[llvm] [Coro] Amortize debug info processing cost in CoroSplit (PR #109032)
Artem Pianykh via llvm-commits
llvm-commits at lists.llvm.org
Sat Nov 23 05:33:15 PST 2024
https://github.com/artempyanykh updated https://github.com/llvm/llvm-project/pull/109032
>From cf1397561ad693d88899436d64071395d5e3874e Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Thu, 12 Sep 2024 15:01:05 -0700
Subject: [PATCH 01/11] [NFC][Utils] Remove DebugInfoFinder parameter from
CloneBasicBlock
Summary:
There was a single usage of CloneBasicBlock with non-default DebugInfoFinder inside
CloneFunctionInto which has been refactored in the previous diff.
Let's remove the parameter to keep the scope of the function more focused.
Test Plan:
ninja check-llvm-unit check-llvm
---
llvm/include/llvm/Transforms/Utils/Cloning.h | 3 +--
llvm/lib/Transforms/Utils/CloneFunction.cpp | 12 ++----------
2 files changed, 3 insertions(+), 12 deletions(-)
diff --git a/llvm/include/llvm/Transforms/Utils/Cloning.h b/llvm/include/llvm/Transforms/Utils/Cloning.h
index 049d68b8a30681..3c8f2cbfaa9b81 100644
--- a/llvm/include/llvm/Transforms/Utils/Cloning.h
+++ b/llvm/include/llvm/Transforms/Utils/Cloning.h
@@ -119,8 +119,7 @@ struct ClonedCodeInfo {
/// parameter.
BasicBlock *CloneBasicBlock(const BasicBlock *BB, ValueToValueMapTy &VMap,
const Twine &NameSuffix = "", Function *F = nullptr,
- ClonedCodeInfo *CodeInfo = nullptr,
- DebugInfoFinder *DIFinder = nullptr);
+ ClonedCodeInfo *CodeInfo = nullptr);
/// Return a copy of the specified function and add it to that
/// function's module. Also, any references specified in the VMap are changed
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index cb6a4e34c226e5..b55776d736e663 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -43,21 +43,16 @@ using namespace llvm;
/// See comments in Cloning.h.
BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, ValueToValueMapTy &VMap,
const Twine &NameSuffix, Function *F,
- ClonedCodeInfo *CodeInfo,
- DebugInfoFinder *DIFinder) {
+ ClonedCodeInfo *CodeInfo) {
BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "", F);
NewBB->IsNewDbgInfoFormat = BB->IsNewDbgInfoFormat;
if (BB->hasName())
NewBB->setName(BB->getName() + NameSuffix);
bool hasCalls = false, hasDynamicAllocas = false, hasMemProfMetadata = false;
- Module *TheModule = F ? F->getParent() : nullptr;
// Loop over all instructions, and copy them over.
for (const Instruction &I : *BB) {
- if (DIFinder && TheModule)
- DIFinder->processInstruction(*TheModule, I);
-
Instruction *NewInst = I.clone();
if (I.hasName())
NewInst->setName(I.getName() + NameSuffix);
@@ -221,10 +216,7 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
for (const BasicBlock &BB : *OldFunc) {
// Create a new basic block and copy instructions into it!
- // NOTE: don't pass DIFinder because instructions' debug info was processed
- // in ProcessSubprogramAttachment. This will be cleaned up further.
- BasicBlock *CBB =
- CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo, nullptr);
+ BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
// Add basic block mapping.
VMap[&BB] = CBB;
>From 692b8615f2f810a493fbbc75c9c769dc93e1a2ce Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Thu, 12 Sep 2024 15:09:44 -0700
Subject: [PATCH 02/11] [NFC][Utils] Clone basic blocks after we're done with
metadata in CloneFunctionInto
Summary:
This makes the flow of the function a bit more straightforward and makes it easier to extract more
into helper functions.
Test Plan:
ninja check-llvm-unit
---
llvm/lib/Transforms/Utils/CloneFunction.cpp | 56 ++++++++++-----------
1 file changed, 28 insertions(+), 28 deletions(-)
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index b55776d736e663..d038117090e4cc 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -210,34 +210,6 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
DISubprogram *SPClonedWithinModule =
CollectDebugInfoForCloning(*OldFunc, Changes, DIFinder);
- // Loop over all of the basic blocks in the function, cloning them as
- // appropriate. Note that we save BE this way in order to handle cloning of
- // recursive functions into themselves.
- for (const BasicBlock &BB : *OldFunc) {
-
- // Create a new basic block and copy instructions into it!
- BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
-
- // Add basic block mapping.
- VMap[&BB] = CBB;
-
- // It is only legal to clone a function if a block address within that
- // function is never referenced outside of the function. Given that, we
- // want to map block addresses from the old function to block addresses in
- // the clone. (This is different from the generic ValueMapper
- // implementation, which generates an invalid blockaddress when
- // cloning a function.)
- if (BB.hasAddressTaken()) {
- Constant *OldBBAddr = BlockAddress::get(const_cast<Function *>(OldFunc),
- const_cast<BasicBlock *>(&BB));
- VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB);
- }
-
- // Note return instructions for the caller.
- if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
- Returns.push_back(RI);
- }
-
if (Changes < CloneFunctionChangeType::DifferentModule &&
DIFinder.subprogram_count() > 0) {
// Turn on module-level changes, since we need to clone (some of) the
@@ -289,6 +261,34 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
TypeMapper, Materializer));
}
+ // Loop over all of the basic blocks in the function, cloning them as
+ // appropriate. Note that we save BE this way in order to handle cloning of
+ // recursive functions into themselves.
+ for (const BasicBlock &BB : *OldFunc) {
+
+ // Create a new basic block and copy instructions into it!
+ BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
+
+ // Add basic block mapping.
