[Coro] Use DebugInfoCache to speed up cloning in CoroSplitPass #118630
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
llvmbot
added
coroutines
artempyanykh
added a commit
that referenced
this pull request
Dec 4, 2024
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of global debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt GlobalDI | Cached CU DIFinder (cur.) | |-----------------+----------+----------------+-------------------+---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | |-----------------+----------+----------------+-------------------+---------------------------| | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
4bc420c to
bec4d02
Compare
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
784c570 to
bdb7970
Compare
This was referenced Dec 4, 2024
|
@llvm/pr-subscribers-debuginfo @llvm/pr-subscribers-llvm-transforms Author: Artem Pianykh (artempyanykh) Changes[Coro] Use DebugInfoCache to speed up cloning in CoroSplitPass Summary: The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample Test Plan: Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. Full diff: https://github.com/llvm/llvm-project/pull/118630.diff 14 Files Affected:
diff --git a/llvm/include/llvm/Transforms/Coroutines/ABI.h b/llvm/include/llvm/Transforms/Coroutines/ABI.h
index 0b2d405f3caec4..2cf614b6bb1e2a 100644
--- a/llvm/include/llvm/Transforms/Coroutines/ABI.h
+++ b/llvm/include/llvm/Transforms/Coroutines/ABI.h
@@ -15,6 +15,7 @@
#ifndef LLVM_TRANSFORMS_COROUTINES_ABI_H
#define LLVM_TRANSFORMS_COROUTINES_ABI_H
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Transforms/Coroutines/CoroShape.h"
#include "llvm/Transforms/Coroutines/MaterializationUtils.h"
@@ -53,7 +54,8 @@ class BaseABI {
// Perform the function splitting according to the ABI.
virtual void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) = 0;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) = 0;
Function &F;
coro::Shape &Shape;
@@ -73,7 +75,8 @@ class SwitchABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
class AsyncABI : public BaseABI {
@@ -86,7 +89,8 @@ class AsyncABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
class AnyRetconABI : public BaseABI {
@@ -99,7 +103,8 @@ class AnyRetconABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
} // end namespace coro
diff --git a/llvm/lib/Analysis/CGSCCPassManager.cpp b/llvm/lib/Analysis/CGSCCPassManager.cpp
index 948bc2435ab275..3ba085cdb0be8b 100644
--- a/llvm/lib/Analysis/CGSCCPassManager.cpp
+++ b/llvm/lib/Analysis/CGSCCPassManager.cpp
@@ -14,6 +14,7 @@
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/iterator_range.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/InstIterator.h"
@@ -139,6 +140,11 @@ ModuleToPostOrderCGSCCPassAdaptor::run(Module &M, ModuleAnalysisManager &AM) {
// Get the call graph for this module.
LazyCallGraph &CG = AM.getResult<LazyCallGraphAnalysis>(M);
+ // Prime DebugInfoCache.
+ // TODO: Currently, the only user is CoroSplitPass. Consider running
+ // conditionally.
+ AM.getResult<DebugInfoCacheAnalysis>(M);
+
// Get Function analysis manager from its proxy.
FunctionAnalysisManager &FAM =
AM.getCachedResult<FunctionAnalysisManagerModuleProxy>(M)->getManager();
@@ -350,6 +356,7 @@ ModuleToPostOrderCGSCCPassAdaptor::run(Module &M, ModuleAnalysisManager &AM) {
// analysis proxies by handling them above and in any nested pass managers.
