[DFAJumpThreading] Use a single lazy DTU #162802
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
|
@llvm/pr-subscribers-llvm-transforms Author: Hongyu Chen (XChy) ChangesResolves #162240 (review) Full diff: https://github.com/llvm/llvm-project/pull/162802.diff 1 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
index f4e05a204c9cf..65e6cc9aa1a7b 100644
--- a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
+++ b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
@@ -148,19 +148,17 @@ class SelectInstToUnfold {
class DFAJumpThreading {
public:
- DFAJumpThreading(AssumptionCache *AC, DominatorTree *DT, LoopInfo *LI,
+ DFAJumpThreading(AssumptionCache *AC, DomTreeUpdater *DTU, LoopInfo *LI,
TargetTransformInfo *TTI, OptimizationRemarkEmitter *ORE)
- : AC(AC), DT(DT), LI(LI), TTI(TTI), ORE(ORE) {}
+ : AC(AC), DTU(DTU), LI(LI), TTI(TTI), ORE(ORE) {}
bool run(Function &F);
bool LoopInfoBroken;
private:
void
- unfoldSelectInstrs(DominatorTree *DT,
- const SmallVector<SelectInstToUnfold, 4> &SelectInsts) {
+ unfoldSelectInstrs(const SmallVector<SelectInstToUnfold, 4> &SelectInsts) {
// TODO: Have everything use a single lazy DTU
- DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Lazy);
SmallVector<SelectInstToUnfold, 4> Stack(SelectInsts);
while (!Stack.empty()) {
@@ -168,7 +166,7 @@ class DFAJumpThreading {
std::vector<SelectInstToUnfold> NewSIsToUnfold;
std::vector<BasicBlock *> NewBBs;
- unfold(&DTU, LI, SIToUnfold, &NewSIsToUnfold, &NewBBs);
+ unfold(DTU, LI, SIToUnfold, &NewSIsToUnfold, &NewBBs);
// Put newly discovered select instructions into the work list.
llvm::append_range(Stack, NewSIsToUnfold);
@@ -181,7 +179,7 @@ class DFAJumpThreading {
std::vector<BasicBlock *> *NewBBs);
AssumptionCache *AC;
- DominatorTree *DT;
+ DomTreeUpdater *DTU;
LoopInfo *LI;
TargetTransformInfo *TTI;
OptimizationRemarkEmitter *ORE;
@@ -869,11 +867,11 @@ struct AllSwitchPaths {
};
struct TransformDFA {
- TransformDFA(AllSwitchPaths *SwitchPaths, DominatorTree *DT,
+ TransformDFA(AllSwitchPaths *SwitchPaths, DomTreeUpdater *DTU,
AssumptionCache *AC, TargetTransformInfo *TTI,
OptimizationRemarkEmitter *ORE,
SmallPtrSet<const Value *, 32> EphValues)
- : SwitchPaths(SwitchPaths), DT(DT), AC(AC), TTI(TTI), ORE(ORE),
+ : SwitchPaths(SwitchPaths), DTU(DTU), AC(AC), TTI(TTI), ORE(ORE),
EphValues(EphValues) {}
bool run() {
@@ -1049,19 +1047,16 @@ struct TransformDFA {
SmallPtrSet<BasicBlock *, 16> BlocksToClean;
BlocksToClean.insert_range(successors(SwitchBlock));
- {
- DomTreeUpdater DTU(*DT, DomTreeUpdater::UpdateStrategy::Lazy);
- for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) {
- createExitPath(NewDefs, TPath, DuplicateMap, BlocksToClean, &DTU);
- NumPaths++;
- }
-
- // After all paths are cloned, now update the last successor of the cloned
- // path so it skips over the switch statement
- for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths())
- updateLastSuccessor(TPath, DuplicateMap, &DTU);
+ for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) {
+ createExitPath(NewDefs, TPath, DuplicateMap, BlocksToClean, DTU);
+ NumPaths++;
}
+ // After all paths are cloned, now update the last successor of the cloned
+ // path so it skips over the switch statement
+ for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths())
+ updateLastSuccessor(TPath, DuplicateMap, DTU);
+
// For each instruction that was cloned and used outside, update its uses
updateSSA(NewDefs);
@@ -1165,7 +1160,7 @@ struct TransformDFA {
}
// SSAUpdater handles phi placement and renaming uses with the appropriate
// value.