+ VMap[&BB] = CBB;
+
+ // It is only legal to clone a function if a block address within that
+ // function is never referenced outside of the function. Given that, we
+ // want to map block addresses from the old function to block addresses in
+ // the clone. (This is different from the generic ValueMapper
+ // implementation, which generates an invalid blockaddress when
+ // cloning a function.)
+ if (BB.hasAddressTaken()) {
+ Constant *OldBBAddr = BlockAddress::get(const_cast<Function *>(OldFunc),
+ const_cast<BasicBlock *>(&BB));
+ VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB);
+ }
+
+ // Note return instructions for the caller.
+ if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
+ Returns.push_back(RI);
+ }
+
// Loop over all of the instructions in the new function, fixing up operand
// references as we go. This uses VMap to do all the hard work.
for (Function::iterator
>From 63efec5bcc802e68945920fac2c9aefcda58f645 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Thu, 12 Sep 2024 15:23:43 -0700
Subject: [PATCH 03/11] [NFC][Utils] Extract BuildDebugInfoMDMap from
CloneFunctionInto
Summary:
Extract the logic to build up a metadta map to use in metadata cloning into a separate function.
Test Plan:
ninja check-llvm-unit
---
llvm/include/llvm/IR/ValueMap.h | 4 +-
llvm/include/llvm/Transforms/Utils/Cloning.h | 7 ++
llvm/lib/Transforms/Utils/CloneFunction.cpp | 88 +++++++++++---------
3 files changed, 59 insertions(+), 40 deletions(-)
diff --git a/llvm/include/llvm/IR/ValueMap.h b/llvm/include/llvm/IR/ValueMap.h
index d12d639aaa8886..fd4c08492e0124 100644
--- a/llvm/include/llvm/IR/ValueMap.h
+++ b/llvm/include/llvm/IR/ValueMap.h
@@ -79,6 +79,9 @@ struct ValueMapConfig {
static mutex_type *getMutex(const ExtraDataT &/*Data*/) { return nullptr; }
};
+/// This type stores Metadata. Used in ValueMap.
+using MDMapT = DenseMap<const Metadata *, TrackingMDRef>;
+
/// See the file comment.
template<typename KeyT, typename ValueT, typename Config =ValueMapConfig<KeyT>>
class ValueMap {
@@ -86,7 +89,6 @@ class ValueMap {
using ValueMapCVH = ValueMapCallbackVH<KeyT, ValueT, Config>;
using MapT = DenseMap<ValueMapCVH, ValueT, DenseMapInfo<ValueMapCVH>>;
- using MDMapT = DenseMap<const Metadata *, TrackingMDRef>;
using ExtraData = typename Config::ExtraData;
MapT Map;
diff --git a/llvm/include/llvm/Transforms/Utils/Cloning.h b/llvm/include/llvm/Transforms/Utils/Cloning.h
index 3c8f2cbfaa9b81..434089138bc521 100644
--- a/llvm/include/llvm/Transforms/Utils/Cloning.h
+++ b/llvm/include/llvm/Transforms/Utils/Cloning.h
@@ -220,6 +220,13 @@ DISubprogram *CollectDebugInfoForCloning(const Function &F,
CloneFunctionChangeType Changes,
DebugInfoFinder &DIFinder);
+/// Build a map of debug info to use during Metadata cloning.
+/// Returns true if cloning would need module level changes and false if there
+/// would only be local changes.
+bool BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
+ DebugInfoFinder &DIFinder,
+ DISubprogram *SPClonedWithinModule);
+
/// This class captures the data input to the InlineFunction call, and records
/// the auxiliary results produced by it.
class InlineFunctionInfo {
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index d038117090e4cc..736c4a13045c14 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -152,6 +152,53 @@ DISubprogram *llvm::CollectDebugInfoForCloning(const Function &F,
return SPClonedWithinModule;
}
+bool llvm::BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
+ DebugInfoFinder &DIFinder,
+ DISubprogram *SPClonedWithinModule) {
+ bool ModuleLevelChanges = Changes > CloneFunctionChangeType::LocalChangesOnly;
+ if (Changes < CloneFunctionChangeType::DifferentModule &&
+ DIFinder.subprogram_count() > 0) {
+ // Turn on module-level changes, since we need to clone (some of) the
+ // debug info metadata.
+ //
+ // FIXME: Metadata effectively owned by a function should be made
+ // local, and only that local metadata should be cloned.
+ ModuleLevelChanges = true;
+
+ auto mapToSelfIfNew = [&MD](MDNode *N) {
+ // Avoid clobbering an existing mapping.
+ (void)MD.try_emplace(N, N);
+ };
+
+ // Avoid cloning types, compile units, and (other) subprograms.
+ SmallPtrSet<const DISubprogram *, 16> MappedToSelfSPs;
+ for (DISubprogram *ISP : DIFinder.subprograms()) {
+ if (ISP != SPClonedWithinModule) {
+ mapToSelfIfNew(ISP);
+ MappedToSelfSPs.insert(ISP);
+ }
+ }
+
+ // If a subprogram isn't going to be cloned skip its lexical blocks as well.
+ for (DIScope *S : DIFinder.scopes()) {
+ auto *LScope = dyn_cast<DILocalScope>(S);
+ if (LScope && MappedToSelfSPs.count(LScope->getSubprogram()))
+ mapToSelfIfNew(S);
+ }
+
+ for (DICompileUnit *CU : DIFinder.compile_units())
+ mapToSelfIfNew(CU);
+
+ for (DIType *Type : DIFinder.types())
+ mapToSelfIfNew(Type);
+ } else {
+ assert(!SPClonedWithinModule &&
+ "Subprogram should be in DIFinder->subprogram_count()...");
+ }
+
+ return ModuleLevelChanges;
+}
+
// Clone OldFunc into NewFunc, transforming the old arguments into references to
// VMap values.
void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
@@ -210,45 +257,8 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
DISubprogram *SPClonedWithinModule =
CollectDebugInfoForCloning(*OldFunc, Changes, DIFinder);
- if (Changes < CloneFunctionChangeType::DifferentModule &&
- DIFinder.subprogram_count() > 0) {
- // Turn on module-level changes, since we need to clone (some of) the
- // debug info metadata.