PA.preserveSet<AllAnalysesOn<LazyCallGraph::SCC>>();
PA.preserve<LazyCallGraphAnalysis>();
+ PA.preserve<DebugInfoCacheAnalysis>();
PA.preserve<CGSCCAnalysisManagerModuleProxy>();
PA.preserve<FunctionAnalysisManagerModuleProxy>();
return PA;
diff --git a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
index 2803b340bd22e0..96e531d498d722 100644
--- a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
+++ b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
@@ -32,6 +32,7 @@
#include "llvm/Analysis/CFG.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/Analysis/OptimizationRemarkEmitter.h"
#include "llvm/Analysis/TargetTransformInfo.h"
@@ -79,15 +80,39 @@ using namespace llvm;
#define DEBUG_TYPE "coro-split"
namespace {
+const DebugInfoFinder *cachedDIFinder(Function &F,
+ const DebugInfoCache *DICache) {
+ if (!DICache)
+ return nullptr;
+
+ auto *SP = F.getSubprogram();
+ auto *CU = SP ? SP->getUnit() : nullptr;
+ if (!CU)
+ return nullptr;
+
+ auto Found = DICache->Result.find(CU);
+ if (Found == DICache->Result.end())
+ return nullptr;
+
+ return &Found->getSecond();
+}
+
/// Collect (a known) subset of global debug info metadata potentially used by
/// the function \p F.
///
/// This metadata set can be used to avoid cloning debug info not owned by \p F
/// and is shared among all potential clones \p F.
-void collectGlobalDebugInfo(Function &F, MetadataSetTy &GlobalDebugInfo) {
+void collectGlobalDebugInfo(Function &F, MetadataSetTy &GlobalDebugInfo,
+ const DebugInfoCache *DICache) {
TimeTraceScope FunctionScope("CollectGlobalDebugInfo");
DebugInfoFinder DIFinder;
+
+ // Copy DIFinder from cache which is primed on F's compile unit when available
+ auto *PrimedDIFinder = cachedDIFinder(F, DICache);
+ if (PrimedDIFinder)
+ DIFinder = *PrimedDIFinder;
+
DISubprogram *SPClonedWithinModule = CollectDebugInfoForCloning(
F, CloneFunctionChangeType::LocalChangesOnly, DIFinder);
@@ -1394,11 +1419,11 @@ namespace {
struct SwitchCoroutineSplitter {
static void split(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI, const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Switch);
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
// Create a resume clone by cloning the body of the original function,
// setting new entry block and replacing coro.suspend an appropriate value
@@ -1712,7 +1737,8 @@ CallInst *coro::createMustTailCall(DebugLoc Loc, Function *MustTailCallFn,
void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Async);
assert(Clones.empty());
// Reset various things that the optimizer might have decided it
@@ -1799,7 +1825,7 @@ void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
assert(Clones.size() == Shape.CoroSuspends.size());
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto *Suspend = CS;
@@ -1812,7 +1838,8 @@ void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
void coro::AnyRetconABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Retcon || Shape.ABI == coro::ABI::RetconOnce);
assert(Clones.empty());
@@ -1934,7 +1961,7 @@ void coro::AnyRetconABI::splitCoroutine(Function &F, coro::Shape &Shape,
assert(Clones.size() == Shape.CoroSuspends.size());
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto Suspend = CS;
@@ -1988,13 +2015,15 @@ static bool hasSafeElideCaller(Function &F) {
void coro::SwitchABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
- SwitchCoroutineSplitter::split(F, Shape, Clones, TTI);
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
+ SwitchCoroutineSplitter::split(F, Shape, Clones, TTI, DICache);
}
static void doSplitCoroutine(Function &F, SmallVectorImpl<Function *> &Clones,
coro::BaseABI &ABI, TargetTransformInfo &TTI,
- bool OptimizeFrame) {
+ bool OptimizeFrame,
+ const DebugInfoCache *DICache) {
PrettyStackTraceFunction prettyStackTrace(F);
auto &Shape = ABI.Shape;
@@ -2019,7 +2048,7 @@ static void doSplitCoroutine(Function &F, SmallVectorImpl<Function *> &Clones,
if (isNoSuspendCoroutine) {
handleNoSuspendCoroutine(Shape);
} else {
- ABI.splitCoroutine(F, Shape, Clones, TTI);
+ ABI.splitCoroutine(F, Shape, Clones, TTI, DICache);
}
// Replace all the swifterror operations in the original function.
@@ -2216,6 +2245,9 @@ PreservedAnalyses CoroSplitPass::run(LazyCallGraph::SCC &C,
auto &FAM =
AM.getResult<FunctionAnalysisManagerCGSCCProxy>(C, CG).getManager();
+ const auto &MAMProxy = AM.getResult<ModuleAnalysisManagerCGSCCProxy>(C, CG);
+ const auto *DICache = MAMProxy.getCachedResult<DebugInfoCacheAnalysis>(M);
+
// Check for uses of llvm.coro.prepare.retcon/async.
