mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-22 02:33:06 +01:00
Make dependency between certain analysis passes transitive (reapply)
LazyBlockFrequenceInfoPass, LazyBranchProbabilityInfoPass and LoopAccessLegacyAnalysis all cache pointers to their nestled required analysis passes. One need to use addRequiredTransitive to describe that the nestled passes can't be freed until those analysis passes no longer are used themselves. There is still a bit of a mess considering the getLazyBPIAnalysisUsage and getLazyBFIAnalysisUsage functions. Those functions are used from both Transform, CodeGen and Analysis passes. I figure it is OK to use addRequiredTransitive also when being used from Transform and CodeGen passes. On the other hand, I figure we must do it when used from other Analysis passes. So using addRequiredTransitive should be more correct here. An alternative solution would be to add a bool option in those functions to let the user tell if it is a analysis pass or not. Since those lazy passes will be obsolete when new PM has conquered the world I figure we can leave it like this right now. Intention with the patch is to fix PR49950. It at least solves the problem for the reproducer in PR49950. However, that reproducer need five passes in a specific order, so there are lots of various "solutions" that could avoid the crash without actually fixing the root cause. This is a reapply of commit 3655f0757f2b4b, that was reverted in 33ff3c20498ef5c2057 due to problems with assertions in the polly lit tests. That problem is supposed to be solved by also adjusting ScopPass to explicitly preserve LazyBlockFrequencyInfo and LazyBranchProbabilityInfo (it already preserved OptimizationRemarkEmitter which depends on those lazy passes). Differential Revision: https://reviews.llvm.org/D100958
This commit is contained in:
parent
6c8dead54f
commit
5908f3a456
@ -45,8 +45,8 @@ void LazyBlockFrequencyInfoPass::getAnalysisUsage(AnalysisUsage &AU) const {
|
||||
// We require DT so it's available when LI is available. The LI updating code
|
||||
// asserts that DT is also present so if we don't make sure that we have DT
|
||||
// here, that assert will trigger.
|
||||
AU.addRequired<DominatorTreeWrapperPass>();
|
||||
AU.addRequired<LoopInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<DominatorTreeWrapperPass>();
|
||||
AU.addRequiredTransitive<LoopInfoWrapperPass>();
|
||||
AU.setPreservesAll();
|
||||
}
|
||||
|
||||
@ -61,8 +61,8 @@ bool LazyBlockFrequencyInfoPass::runOnFunction(Function &F) {
|
||||
|
||||
void LazyBlockFrequencyInfoPass::getLazyBFIAnalysisUsage(AnalysisUsage &AU) {
|
||||
LazyBranchProbabilityInfoPass::getLazyBPIAnalysisUsage(AU);
|
||||
AU.addRequired<LazyBlockFrequencyInfoPass>();
|
||||
AU.addRequired<LoopInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<LazyBlockFrequencyInfoPass>();
|
||||
AU.addRequiredTransitive<LoopInfoWrapperPass>();
|
||||
}
|
||||
|
||||
void llvm::initializeLazyBFIPassPass(PassRegistry &Registry) {
|
||||
|
@ -46,9 +46,9 @@ void LazyBranchProbabilityInfoPass::getAnalysisUsage(AnalysisUsage &AU) const {
|
||||
// We require DT so it's available when LI is available. The LI updating code
|
||||
// asserts that DT is also present so if we don't make sure that we have DT
|
||||
// here, that assert will trigger.