- //
- // FIXME: Metadata effectively owned by a function should be made
- // local, and only that local metadata should be cloned.
- ModuleLevelChanges = true;
-
- auto mapToSelfIfNew = [&VMap](MDNode *N) {
- // Avoid clobbering an existing mapping.
- (void)VMap.MD().try_emplace(N, N);
- };
-
- // Avoid cloning types, compile units, and (other) subprograms.
- SmallPtrSet<const DISubprogram *, 16> MappedToSelfSPs;
- for (DISubprogram *ISP : DIFinder.subprograms()) {
- if (ISP != SPClonedWithinModule) {
- mapToSelfIfNew(ISP);
- MappedToSelfSPs.insert(ISP);
- }
- }
-
- // If a subprogram isn't going to be cloned skip its lexical blocks as well.
- for (DIScope *S : DIFinder.scopes()) {
- auto *LScope = dyn_cast<DILocalScope>(S);
- if (LScope && MappedToSelfSPs.count(LScope->getSubprogram()))
- mapToSelfIfNew(S);
- }
-
- for (DICompileUnit *CU : DIFinder.compile_units())
- mapToSelfIfNew(CU);
-
- for (DIType *Type : DIFinder.types())
- mapToSelfIfNew(Type);
- } else {
- assert(!SPClonedWithinModule &&
- "Subprogram should be in DIFinder->subprogram_count()...");
- }
+ ModuleLevelChanges =
+ BuildDebugInfoMDMap(VMap.MD(), Changes, DIFinder, SPClonedWithinModule);
const auto RemapFlag = ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges;
// Duplicate the metadata that is attached to the cloned function.
>From fbbbf4557964e8e0d0ae7561050b6814b5ac8ae7 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Thu, 12 Sep 2024 15:35:38 -0700
Subject: [PATCH 04/11] [NFC][Utils] Extract CloneFunctionMetadataInto from
CloneFunctionInto
Summary:
One potentially contentious point in the new function's API is that it expects the caller ot
populate the VMap rather than doing metadata cloning wholesale inside it.
We'll need it this way for a subsequent change.
Test Plan:
ninja check-llvm-unit
---
llvm/include/llvm/Transforms/Utils/Cloning.h | 12 +++++++++
llvm/lib/Transforms/Utils/CloneFunction.cpp | 28 +++++++++++++-------
2 files changed, 31 insertions(+), 9 deletions(-)
diff --git a/llvm/include/llvm/Transforms/Utils/Cloning.h b/llvm/include/llvm/Transforms/Utils/Cloning.h
index 434089138bc521..698d773525e80b 100644
--- a/llvm/include/llvm/Transforms/Utils/Cloning.h
+++ b/llvm/include/llvm/Transforms/Utils/Cloning.h
@@ -182,6 +182,18 @@ void CloneFunctionAttributesInto(Function *NewFunc, const Function *OldFunc,
ValueMapTypeRemapper *TypeMapper = nullptr,
ValueMaterializer *Materializer = nullptr);
+/// Clone OldFunc's metadata into NewFunc.
+///
+/// The caller is expected to populate \p VMap beforehand and set an appropriate
+/// \p RemapFlag.
+///
+/// NOTE: This function doesn't clone !llvm.dbg.cu when cloning into a different
+/// module. Use CloneFunctionInto for that behavior.
+void CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
+ ValueToValueMapTy &VMap, RemapFlags RemapFlag,
+ ValueMapTypeRemapper *TypeMapper = nullptr,
+ ValueMaterializer *Materializer = nullptr);
+
void CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
const Instruction *StartingInst,
ValueToValueMapTy &VMap, bool ModuleLevelChanges,
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index 736c4a13045c14..4306e5d2abdd18 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -199,6 +199,22 @@ bool llvm::BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
return ModuleLevelChanges;
}
+void llvm::CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
+ ValueToValueMapTy &VMap,
+ RemapFlags RemapFlag,
+ ValueMapTypeRemapper *TypeMapper,
+ ValueMaterializer *Materializer) {
+ // Duplicate the metadata that is attached to the cloned function.
+ // Subprograms/CUs/types that were already mapped to themselves won't be
+ // duplicated.
+ SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
+ OldFunc->getAllMetadata(MDs);
+ for (auto MD : MDs) {
+ NewFunc->addMetadata(MD.first, *MapMetadata(MD.second, VMap, RemapFlag,
+ TypeMapper, Materializer));
+ }
+}
+
// Clone OldFunc into NewFunc, transforming the old arguments into references to
// VMap values.
void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
@@ -261,15 +277,9 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
BuildDebugInfoMDMap(VMap.MD(), Changes, DIFinder, SPClonedWithinModule);
const auto RemapFlag = ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges;
- // Duplicate the metadata that is attached to the cloned function.
- // Subprograms/CUs/types that were already mapped to themselves won't be
- // duplicated.
- SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
- OldFunc->getAllMetadata(MDs);
- for (auto MD : MDs) {
- NewFunc->addMetadata(MD.first, *MapMetadata(MD.second, VMap, RemapFlag,
- TypeMapper, Materializer));
- }
+
+ CloneFunctionMetadataInto(NewFunc, OldFunc, VMap, RemapFlag, TypeMapper,
+ Materializer);
// Loop over all of the basic blocks in the function, cloning them as
// appropriate. Note that we save BE this way in order to handle cloning of
>From 9b1f62f17b450a5d5f66c135572b31b49360c339 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Thu, 12 Sep 2024 15:50:25 -0700
Subject: [PATCH 05/11] [NFC][Utils] Extract CloneFunctionBodyInto from
CloneFunctionInto
Summary:
This and previously extracted functions will be used in a later diff.