SmallVector<Function *, 2> PrepareFns;
addPrepareFunction(M, PrepareFns, "llvm.coro.prepare.retcon");
@@ -2252,7 +2284,7 @@ PreservedAnalyses CoroSplitPass::run(LazyCallGraph::SCC &C,
SmallVector<Function *, 4> Clones;
auto &TTI = FAM.getResult<TargetIRAnalysis>(F);
- doSplitCoroutine(F, Clones, *ABI, TTI, OptimizeFrame);
+ doSplitCoroutine(F, Clones, *ABI, TTI, OptimizeFrame, DICache);
CurrentSCC = &updateCallGraphAfterCoroutineSplit(
*N, Shape, Clones, *CurrentSCC, CG, AM, UR, FAM);
diff --git a/llvm/test/Other/new-pass-manager.ll b/llvm/test/Other/new-pass-manager.ll
index f0fe708806f1b6..53fd6fe2a317ec 100644
--- a/llvm/test/Other/new-pass-manager.ll
+++ b/llvm/test/Other/new-pass-manager.ll
@@ -23,6 +23,7 @@
; CHECK-CGSCC-PASS-NEXT: Running analysis: InnerAnalysisManagerProxy<{{.*(FunctionAnalysisManager|AnalysisManager<.*Function.*>).*}},{{.*}}Module>
; CHECK-CGSCC-PASS-NEXT: Running analysis: LazyCallGraphAnalysis
; CHECK-CGSCC-PASS-NEXT: Running analysis: TargetLibraryAnalysis
+; CHECK-CGSCC-PASS-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-CGSCC-PASS-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-CGSCC-PASS-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-CGSCC-PASS-NEXT: Running pass: NoOpCGSCCPass
diff --git a/llvm/test/Other/new-pm-defaults.ll b/llvm/test/Other/new-pm-defaults.ll
index 7cf035b0c6f376..19bfec3ab718e0 100644
--- a/llvm/test/Other/new-pm-defaults.ll
+++ b/llvm/test/Other/new-pm-defaults.ll
@@ -139,6 +139,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-lto-defaults.ll b/llvm/test/Other/new-pm-lto-defaults.ll
index f788db1e338a1e..8f4fa763b52098 100644
--- a/llvm/test/Other/new-pm-lto-defaults.ll
+++ b/llvm/test/Other/new-pm-lto-defaults.ll
@@ -43,6 +43,7 @@
; CHECK-O23SZ-NEXT: Running analysis: OptimizationRemarkEmitterAnalysis
; CHECK-O23SZ-NEXT: Running analysis: InnerAnalysisManagerProxy<{{.*}}SCC
; CHECK-O23SZ-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O23SZ-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O23SZ-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O23SZ-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph{{.*}}>
; CHECK-O23SZ-NEXT: Running pass: PostOrderFunctionAttrsPass
diff --git a/llvm/test/Other/new-pm-pgo-preinline.ll b/llvm/test/Other/new-pm-pgo-preinline.ll
index f07a3728ba3d48..97813bb2433642 100644
--- a/llvm/test/Other/new-pm-pgo-preinline.ll
+++ b/llvm/test/Other/new-pm-pgo-preinline.ll
@@ -5,6 +5,7 @@
; CHECK-Osz-NEXT: Running analysis: InlineAdvisorAnalysis
; CHECK-Osz-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-Osz-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-Osz-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-Osz-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy on (foo)
; CHECK-Osz-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-Osz-NEXT: Running pass: InlinerPass on (foo)
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
index ed13402e1c4b15..e1ad86015fda95 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
@@ -74,6 +74,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
index c82c34f7ff01e7..3f6c5351e8e8e0 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
@@ -62,6 +62,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
index d375747547d61f..371dde305b0990 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
@@ -71,6 +71,7 @@
; CHECK-O-NEXT: Invalidating analysis: AAManager
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
index 5aacd26def2be5..860fb99525030f 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
@@ -106,6 +106,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
index f6a94065968038..d97cc97169b56b 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
@@ -62,6 +62,7 @@
; CHECK-O-NEXT: Running analysis: InlineAdvisorAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy on (foo)