|
||||
AU.addRequired<DominatorTreeWrapperPass>();
|
||||
AU.addRequired<LoopInfoWrapperPass>();
|
||||
AU.addRequired<TargetLibraryInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<DominatorTreeWrapperPass>();
|
||||
AU.addRequiredTransitive<LoopInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<TargetLibraryInfoWrapperPass>();
|
||||
AU.setPreservesAll();
|
||||
}
|
||||
|
||||
@ -63,9 +63,9 @@ bool LazyBranchProbabilityInfoPass::runOnFunction(Function &F) {
|
||||
}
|
||||
|
||||
void LazyBranchProbabilityInfoPass::getLazyBPIAnalysisUsage(AnalysisUsage &AU) {
|
||||
AU.addRequired<LazyBranchProbabilityInfoPass>();
|
||||
AU.addRequired<LoopInfoWrapperPass>();
|
||||
AU.addRequired<TargetLibraryInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<LazyBranchProbabilityInfoPass>();
|
||||
AU.addRequiredTransitive<LoopInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<TargetLibraryInfoWrapperPass>();
|
||||
}
|
||||
|
||||
void llvm::initializeLazyBPIPassPass(PassRegistry &Registry) {
|
||||
|
@ -2281,10 +2281,10 @@ bool LoopAccessLegacyAnalysis::runOnFunction(Function &F) {
|
||||
}
|
||||
|
||||
void LoopAccessLegacyAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
|
||||
AU.addRequired<ScalarEvolutionWrapperPass>();
|
||||
AU.addRequired<AAResultsWrapperPass>();
|
||||
AU.addRequired<DominatorTreeWrapperPass>();
|
||||
AU.addRequired<LoopInfoWrapperPass>();
|
||||
AU.addRequiredTransitive<ScalarEvolutionWrapperPass>();
|
||||
AU.addRequiredTransitive<AAResultsWrapperPass>();
|
||||
AU.addRequiredTransitive<DominatorTreeWrapperPass>();
|
||||
AU.addRequiredTransitive<LoopInfoWrapperPass>();
|
||||
|
||||
AU.setPreservesAll();
|
||||
}
|
||||
|
78
test/Other/pr49950.ll
Normal file
78
test/Other/pr49950.ll
Normal file
@ -0,0 +1,78 @@
|
||||
; RUN: opt < %s -o /dev/null -enable-new-pm=0 -block-freq -opt-remark-emitter -memoryssa -inject-tli-mappings -pgo-memop-opt -verify-loop-info -debug-pass=Details 2>&1 | FileCheck %s
|
||||
|
||||
; REQUIRES: asserts
|
||||
|
||||
; This is a heavily reduced reproducer for the problem found in
|
||||
; https://bugs.llvm.org/show_bug.cgi?id=49950 when doing fuzzy
|
||||
; testing (including non-standard pipelines).
|
||||
;
|
||||
; The problem manifested as having a pass structure like this
|
||||
; when it failed (as given by using -debug-pass=Details):
|
||||
;
|
||||
; Target Library Information
|
||||
; Target Transform Information
|
||||
; Profile summary info
|
||||
; Assumption Cache Tracker
|
||||
; ModulePass Manager
|
||||
; FunctionPass Manager
|
||||
; Dominator Tree Construction
|
||||
; Natural Loop Information
|
||||
; Post-Dominator Tree Construction
|
||||
; Branch Probability Analysis
|
||||
; Block Frequency Analysis
|
||||
; -- Branch Probability Analysis
|
||||
; Lazy Branch Probability Analysis
|
||||
; Lazy Block Frequency Analysis
|
||||
; Optimization Remark Emitter
|
||||
; Basic Alias Analysis (stateless AA impl)
|
||||
; Function Alias Analysis Results
|
||||
; Memory SSA
|
||||
; -- Dominator Tree Construction
|
||||
; -- Function Alias Analysis Results
|
||||
; -- Basic Alias Analysis (stateless AA impl)
|
||||
; -- Memory SSA
|
||||
; Inject TLI Mappings
|
||||
; -- Inject TLI Mappings
|
||||
; PGOMemOPSize
|
||||
; -- Block Frequency Analysis
|
||||
; -- Post-Dominator Tree Construction
|
||||
; -- Optimization Remark Emitter
|
||||
; -- Lazy Branch Probability Analysis
|
||||
; -- Natural Loop Information
|
||||
; -- Lazy Block Frequency Analysis
|
||||
; -- PGOMemOPSize
|
||||
; Module Verifier
|
||||
; -- Module Verifier
|
||||
; -- Target Library Information
|
||||
; -- Profile summary info
|
||||
; -- Assumption Cache Tracker
|
||||
; Bitcode Writer
|
||||
; -- Bitcode Writer
|
||||
;
|
||||
; One might notice that "Dominator Tree Construction" is dropped after
|
||||
; "Memory SSA", while for example "Natural Loop Information" stick around
|
||||
; a bit longer. This despite "Dominator Tree Construction" being transitively
|
||||
; required by "Natural Loop Information".
|
||||
; The end result was that we got crashes when doing verification of loop
|
||||
; info after "Inject TLI Mappings" (since the dominator tree had been
|
||||
; removed too early).
|
||||
|
||||
; Verify that both domintator tree and loop info are kept until after
|
||||
; PGOMemOPSize:
|
||||
;
|
||||
; CHECK: Dominator Tree Construction
|
||||
; CHECK-NOT: -- Dominator Tree Construction
|
||||
; CHECK: Memory SSA
|
||||
; CHECK-NOT: -- Dominator Tree Construction
|
||||
; CHECK: Inject TLI Mappings
|
||||
; CHECK-NOT: -- Dominator Tree Construction
|
||||
; CHECK: PGOMemOPSize
|
||||
; CHECK-DAG: -- Dominator Tree Construction
|
||||
; CHECK-DAG: -- Natural Loop Information
|
||||
; CHECK-DAG: -- PGOMemOPSize
|
||||
; CHECK: Bitcode Writer
|
||||
|
||||
define void @foo() {
|
||||
entry:
|
||||
ret void
|
||||
}
|
Loading…
Reference in New Issue
Block a user