Test Plan:
ninja check-llvm-unit
---
llvm/include/llvm/Transforms/Utils/Cloning.h | 9 ++
llvm/lib/Transforms/Utils/CloneFunction.cpp | 96 +++++++++++---------
2 files changed, 64 insertions(+), 41 deletions(-)
diff --git a/llvm/include/llvm/Transforms/Utils/Cloning.h b/llvm/include/llvm/Transforms/Utils/Cloning.h
index 698d773525e80b..6c73328052380a 100644
--- a/llvm/include/llvm/Transforms/Utils/Cloning.h
+++ b/llvm/include/llvm/Transforms/Utils/Cloning.h
@@ -194,6 +194,15 @@ void CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
ValueMapTypeRemapper *TypeMapper = nullptr,
ValueMaterializer *Materializer = nullptr);
+/// Clone OldFunc's body NewFunct.
+void CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
+ ValueToValueMapTy &VMap, RemapFlags RemapFlag,
+ SmallVectorImpl<ReturnInst *> &Returns,
+ const char *NameSuffix = "",
+ ClonedCodeInfo *CodeInfo = nullptr,
+ ValueMapTypeRemapper *TypeMapper = nullptr,
+ ValueMaterializer *Materializer = nullptr);
+
void CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
const Instruction *StartingInst,
ValueToValueMapTy &VMap, bool ModuleLevelChanges,
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index 4306e5d2abdd18..1f4cdec333a2c2 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -215,6 +215,59 @@ void llvm::CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
}
}
+void llvm::CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
+ ValueToValueMapTy &VMap, RemapFlags RemapFlag,
+ SmallVectorImpl<ReturnInst *> &Returns,
+ const char *NameSuffix,
+ ClonedCodeInfo *CodeInfo,
+ ValueMapTypeRemapper *TypeMapper,
+ ValueMaterializer *Materializer) {
+ if (OldFunc->isDeclaration())
+ return;
+
+ // Loop over all of the basic blocks in the function, cloning them as
+ // appropriate. Note that we save BE this way in order to handle cloning of
+ // recursive functions into themselves.
+ for (const BasicBlock &BB : *OldFunc) {
+
+ // Create a new basic block and copy instructions into it!
+ BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
+
+ // Add basic block mapping.
+ VMap[&BB] = CBB;
+
+ // It is only legal to clone a function if a block address within that
+ // function is never referenced outside of the function. Given that, we
+ // want to map block addresses from the old function to block addresses in
+ // the clone. (This is different from the generic ValueMapper
+ // implementation, which generates an invalid blockaddress when
+ // cloning a function.)
+ if (BB.hasAddressTaken()) {
+ Constant *OldBBAddr = BlockAddress::get(const_cast<Function *>(OldFunc),
+ const_cast<BasicBlock *>(&BB));
+ VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB);
+ }
+
+ // Note return instructions for the caller.
+ if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
+ Returns.push_back(RI);
+ }
+
+ // Loop over all of the instructions in the new function, fixing up operand
+ // references as we go. This uses VMap to do all the hard work.
+ for (Function::iterator
+ BB = cast<BasicBlock>(VMap[&OldFunc->front()])->getIterator(),
+ BE = NewFunc->end();
+ BB != BE; ++BB)
+ // Loop over all instructions, fixing each one as we find it, and any
+ // attached debug-info records.
+ for (Instruction &II : *BB) {
+ RemapInstruction(&II, VMap, RemapFlag, TypeMapper, Materializer);
+ RemapDbgRecordRange(II.getModule(), II.getDbgRecordRange(), VMap,
+ RemapFlag, TypeMapper, Materializer);
+ }
+}
+
// Clone OldFunc into NewFunc, transforming the old arguments into references to
// VMap values.
void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
@@ -281,47 +334,8 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
CloneFunctionMetadataInto(NewFunc, OldFunc, VMap, RemapFlag, TypeMapper,
Materializer);
- // Loop over all of the basic blocks in the function, cloning them as
- // appropriate. Note that we save BE this way in order to handle cloning of
- // recursive functions into themselves.
- for (const BasicBlock &BB : *OldFunc) {
-
- // Create a new basic block and copy instructions into it!
- BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
-
- // Add basic block mapping.
- VMap[&BB] = CBB;
-
- // It is only legal to clone a function if a block address within that
- // function is never referenced outside of the function. Given that, we
- // want to map block addresses from the old function to block addresses in
- // the clone. (This is different from the generic ValueMapper
- // implementation, which generates an invalid blockaddress when
- // cloning a function.)
- if (BB.hasAddressTaken()) {
- Constant *OldBBAddr = BlockAddress::get(const_cast<Function *>(OldFunc),
- const_cast<BasicBlock *>(&BB));
- VMap[OldBBAddr] = BlockAddress::get(NewFunc, CBB);
- }
-
- // Note return instructions for the caller.
- if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
- Returns.push_back(RI);
- }
-
- // Loop over all of the instructions in the new function, fixing up operand
- // references as we go. This uses VMap to do all the hard work.
- for (Function::iterator
- BB = cast<BasicBlock>(VMap[&OldFunc->front()])->getIterator(),
- BE = NewFunc->end();
- BB != BE; ++BB)
- // Loop over all instructions, fixing each one as we find it, and any
- // attached debug-info records.
- for (Instruction &II : *BB) {
- RemapInstruction(&II, VMap, RemapFlag, TypeMapper, Materializer);
- RemapDbgRecordRange(II.getModule(), II.getDbgRecordRange(), VMap,
- RemapFlag, TypeMapper, Materializer);
- }
+ CloneFunctionBodyInto(NewFunc, OldFunc, VMap, RemapFlag, Returns, NameSuffix,
+ CodeInfo, TypeMapper, Materializer);
// Only update !llvm.dbg.cu for DifferentModule (not CloneModule). In the
// same module, the compile unit will already be listed (or not). When
>From 191cecf38e7d5a35ed4c026fb326a5d26b8ce65a Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Sat, 14 Sep 2024 16:02:51 -0700
Subject: [PATCH 06/11] [NFC][Utils] Eliminate DISubprogram set from
BuildDebugInfoMDMap
Summary:
Previously, we'd add all SPs distinct from the cloned one into a set.