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: InlinerPass on (foo)
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
index 48a9433d249996..d338817d076463 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
@@ -76,6 +76,7 @@
; CHECK-O-NEXT: Invalidating analysis: AAManager
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp b/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
index 5c71bc8063d6c9..7212107d992638 100644
--- a/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
+++ b/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
@@ -7,6 +7,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/CGSCCPassManager.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/AsmParser/Parser.h"
@@ -16,8 +17,8 @@
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
-#include "llvm/IR/PassManager.h"
#include "llvm/IR/PassInstrumentation.h"
+#include "llvm/IR/PassManager.h"
#include "llvm/IR/Verifier.h"
#include "llvm/Support/SourceMgr.h"
#include "llvm/Transforms/Utils/CallGraphUpdater.h"
@@ -255,6 +256,7 @@ class CGSCCPassManagerTest : public ::testing::Test {
"}\n")) {
FAM.registerPass([&] { return TargetLibraryAnalysis(); });
MAM.registerPass([&] { return LazyCallGraphAnalysis(); });
+ MAM.registerPass([&] { return DebugInfoCacheAnalysis(); });
MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); });
// Register required pass instrumentation analysis.
|
|
@llvm/pr-subscribers-coroutines Author: Artem Pianykh (artempyanykh) Changes[Coro] Use DebugInfoCache to speed up cloning in CoroSplitPass Summary: The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample Test Plan: Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. Full diff: https://github.com/llvm/llvm-project/pull/118630.diff 14 Files Affected:
diff --git a/llvm/include/llvm/Transforms/Coroutines/ABI.h b/llvm/include/llvm/Transforms/Coroutines/ABI.h
index 0b2d405f3caec4..2cf614b6bb1e2a 100644
--- a/llvm/include/llvm/Transforms/Coroutines/ABI.h
+++ b/llvm/include/llvm/Transforms/Coroutines/ABI.h
@@ -15,6 +15,7 @@
#ifndef LLVM_TRANSFORMS_COROUTINES_ABI_H
#define LLVM_TRANSFORMS_COROUTINES_ABI_H
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Transforms/Coroutines/CoroShape.h"
#include "llvm/Transforms/Coroutines/MaterializationUtils.h"
@@ -53,7 +54,8 @@ class BaseABI {
// Perform the function splitting according to the ABI.
virtual void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) = 0;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) = 0;
Function &F;
coro::Shape &Shape;
@@ -73,7 +75,8 @@ class SwitchABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
class AsyncABI : public BaseABI {
@@ -86,7 +89,8 @@ class AsyncABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
class AnyRetconABI : public BaseABI {
@@ -99,7 +103,8 @@ class AnyRetconABI : public BaseABI {
void splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) override;
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) override;
};
} // end namespace coro
diff --git a/llvm/lib/Analysis/CGSCCPassManager.cpp b/llvm/lib/Analysis/CGSCCPassManager.cpp
index 948bc2435ab275..3ba085cdb0be8b 100644
--- a/llvm/lib/Analysis/CGSCCPassManager.cpp
+++ b/llvm/lib/Analysis/CGSCCPassManager.cpp
@@ -14,6 +14,7 @@
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/iterator_range.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/InstIterator.h"
@@ -139,6 +140,11 @@ ModuleToPostOrderCGSCCPassAdaptor::run(Module &M, ModuleAnalysisManager &AM) {
// Get the call graph for this module.
LazyCallGraph &CG = AM.getResult<LazyCallGraphAnalysis>(M);
+ // Prime DebugInfoCache.
+ // TODO: Currently, the only user is CoroSplitPass. Consider running
+ // conditionally.
+ AM.getResult<DebugInfoCacheAnalysis>(M);
+
// Get Function analysis manager from its proxy.
