[Polly] Switch DT/LI in RegionGenerator for parallel subfn (#120413)

The patch #102460 already implements separate DT/LI/SE for parallel sub
function. Crashes have been reported while region generator tries using
oringinal function's DT while creating new parallel sub function due to
checks in #101198. This patch aims at fixing those cases by switching
the DT/LI while generating parallel function using Region Generator.

Fixes #117877
This commit is contained in:
Karthika Devi C 2025-01-08 13:59:50 +05:30 committed by GitHub
parent a8dab1aa03
commit 49668d5efe
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 49 additions and 6 deletions

View File

@ -632,7 +632,7 @@ public:
};
/// Generator for new versions of polyhedral region statements.
class RegionGenerator final : BlockGenerator {
class RegionGenerator final : public BlockGenerator {
public:
/// Create a generator for regions.
///

View File

@ -1000,7 +1000,7 @@ BasicBlock *RegionGenerator::repairDominance(BasicBlock *BB,
BasicBlock *BBCopyIDom = EndBlockMap.lookup(BBIDom);
if (BBCopyIDom)
DT.changeImmediateDominator(BBCopy, BBCopyIDom);
GenDT->changeImmediateDominator(BBCopy, BBCopyIDom);
return StartBlockMap.lookup(BBIDom);
}
@ -1069,8 +1069,8 @@ void RegionGenerator::copyStmt(ScopStmt &Stmt, LoopToScevMapT &LTS,
// Create a dedicated entry for the region where we can reload all demoted
// inputs.
BasicBlock *EntryBB = R->getEntry();
BasicBlock *EntryBBCopy = SplitBlock(Builder.GetInsertBlock(),
&*Builder.GetInsertPoint(), &DT, &LI);
BasicBlock *EntryBBCopy = SplitBlock(
Builder.GetInsertBlock(), &*Builder.GetInsertPoint(), GenDT, GenLI);
EntryBBCopy->setName("polly.stmt." + EntryBB->getName() + ".entry");
Builder.SetInsertPoint(&EntryBBCopy->front());
@ -1136,7 +1136,7 @@ void RegionGenerator::copyStmt(ScopStmt &Stmt, LoopToScevMapT &LTS,
// Now create a new dedicated region exit block and add it to the region map.
BasicBlock *ExitBBCopy = SplitBlock(Builder.GetInsertBlock(),
&*Builder.GetInsertPoint(), &DT, &LI);
&*Builder.GetInsertPoint(), GenDT, GenLI);
ExitBBCopy->setName("polly.stmt." + R->getExit()->getName() + ".exit");
StartBlockMap[R->getExit()] = ExitBBCopy;
EndBlockMap[R->getExit()] = ExitBBCopy;
@ -1145,7 +1145,7 @@ void RegionGenerator::copyStmt(ScopStmt &Stmt, LoopToScevMapT &LTS,
assert(ExitDomBBCopy &&
"Common exit dominator must be within region; at least the entry node "
"must match");
DT.changeImmediateDominator(ExitBBCopy, ExitDomBBCopy);
GenDT->changeImmediateDominator(ExitBBCopy, ExitDomBBCopy);
// As the block generator doesn't handle control flow we need to add the
// region control flow by hand after all blocks have been copied.

View File

@ -612,6 +612,7 @@ void IslNodeBuilder::createForParallel(__isl_take isl_ast_node *For) {
GenLI = SubLI;
GenSE = SubSE.get();
BlockGen.switchGeneratedFunc(SubFn, GenDT, GenLI, GenSE);
RegionGen.switchGeneratedFunc(SubFn, GenDT, GenLI, GenSE);
ExprBuilder.switchGeneratedFunc(SubFn, GenDT, GenLI, GenSE);
Builder.SetInsertPoint(&*LoopBody);
@ -681,6 +682,7 @@ void IslNodeBuilder::createForParallel(__isl_take isl_ast_node *For) {
IDToValue = std::move(IDToValueCopy);
ValueMap = std::move(CallerGlobals);
ExprBuilder.switchGeneratedFunc(CallerFn, CallerDT, CallerLI, CallerSE);
RegionGen.switchGeneratedFunc(CallerFn, CallerDT, CallerLI, CallerSE);
BlockGen.switchGeneratedFunc(CallerFn, CallerDT, CallerLI, CallerSE);
Builder.SetInsertPoint(&*AfterLoop);

View File

@ -0,0 +1,41 @@
; RUN: opt %loadNPMPolly -passes=polly-codegen -polly-parallel -S < %s | FileCheck %s
; CHECK: define ptr @ham(ptr %arg, i64 %arg1, i1 %arg2)
; This test is added to verify if the following IR does not crash on using different Dominator Tree when using polly parallel flag.
; ModuleID = '<stdin>'
source_filename = "<stdin>"
define ptr @ham(ptr %arg, i64 %arg1, i1 %arg2) {
bb:
br label %bb3
bb3: ; preds = %bb8, %bb
%phi = phi i64 [ 0, %bb ], [ %add9, %bb8 ]
%getelementptr = getelementptr [64 x i16], ptr %arg, i64 %phi
br label %bb4
bb4: ; preds = %bb7, %bb3
%phi5 = phi i64 [ %add, %bb7 ], [ 0, %bb3 ]
%load = load i16, ptr null, align 2
br i1 %arg2, label %bb7, label %bb6
bb6: ; preds = %bb4
store i16 0, ptr %getelementptr, align 2
br label %bb7
bb7: ; preds = %bb6, %bb4
%add = add i64 %phi5, 1
%icmp = icmp ne i64 %phi5, 64
br i1 %icmp, label %bb4, label %bb8
bb8: ; preds = %bb7
%add9 = add i64 %phi, 1
%icmp10 = icmp ult i64 %phi, %arg1
br i1 %icmp10, label %bb3, label %bb11
bb11: ; preds = %bb8
ret ptr null
}