Then when cloning a local scope we'd check if it's from one of those
'distinct' SPs by checking if it's in the set. We don't need to do that.
We can just check against the cloned SP directly and drop the set.
Test Plan:
ninja check-llvm-unit check-llvm
---
llvm/lib/Transforms/Utils/CloneFunction.cpp | 7 ++-----
1 file changed, 2 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index 1f4cdec333a2c2..5bc90bc7ae23da 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -171,18 +171,15 @@ bool llvm::BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
};
// Avoid cloning types, compile units, and (other) subprograms.
- SmallPtrSet<const DISubprogram *, 16> MappedToSelfSPs;
for (DISubprogram *ISP : DIFinder.subprograms()) {
- if (ISP != SPClonedWithinModule) {
+ if (ISP != SPClonedWithinModule)
mapToSelfIfNew(ISP);
- MappedToSelfSPs.insert(ISP);
- }
}
// If a subprogram isn't going to be cloned skip its lexical blocks as well.
for (DIScope *S : DIFinder.scopes()) {
auto *LScope = dyn_cast<DILocalScope>(S);
- if (LScope && MappedToSelfSPs.count(LScope->getSubprogram()))
+ if (LScope && LScope->getSubprogram() != SPClonedWithinModule)
mapToSelfIfNew(S);
}
>From 3a9d87679129650e01279c51f725d7f5e654d305 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Sun, 15 Sep 2024 02:59:24 -0700
Subject: [PATCH 07/11] [NFC] Remove adhoc definition of MDMapT in IRMover
Summary:
The typedef was there probably because the type alias in ValueMap was private.
Test Plan:
ninja check-llvm-unit
---
llvm/include/llvm/Linker/IRMover.h | 4 +---
llvm/lib/Linker/IRMover.cpp | 3 ---
2 files changed, 1 insertion(+), 6 deletions(-)
diff --git a/llvm/include/llvm/Linker/IRMover.h b/llvm/include/llvm/Linker/IRMover.h
index 1e3c5394ffa2af..89e9cbe0be18e1 100644
--- a/llvm/include/llvm/Linker/IRMover.h
+++ b/llvm/include/llvm/Linker/IRMover.h
@@ -12,6 +12,7 @@
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/FunctionExtras.h"
+#include "llvm/IR/ValueMap.h"
#include <functional>
namespace llvm {
@@ -41,9 +42,6 @@ class IRMover {
static bool isEqual(const StructType *LHS, const StructType *RHS);
};
- /// Type of the Metadata map in \a ValueToValueMapTy.
- typedef DenseMap<const Metadata *, TrackingMDRef> MDMapT;
-
public:
class IdentifiedStructTypeSet {
// The set of opaque types is the composite module.
diff --git a/llvm/lib/Linker/IRMover.cpp b/llvm/lib/Linker/IRMover.cpp
index a0c3f2c5b0baf6..76af83df00cba2 100644
--- a/llvm/lib/Linker/IRMover.cpp
+++ b/llvm/lib/Linker/IRMover.cpp
@@ -373,9 +373,6 @@ class LocalValueMaterializer final : public ValueMaterializer {
Value *materialize(Value *V) override;
};
-/// Type of the Metadata map in \a ValueToValueMapTy.
-typedef DenseMap<const Metadata *, TrackingMDRef> MDMapT;
-
/// This is responsible for keeping track of the state used for moving data
/// from SrcM to DstM.
class IRLinker {
>From 29e12aae38adf2b850440f6af92bfa0da7814d60 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Sun, 15 Sep 2024 04:39:20 -0700
Subject: [PATCH 08/11] [Utils] Identity map global debug info on first use in
CloneFunction*
Summary:
To avoid cloning 'global' debug info, CloneFunction implementation used to eagerly identity map a known
subset of global debug into into ValueMap's MD map. In larger modules with meaningful volume of
debug info this gets very expensive.
By passing such global metadata via an IdentityMD set for the ValueMapper to map on first use, we
get several benefits:
1. Mapping metadata is not cheap, particularly because of tracking. When cloning a Function we
identity map lots of global module-level metadata to avoid cloning it, while only a fraction of it
is actually used by the function. Mapping on first use is a lot faster for modules with meaningful
amount of debug info.
2. Eagerly identity mapping metadata makes it harder to cache module-level data (e.g. a set of
metadata nodes in a \a DICompileUnit). With this patch we can cache certain module-level metadata
calculations to speed things up further.
Anecdata from compiling a sample cpp file with full debug info shows that this moderately speeds up
CoroSplitPass which is one of the heavier users of cloning:
| | Baseline | IdentityMD set |
|-----------------+----------+----------------|
| CoroSplitPass | 306ms | 221ms |
| CoroCloner | 101ms | 72ms |
|-----------------+----------+----------------|
| Speed up | 1x | 1.4x |
Test Plan:
ninja check-llvm-unit
ninja check-llvm
---
llvm/include/llvm/Transforms/Utils/Cloning.h | 18 ++---
.../llvm/Transforms/Utils/ValueMapper.h | 67 ++++++++++++++-----
llvm/lib/Transforms/Utils/CloneFunction.cpp | 58 ++++++++--------
llvm/lib/Transforms/Utils/ValueMapper.cpp | 19 ++++--
4 files changed, 103 insertions(+), 59 deletions(-)
diff --git a/llvm/include/llvm/Transforms/Utils/Cloning.h b/llvm/include/llvm/Transforms/Utils/Cloning.h
index 6c73328052380a..9862165cd4638c 100644
--- a/llvm/include/llvm/Transforms/Utils/Cloning.h
+++ b/llvm/include/llvm/Transforms/Utils/Cloning.h
@@ -192,7 +192,8 @@ void CloneFunctionAttributesInto(Function *NewFunc, const Function *OldFunc,
void CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap, RemapFlags RemapFlag,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr);
/// Clone OldFunc's body NewFunct.
void CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
@@ -201,7 +202,8 @@ void CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
const char *NameSuffix = "",
ClonedCodeInfo *CodeInfo = nullptr,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr);
void CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
const Instruction *StartingInst,
@@ -241,12 +243,12 @@ DISubprogram *CollectDebugInfoForCloning(const Function &F,
CloneFunctionChangeType Changes,
DebugInfoFinder &DIFinder);
-/// Build a map of debug info to use during Metadata cloning.