FunctionAnalysisManager &FAM =
AM.getCachedResult<FunctionAnalysisManagerModuleProxy>(M)->getManager();
@@ -350,6 +356,7 @@ ModuleToPostOrderCGSCCPassAdaptor::run(Module &M, ModuleAnalysisManager &AM) {
// analysis proxies by handling them above and in any nested pass managers.
PA.preserveSet<AllAnalysesOn<LazyCallGraph::SCC>>();
PA.preserve<LazyCallGraphAnalysis>();
+ PA.preserve<DebugInfoCacheAnalysis>();
PA.preserve<CGSCCAnalysisManagerModuleProxy>();
PA.preserve<FunctionAnalysisManagerModuleProxy>();
return PA;
diff --git a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
index 2803b340bd22e0..96e531d498d722 100644
--- a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
+++ b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
@@ -32,6 +32,7 @@
#include "llvm/Analysis/CFG.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/Analysis/OptimizationRemarkEmitter.h"
#include "llvm/Analysis/TargetTransformInfo.h"
@@ -79,15 +80,39 @@ using namespace llvm;
#define DEBUG_TYPE "coro-split"
namespace {
+const DebugInfoFinder *cachedDIFinder(Function &F,
+ const DebugInfoCache *DICache) {
+ if (!DICache)
+ return nullptr;
+
+ auto *SP = F.getSubprogram();
+ auto *CU = SP ? SP->getUnit() : nullptr;
+ if (!CU)
+ return nullptr;
+
+ auto Found = DICache->Result.find(CU);
+ if (Found == DICache->Result.end())
+ return nullptr;
+
+ return &Found->getSecond();
+}
+
/// Collect (a known) subset of global debug info metadata potentially used by
/// the function \p F.
///
/// This metadata set can be used to avoid cloning debug info not owned by \p F
/// and is shared among all potential clones \p F.
-void collectGlobalDebugInfo(Function &F, MetadataSetTy &GlobalDebugInfo) {
+void collectGlobalDebugInfo(Function &F, MetadataSetTy &GlobalDebugInfo,
+ const DebugInfoCache *DICache) {
TimeTraceScope FunctionScope("CollectGlobalDebugInfo");
DebugInfoFinder DIFinder;
+
+ // Copy DIFinder from cache which is primed on F's compile unit when available
+ auto *PrimedDIFinder = cachedDIFinder(F, DICache);
+ if (PrimedDIFinder)
+ DIFinder = *PrimedDIFinder;
+
DISubprogram *SPClonedWithinModule = CollectDebugInfoForCloning(
F, CloneFunctionChangeType::LocalChangesOnly, DIFinder);
@@ -1394,11 +1419,11 @@ namespace {
struct SwitchCoroutineSplitter {
static void split(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI, const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Switch);
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
// Create a resume clone by cloning the body of the original function,
// setting new entry block and replacing coro.suspend an appropriate value
@@ -1712,7 +1737,8 @@ CallInst *coro::createMustTailCall(DebugLoc Loc, Function *MustTailCallFn,
void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Async);
assert(Clones.empty());
// Reset various things that the optimizer might have decided it
@@ -1799,7 +1825,7 @@ void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
assert(Clones.size() == Shape.CoroSuspends.size());
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto *Suspend = CS;
@@ -1812,7 +1838,8 @@ void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
void coro::AnyRetconABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
assert(Shape.ABI == coro::ABI::Retcon || Shape.ABI == coro::ABI::RetconOnce);
assert(Clones.empty());
@@ -1934,7 +1961,7 @@ void coro::AnyRetconABI::splitCoroutine(Function &F, coro::Shape &Shape,
assert(Clones.size() == Shape.CoroSuspends.size());
MetadataSetTy GlobalDebugInfo;
- collectGlobalDebugInfo(F, GlobalDebugInfo);
+ collectGlobalDebugInfo(F, GlobalDebugInfo, DICache);
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto Suspend = CS;
@@ -1988,13 +2015,15 @@ static bool hasSafeElideCaller(Function &F) {
void coro::SwitchABI::splitCoroutine(Function &F, coro::Shape &Shape,
SmallVectorImpl<Function *> &Clones,
- TargetTransformInfo &TTI) {
- SwitchCoroutineSplitter::split(F, Shape, Clones, TTI);
+ TargetTransformInfo &TTI,
+ const DebugInfoCache *DICache) {
+ SwitchCoroutineSplitter::split(F, Shape, Clones, TTI, DICache);
}
static void doSplitCoroutine(Function &F, SmallVectorImpl<Function *> &Clones,
coro::BaseABI &ABI, TargetTransformInfo &TTI,
- bool OptimizeFrame) {
+ bool OptimizeFrame,
+ const DebugInfoCache *DICache) {
PrettyStackTraceFunction prettyStackTrace(F);
auto &Shape = ABI.Shape;
@@ -2019,7 +2048,7 @@ static void doSplitCoroutine(Function &F, SmallVectorImpl<Function *> &Clones,
if (isNoSuspendCoroutine) {
handleNoSuspendCoroutine(Shape);
} else {
- ABI.splitCoroutine(F, Shape, Clones, TTI);
+ ABI.splitCoroutine(F, Shape, Clones, TTI, DICache);
}
// Replace all the swifterror operations in the original function.
