree updater handled this for you?
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/169/builds/17809 Here is the relevant piece of the build log for the reference |
Contributor
Author
|
(failure seems unrelated, next run is green again https://lab.llvm.org/buildbot/#/builders/169/builds/17810) |
honeygoyal
pushed a commit
to honeygoyal/llvm-project
that referenced
this pull request
Dec 9, 2025
Fixes llvm#170730 Maybe I can also come up with a more elegant solution. But I think it wouldn't make a lot of sense trying to avoid multiple "fake" target blocks. Would require a mapping between original and fake target blocks...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #170730
Maybe I can also come up with a more elegant solution. But I think it wouldn't make a lot of sense trying to avoid multiple "fake" target blocks. Would require a mapping between original and fake target blocks...