-/// Returns true if cloning would need module level changes and false if there
-/// would only be local changes.
-bool BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
- DebugInfoFinder &DIFinder,
- DISubprogram *SPClonedWithinModule);
+/// Based on \p Changes and \p DIFinder populate \p MD with debug info that
+/// needs to be identity mapped during Metadata cloning.
+void FindDebugInfoToIdentityMap(MetadataSetTy &MD,
+ CloneFunctionChangeType Changes,
+ DebugInfoFinder &DIFinder,
+ DISubprogram *SPClonedWithinModule);
/// This class captures the data input to the InlineFunction call, and records
/// the auxiliary results produced by it.
diff --git a/llvm/include/llvm/Transforms/Utils/ValueMapper.h b/llvm/include/llvm/Transforms/Utils/ValueMapper.h
index 743cfeb7ef3f02..b8d612f11d519f 100644
--- a/llvm/include/llvm/Transforms/Utils/ValueMapper.h
+++ b/llvm/include/llvm/Transforms/Utils/ValueMapper.h
@@ -15,6 +15,7 @@
#define LLVM_TRANSFORMS_UTILS_VALUEMAPPER_H
#include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/simple_ilist.h"
#include "llvm/IR/ValueHandle.h"
#include "llvm/IR/ValueMap.h"
@@ -35,6 +36,7 @@ class Value;
using ValueToValueMapTy = ValueMap<const Value *, WeakTrackingVH>;
using DbgRecordIterator = simple_ilist<DbgRecord>::iterator;
+using MetadataSetTy = SmallPtrSet<const Metadata *, 16>;
/// This is a class that can be implemented by clients to remap types when
/// cloning constants and instructions.
@@ -136,6 +138,18 @@ inline RemapFlags operator|(RemapFlags LHS, RemapFlags RHS) {
/// alternate \a ValueToValueMapTy and \a ValueMaterializer and returns a ID to
/// pass into the schedule*() functions.
///
+/// NOTE: \c IdentityMD is used by CloneFunction* to directly specify metadata
+/// that should be identity mapped (and hence not cloned). The metadata will be
+/// identity mapped in \c VM on first use. There are several reasons for doing
+/// it this way rather than eagerly identity mapping metadata nodes in \c VM:
+/// 1. Mapping metadata is not cheap, particularly because of tracking.
+/// 2. When cloning a Function we identity map lots of global module-level
+/// metadata to avoid cloning it, while only a fraction of it is actually
+/// used by the function. Mapping on first use is a lot faster for modules
+/// with meaningful amount of debug info.
+/// 3. Eagerly identity mapping metadata makes it harder to cache module-level
+/// data (e.g. a set of metadata nodes in a \a DICompileUnit).
+///
/// TODO: lib/Linker really doesn't need the \a ValueHandle in the \a
/// ValueToValueMapTy. We should template \a ValueMapper (and its
/// implementation classes), and explicitly instantiate on two concrete
@@ -152,7 +166,8 @@ class ValueMapper {
public:
ValueMapper(ValueToValueMapTy &VM, RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr);
ValueMapper(ValueMapper &&) = delete;
ValueMapper(const ValueMapper &) = delete;
ValueMapper &operator=(ValueMapper &&) = delete;
@@ -218,8 +233,10 @@ class ValueMapper {
inline Value *MapValue(const Value *V, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- return ValueMapper(VM, Flags, TypeMapper, Materializer).mapValue(*V);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ return ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .mapValue(*V);
}
/// Lookup or compute a mapping for a piece of metadata.
@@ -231,7 +248,9 @@ inline Value *MapValue(const Value *V, ValueToValueMapTy &VM,
/// \c MD.
/// 3. Else if \c MD is a \a ConstantAsMetadata, call \a MapValue() and
/// re-wrap its return (returning nullptr on nullptr).
-/// 4. Else, \c MD is an \a MDNode. These are remapped, along with their
+/// 4. Else if \c MD is in \c IdentityMD then add an identity mapping for it
+/// and return it.
+/// 5. Else, \c MD is an \a MDNode. These are remapped, along with their
/// transitive operands. Distinct nodes are duplicated or moved depending
/// on \a RF_MoveDistinctNodes. Uniqued nodes are remapped like constants.
///
@@ -240,16 +259,20 @@ inline Value *MapValue(const Value *V, ValueToValueMapTy &VM,
inline Metadata *MapMetadata(const Metadata *MD, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- return ValueMapper(VM, Flags, TypeMapper, Materializer).mapMetadata(*MD);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ return ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .mapMetadata(*MD);
}
/// Version of MapMetadata with type safety for MDNode.