@@ -2216,6 +2245,9 @@ PreservedAnalyses CoroSplitPass::run(LazyCallGraph::SCC &C,
auto &FAM =
AM.getResult<FunctionAnalysisManagerCGSCCProxy>(C, CG).getManager();
+ const auto &MAMProxy = AM.getResult<ModuleAnalysisManagerCGSCCProxy>(C, CG);
+ const auto *DICache = MAMProxy.getCachedResult<DebugInfoCacheAnalysis>(M);
+
// Check for uses of llvm.coro.prepare.retcon/async.
SmallVector<Function *, 2> PrepareFns;
addPrepareFunction(M, PrepareFns, "llvm.coro.prepare.retcon");
@@ -2252,7 +2284,7 @@ PreservedAnalyses CoroSplitPass::run(LazyCallGraph::SCC &C,
SmallVector<Function *, 4> Clones;
auto &TTI = FAM.getResult<TargetIRAnalysis>(F);
- doSplitCoroutine(F, Clones, *ABI, TTI, OptimizeFrame);
+ doSplitCoroutine(F, Clones, *ABI, TTI, OptimizeFrame, DICache);
CurrentSCC = &updateCallGraphAfterCoroutineSplit(
*N, Shape, Clones, *CurrentSCC, CG, AM, UR, FAM);
diff --git a/llvm/test/Other/new-pass-manager.ll b/llvm/test/Other/new-pass-manager.ll
index f0fe708806f1b6..53fd6fe2a317ec 100644
--- a/llvm/test/Other/new-pass-manager.ll
+++ b/llvm/test/Other/new-pass-manager.ll
@@ -23,6 +23,7 @@
; CHECK-CGSCC-PASS-NEXT: Running analysis: InnerAnalysisManagerProxy<{{.*(FunctionAnalysisManager|AnalysisManager<.*Function.*>).*}},{{.*}}Module>
; CHECK-CGSCC-PASS-NEXT: Running analysis: LazyCallGraphAnalysis
; CHECK-CGSCC-PASS-NEXT: Running analysis: TargetLibraryAnalysis
+; CHECK-CGSCC-PASS-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-CGSCC-PASS-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-CGSCC-PASS-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-CGSCC-PASS-NEXT: Running pass: NoOpCGSCCPass
diff --git a/llvm/test/Other/new-pm-defaults.ll b/llvm/test/Other/new-pm-defaults.ll
index 7cf035b0c6f376..19bfec3ab718e0 100644
--- a/llvm/test/Other/new-pm-defaults.ll
+++ b/llvm/test/Other/new-pm-defaults.ll
@@ -139,6 +139,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-lto-defaults.ll b/llvm/test/Other/new-pm-lto-defaults.ll
index f788db1e338a1e..8f4fa763b52098 100644
--- a/llvm/test/Other/new-pm-lto-defaults.ll
+++ b/llvm/test/Other/new-pm-lto-defaults.ll
@@ -43,6 +43,7 @@
; CHECK-O23SZ-NEXT: Running analysis: OptimizationRemarkEmitterAnalysis
; CHECK-O23SZ-NEXT: Running analysis: InnerAnalysisManagerProxy<{{.*}}SCC
; CHECK-O23SZ-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O23SZ-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O23SZ-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O23SZ-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph{{.*}}>
; CHECK-O23SZ-NEXT: Running pass: PostOrderFunctionAttrsPass
diff --git a/llvm/test/Other/new-pm-pgo-preinline.ll b/llvm/test/Other/new-pm-pgo-preinline.ll
index f07a3728ba3d48..97813bb2433642 100644
--- a/llvm/test/Other/new-pm-pgo-preinline.ll
+++ b/llvm/test/Other/new-pm-pgo-preinline.ll
@@ -5,6 +5,7 @@
; CHECK-Osz-NEXT: Running analysis: InlineAdvisorAnalysis
; CHECK-Osz-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-Osz-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-Osz-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-Osz-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy on (foo)