- SSAUpdate.RewriteAllUses(DT);
+ SSAUpdate.RewriteAllUses(&DTU->getDomTree());
}
/// Clones a basic block, and adds it to the CFG.
@@ -1388,7 +1383,7 @@ struct TransformDFA {
}
AllSwitchPaths *SwitchPaths;
- DominatorTree *DT;
+ DomTreeUpdater *DTU;
AssumptionCache *AC;
TargetTransformInfo *TTI;
OptimizationRemarkEmitter *ORE;
@@ -1431,7 +1426,7 @@ bool DFAJumpThreading::run(Function &F) {
<< "candidate for jump threading\n");
LLVM_DEBUG(SI->dump());
- unfoldSelectInstrs(DT, Switch.getSelectInsts());
+ unfoldSelectInstrs(Switch.getSelectInsts());
if (!Switch.getSelectInsts().empty())
MadeChanges = true;
@@ -1453,7 +1448,7 @@ bool DFAJumpThreading::run(Function &F) {
}
#ifdef NDEBUG
- LI->verify(*DT);
+ LI->verify(DTU->getDomTree());
#endif
SmallPtrSet<const Value *, 32> EphValues;
@@ -1461,13 +1456,15 @@ bool DFAJumpThreading::run(Function &F) {
CodeMetrics::collectEphemeralValues(&F, AC, EphValues);
for (AllSwitchPaths SwitchPaths : ThreadableLoops) {
- TransformDFA Transform(&SwitchPaths, DT, AC, TTI, ORE, EphValues);
+ TransformDFA Transform(&SwitchPaths, DTU, AC, TTI, ORE, EphValues);
if (Transform.run())
MadeChanges = LoopInfoBroken = true;
}
+ DTU->flush();
+
#ifdef EXPENSIVE_CHECKS
- assert(DT->verify(DominatorTree::VerificationLevel::Full));
+ assert(DTU->getDomTree().verify(DominatorTree::VerificationLevel::Full));
verifyFunction(F, &dbgs());
#endif
@@ -1482,7 +1479,9 @@ PreservedAnalyses DFAJumpThreadingPass::run(Function &F,
LoopInfo &LI = AM.getResult<LoopAnalysis>(F);
TargetTransformInfo &TTI = AM.getResult<TargetIRAnalysis>(F);
OptimizationRemarkEmitter ORE(&F);
- DFAJumpThreading ThreadImpl(&AC, &DT, &LI, &TTI, &ORE);
+
+ DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Lazy);
+ DFAJumpThreading ThreadImpl(&AC, &DTU, &LI, &TTI, &ORE);
if (!ThreadImpl.run(F))
return PreservedAnalyses::all();
|
|
Hi @XChy The following starts failing with this patch: |
|
Another crash like this: Result: |
|
This patch seems not to be blamed. It looks like we insert an additional edge to domtree during unfolding selects, though I am on a vacation recently and have no chance to debug it now. |
Well, both reproducers I attached at least starts crashing with this patch. |
That's OK for bisection, as this patch exposes the problems. Thanks for the reports anyway. I will fix them once I have a chance. |
DharuniRAcharya
pushed a commit
to DharuniRAcharya/llvm-project
that referenced
this pull request
Oct 13, 2025
UsmanNadeem
added a commit
that referenced
this pull request
Oct 14, 2025
…ing (#163296) The edge `StartBlock -> EndBlock` already exists before unfolding. The instructions for `applyUpdates()` say that you are supposed not to insert an existing edge. Fixes issues reported by @mikaelholmen in #162802
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Oct 14, 2025
…hile unfolding (#163296) The edge `StartBlock -> EndBlock` already exists before unfolding. The instructions for `applyUpdates()` say that you are supposed not to insert an existing edge. Fixes issues reported by @mikaelholmen in llvm/llvm-project#162802
akadutta
pushed a commit
to akadutta/llvm-project
that referenced
this pull request
Oct 14, 2025
akadutta
pushed a commit
to akadutta/llvm-project
that referenced
this pull request
Oct 14, 2025
…ing (llvm#163296) The edge `StartBlock -> EndBlock` already exists before unfolding. The instructions for `applyUpdates()` say that you are supposed not to insert an existing edge. Fixes issues reported by @mikaelholmen in llvm#162802
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Resolves #162240 (review)