inline MDNode *MapMetadata(const MDNode *MD, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- return ValueMapper(VM, Flags, TypeMapper, Materializer).mapMDNode(*MD);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ return ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .mapMDNode(*MD);
}
/// Convert the instruction operands from referencing the current values into
@@ -263,8 +286,10 @@ inline MDNode *MapMetadata(const MDNode *MD, ValueToValueMapTy &VM,
inline void RemapInstruction(Instruction *I, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- ValueMapper(VM, Flags, TypeMapper, Materializer).remapInstruction(*I);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .remapInstruction(*I);
}
/// Remap the Values used in the DbgRecord \a DR using the value map \a
@@ -272,8 +297,10 @@ inline void RemapInstruction(Instruction *I, ValueToValueMapTy &VM,
inline void RemapDbgRecord(Module *M, DbgRecord *DR, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- ValueMapper(VM, Flags, TypeMapper, Materializer).remapDbgRecord(M, *DR);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .remapDbgRecord(M, *DR);
}
/// Remap the Values used in the DbgRecords \a Range using the value map \a
@@ -283,8 +310,9 @@ inline void RemapDbgRecordRange(Module *M,
ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- ValueMapper(VM, Flags, TypeMapper, Materializer)
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
.remapDbgRecordRange(M, Range);
}
@@ -297,16 +325,19 @@ inline void RemapDbgRecordRange(Module *M,
inline void RemapFunction(Function &F, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- ValueMapper(VM, Flags, TypeMapper, Materializer).remapFunction(F);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD).remapFunction(F);
}
/// Version of MapValue with type safety for Constant.
inline Constant *MapValue(const Constant *V, ValueToValueMapTy &VM,
RemapFlags Flags = RF_None,
ValueMapTypeRemapper *TypeMapper = nullptr,
- ValueMaterializer *Materializer = nullptr) {
- return ValueMapper(VM, Flags, TypeMapper, Materializer).mapConstant(*V);
+ ValueMaterializer *Materializer = nullptr,
+ const MetadataSetTy *IdentityMD = nullptr) {
+ return ValueMapper(VM, Flags, TypeMapper, Materializer, IdentityMD)
+ .mapConstant(*V);
}
} // end namespace llvm
diff --git a/llvm/lib/Transforms/Utils/CloneFunction.cpp b/llvm/lib/Transforms/Utils/CloneFunction.cpp
index 5bc90bc7ae23da..057bd4fae33533 100644
--- a/llvm/lib/Transforms/Utils/CloneFunction.cpp
+++ b/llvm/lib/Transforms/Utils/CloneFunction.cpp
@@ -152,63 +152,57 @@ DISubprogram *llvm::CollectDebugInfoForCloning(const Function &F,
return SPClonedWithinModule;
}
-bool llvm::BuildDebugInfoMDMap(MDMapT &MD, CloneFunctionChangeType Changes,
- DebugInfoFinder &DIFinder,
- DISubprogram *SPClonedWithinModule) {
- bool ModuleLevelChanges = Changes > CloneFunctionChangeType::LocalChangesOnly;
+void llvm::FindDebugInfoToIdentityMap(MetadataSetTy &MD,
+ CloneFunctionChangeType Changes,
+ DebugInfoFinder &DIFinder,
+ DISubprogram *SPClonedWithinModule) {
if (Changes < CloneFunctionChangeType::DifferentModule &&
DIFinder.subprogram_count() > 0) {
- // Turn on module-level changes, since we need to clone (some of) the
- // debug info metadata.
+ // Even if Changes are local only, we turn on module-level changes, since we
+ // need to clone (some of) the debug info metadata.
//
// FIXME: Metadata effectively owned by a function should be made
// local, and only that local metadata should be cloned.
- ModuleLevelChanges = true;
-
- auto mapToSelfIfNew = [&MD](MDNode *N) {
- // Avoid clobbering an existing mapping.
- (void)MD.try_emplace(N, N);
- };
// Avoid cloning types, compile units, and (other) subprograms.
for (DISubprogram *ISP : DIFinder.subprograms()) {
if (ISP != SPClonedWithinModule)
- mapToSelfIfNew(ISP);
+ MD.insert(ISP);
}
// If a subprogram isn't going to be cloned skip its lexical blocks as well.
for (DIScope *S : DIFinder.scopes()) {
auto *LScope = dyn_cast<DILocalScope>(S);
if (LScope && LScope->getSubprogram() != SPClonedWithinModule)
- mapToSelfIfNew(S);
+ MD.insert(S);
}
for (DICompileUnit *CU : DIFinder.compile_units())
- mapToSelfIfNew(CU);
+ MD.insert(CU);
for (DIType *Type : DIFinder.types())
- mapToSelfIfNew(Type);
+ MD.insert(Type);
} else {
assert(!SPClonedWithinModule &&
"Subprogram should be in DIFinder->subprogram_count()...");
}
-
- return ModuleLevelChanges;
}
void llvm::CloneFunctionMetadataInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap,
RemapFlags RemapFlag,
ValueMapTypeRemapper *TypeMapper,
- ValueMaterializer *Materializer) {
+ ValueMaterializer *Materializer,
+ const MetadataSetTy *IdentityMD) {
// Duplicate the metadata that is attached to the cloned function.
// Subprograms/CUs/types that were already mapped to themselves won't be
// duplicated.
SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
OldFunc->getAllMetadata(MDs);
for (auto MD : MDs) {
- NewFunc->addMetadata(MD.first, *MapMetadata(MD.second, VMap, RemapFlag,
- TypeMapper, Materializer));
+ NewFunc->addMetadata(MD.first,
+ *MapMetadata(MD.second, VMap, RemapFlag, TypeMapper,
+ Materializer, IdentityMD));
}
}
@@ -218,7 +212,8 @@ void llvm::CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
const char *NameSuffix,
ClonedCodeInfo *CodeInfo,
ValueMapTypeRemapper *TypeMapper,
- ValueMaterializer *Materializer) {
+ ValueMaterializer *Materializer,
+ const MetadataSetTy *IdentityMD) {
if (OldFunc->isDeclaration())
return;
@@ -259,9 +254,10 @@ void llvm::CloneFunctionBodyInto(Function *NewFunc, const Function *OldFunc,
// Loop over all instructions, fixing each one as we find it, and any
// attached debug-info records.