; CHECK-Osz-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-Osz-NEXT: Running pass: InlinerPass on (foo)
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
index ed13402e1c4b15..e1ad86015fda95 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
@@ -74,6 +74,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
index c82c34f7ff01e7..3f6c5351e8e8e0 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
@@ -62,6 +62,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
index d375747547d61f..371dde305b0990 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
@@ -71,6 +71,7 @@
; CHECK-O-NEXT: Invalidating analysis: AAManager
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
index 5aacd26def2be5..860fb99525030f 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-defaults.ll
@@ -106,6 +106,7 @@
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
index f6a94065968038..d97cc97169b56b 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-pgo-defaults.ll
@@ -62,6 +62,7 @@
; CHECK-O-NEXT: Running analysis: InlineAdvisorAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
; CHECK-O-NEXT: Running analysis: LazyCallGraphAnalysis
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy on (foo)
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
; CHECK-O-NEXT: Running pass: InlinerPass on (foo)
diff --git a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
index 48a9433d249996..d338817d076463 100644
--- a/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-prelink-samplepgo-defaults.ll
@@ -76,6 +76,7 @@
; CHECK-O-NEXT: Invalidating analysis: AAManager
; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}ProfileSummaryAnalysis
; CHECK-O-NEXT: Running analysis: InnerAnalysisManagerProxy
+; CHECK-O-NEXT: Running analysis: DebugInfoCacheAnalysis
; CHECK-O-NEXT: Running analysis: FunctionAnalysisManagerCGSCCProxy
; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy<{{.*}}LazyCallGraph::SCC{{.*}}>
; CHECK-O-NEXT: Running pass: DevirtSCCRepeatedPass
diff --git a/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp b/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
index 5c71bc8063d6c9..7212107d992638 100644
--- a/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
+++ b/llvm/unittests/Analysis/CGSCCPassManagerTest.cpp
@@ -7,6 +7,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/CGSCCPassManager.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/AsmParser/Parser.h"
@@ -16,8 +17,8 @@
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
-#include "llvm/IR/PassManager.h"
#include "llvm/IR/PassInstrumentation.h"
+#include "llvm/IR/PassManager.h"
#include "llvm/IR/Verifier.h"
#include "llvm/Support/SourceMgr.h"
#include "llvm/Transforms/Utils/CallGraphUpdater.h"
@@ -255,6 +256,7 @@ class CGSCCPassManagerTest : public ::testing::Test {
"}\n")) {
FAM.registerPass([&] { return TargetLibraryAnalysis(); });
MAM.registerPass([&] { return LazyCallGraphAnalysis(); });
+ MAM.registerPass([&] { return DebugInfoCacheAnalysis(); });
MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); });
// Register required pass instrumentation analysis.
|
This was referenced Dec 4, 2024
Merged
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
bdb7970 to
22aaf36
Compare
artempyanykh
added a commit
that referenced
this pull request
Dec 6, 2024
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of global debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt GlobalDI | Cached CU DIFinder (cur.) | |-----------------+----------+----------------+-------------------+---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | |-----------------+----------+----------------+-------------------+---------------------------| | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
bec4d02 to
4cf7f3f
Compare
artempyanykh
changed the base branch from
users/artempyanykh/fast-coro-upstream/10
to
main
artempyanykh
added a commit
that referenced
this pull request
Dec 6, 2024
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of global debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt GlobalDI | Cached CU DIFinder (cur.) | |-----------------+----------+----------------+-------------------+---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | |-----------------+----------+----------------+-------------------+---------------------------| | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
4cf7f3f to
5d378de
Compare
artempyanykh
changed the base branch from
main
to
users/artempyanykh/fast-coro-upstream/10
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
b316815 to
c32803f
Compare
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
40407df to
2190fe0
Compare
artempyanykh
added a commit
that referenced
this pull request
Dec 17, 2024
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
b81f6f7 to
daedf07
Compare
artempyanykh
changed the base branch from
users/artempyanykh/fast-coro-upstream/10
to
main
artempyanykh
added a commit
that referenced
this pull request
Jan 9, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
daedf07 to
d2fd415
Compare
artempyanykh
changed the base branch from
main
to
users/artempyanykh/fast-coro-upstream/10
artempyanykh
changed the base branch from
users/artempyanykh/fast-coro-upstream/10
to
main
artempyanykh
added a commit
that referenced
this pull request
Jan 23, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
d2fd415 to
83cfc8d
Compare
artempyanykh
changed the base branch from
main
to
users/artempyanykh/fast-coro-upstream/10
artempyanykh
changed the base branch from
users/artempyanykh/fast-coro-upstream/10
to
main
artempyanykh
added a commit
that referenced
this pull request
Jan 23, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
83cfc8d to
809206d
Compare
artempyanykh
changed the base branch from
main
to
users/artempyanykh/fast-coro-upstream/10
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
5402d09 to
54bc13d
Compare
artempyanykh
added a commit
that referenced
this pull request
Jan 24, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of global debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectGlobalDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
809206d to
fc245ef
Compare
artempyanykh
changed the base branch from
users/artempyanykh/fast-coro-upstream/10
to
main
artempyanykh
added a commit
that referenced
this pull request
Jan 24, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of common debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectCommonDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
fc245ef to
81f1380
Compare
artempyanykh
changed the base branch from
main
to
users/artempyanykh/fast-coro-upstream/10
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/10
branch
from
f727512 to
7d38fe3
Compare
artempyanykh
added a commit
that referenced
this pull request
Jan 24, 2025
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of common debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectCommonDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
81f1380 to
27e9907
Compare
Summary: We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed up collection of module-level debug info. The pass could likely be another 2x+ faster if we avoid rebuilding the set of common debug info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done incrementally on top of this. Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample cpp file compiled with full debug info: | | Baseline | IdentityMD set | Prebuilt CommonDI | Cached CU DIFinder (cur.) | |-----------------|----------|----------------|-------------------|---------------------------| | CoroSplitPass | 306ms | 221ms | 68ms | 17ms | | CoroCloner | 101ms | 72ms | 0.5ms | 0.5ms | | CollectCommonDI | - | - | 63ms | 13ms | | Speed up | 1x | 1.4x | 4.5x | 18x | Test Plan: ninja check-llvm-unit ninja check-llvm Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes. stack-info: PR: #118630, branch: users/artempyanykh/fast-coro-upstream/11
artempyanykh
force-pushed
the
users/artempyanykh/fast-coro-upstream/11
branch
from
27e9907 to
78eb7e9
Compare
|
Dropping this in favor of #129148 |
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.
Stacked PRs:
[Coro] Use DebugInfoCache to speed up cloning in CoroSplitPass
Summary:
We can use a DebugInfoFinder from DebugInfoCache which is already primed on a compile unit to speed
up collection of module-level debug info.
The pass could likely be another 2x+ faster if we avoid rebuilding the set of common debug
info. This needs further massaging of CloneFunction and ValueMapper, though, and can be done
incrementally on top of this.
Comparing performance of CoroSplitPass at various points in this stack, this is anecdata from a sample
cpp file compiled with full debug info:
Test Plan:
ninja check-llvm-unit
ninja check-llvm
Compiled a sample cpp file with time trace to get the avg. duration of the pass and inner scopes.