for (Instruction &II : *BB) {
- RemapInstruction(&II, VMap, RemapFlag, TypeMapper, Materializer);
+ RemapInstruction(&II, VMap, RemapFlag, TypeMapper, Materializer,
+ IdentityMD);
RemapDbgRecordRange(II.getModule(), II.getDbgRecordRange(), VMap,
- RemapFlag, TypeMapper, Materializer);
+ RemapFlag, TypeMapper, Materializer, IdentityMD);
}
}
@@ -323,16 +319,20 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
DISubprogram *SPClonedWithinModule =
CollectDebugInfoForCloning(*OldFunc, Changes, DIFinder);
- ModuleLevelChanges =
- BuildDebugInfoMDMap(VMap.MD(), Changes, DIFinder, SPClonedWithinModule);
+ MetadataSetTy IdentityMD;
+ FindDebugInfoToIdentityMap(IdentityMD, Changes, DIFinder,
+ SPClonedWithinModule);
- const auto RemapFlag = ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges;
+ // Current implementation always upgrades from local changes to module level
+ // changes due to the way metadata cloning is done. See
+ // BuildDebugInfoToIdentityMap for more details.
+ const auto RemapFlag = RF_None;
CloneFunctionMetadataInto(NewFunc, OldFunc, VMap, RemapFlag, TypeMapper,
- Materializer);
+ Materializer, &IdentityMD);
CloneFunctionBodyInto(NewFunc, OldFunc, VMap, RemapFlag, Returns, NameSuffix,
- CodeInfo, TypeMapper, Materializer);
+ CodeInfo, TypeMapper, Materializer, &IdentityMD);
// Only update !llvm.dbg.cu for DifferentModule (not CloneModule). In the
// same module, the compile unit will already be listed (or not). When
diff --git a/llvm/lib/Transforms/Utils/ValueMapper.cpp b/llvm/lib/Transforms/Utils/ValueMapper.cpp
index 3faea48466ba9d..f6ab5bd284597a 100644
--- a/llvm/lib/Transforms/Utils/ValueMapper.cpp
+++ b/llvm/lib/Transforms/Utils/ValueMapper.cpp
@@ -120,12 +120,14 @@ class Mapper {
SmallVector<WorklistEntry, 4> Worklist;
SmallVector<DelayedBasicBlock, 1> DelayedBBs;
SmallVector<Constant *, 16> AppendingInits;
+ const MetadataSetTy *IdentityMD;
public:
Mapper(ValueToValueMapTy &VM, RemapFlags Flags,
- ValueMapTypeRemapper *TypeMapper, ValueMaterializer *Materializer)
+ ValueMapTypeRemapper *TypeMapper, ValueMaterializer *Materializer,
+ const MetadataSetTy *IdentityMD)
: Flags(Flags), TypeMapper(TypeMapper),
- MCs(1, MappingContext(VM, Materializer)) {}
+ MCs(1, MappingContext(VM, Materializer)), IdentityMD(IdentityMD) {}
/// ValueMapper should explicitly call \a flush() before destruction.
~Mapper() { assert(!hasWorkToDo() && "Expected to be flushed"); }
@@ -899,6 +901,14 @@ std::optional<Metadata *> Mapper::mapSimpleMetadata(const Metadata *MD) {
return wrapConstantAsMetadata(*CMD, mapValue(CMD->getValue()));
}
+ // Map metadata from IdentityMD on first use. We need to add these nodes to
+ // the mapping as otherwise metadata nodes numbering gets messed up. This is
+ // still economical because the amount of data in IdentityMD may be a lot
+ // larger than what will actually get used.
+ if (IdentityMD && IdentityMD->contains(MD)) {
+ return getVM().MD()[MD] = TrackingMDRef(const_cast<Metadata *>(MD));
+ }
+
assert(isa<MDNode>(MD) && "Expected a metadata node");
return std::nullopt;
@@ -1198,8 +1208,9 @@ class FlushingMapper {
ValueMapper::ValueMapper(ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
- ValueMaterializer *Materializer)
- : pImpl(new Mapper(VM, Flags, TypeMapper, Materializer)) {}
+ ValueMaterializer *Materializer,
+ const MetadataSetTy *IdentityMD)
+ : pImpl(new Mapper(VM, Flags, TypeMapper, Materializer, IdentityMD)) {}
ValueMapper::~ValueMapper() { delete getAsMapper(pImpl); }
>From a9330650c1fbd3f0013c1de4102fe759767934de Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Tue, 19 Nov 2024 17:19:27 -0700
Subject: [PATCH 09/11] [Coro] Prebuild a global debug info set and share it
between all coroutine clones
Summary:
CoroCloner, by calling into CloneFunctionInto, does a lot of repeated work priming DIFinder and building
a list of global debug info metadata. For programs compiled with full debug info this gets very
expensive.
This diff builds the data once and shares it between all clones.
Anecdata for a sample cpp source file compiled with full debug info:
| | Baseline | IdentityMD set | Prebuilt GlobalDI (cur.) |
|-----------------+----------+----------------+--------------------------|
| CoroSplitPass | 306ms | 221ms | 68ms |
| CoroCloner | 101ms | 72ms | 0.5ms |
| CollectGlobalDI | - | - | 63ms |
|-----------------+----------+----------------+--------------------------|
| Speed up | 1x | 1.4x | 4.5x |
Note that CollectGlobalDI happens once *per coroutine* rather than per clone.
Test Plan:
ninja check-llvm-unit
ninja check-llvm
Compiled a sample internal source file, checked time trace output for scope timings.
---
llvm/lib/Transforms/Coroutines/CoroSplit.cpp | 77 +++++++++++++++-----
1 file changed, 60 insertions(+), 17 deletions(-)
diff --git a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
index 8a5bae9f6f0d4b..99f0b0b830b571 100644
--- a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
+++ b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
@@ -42,6 +42,7 @@
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DebugInfo.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Function.h"
@@ -84,6 +85,23 @@ using namespace llvm;
namespace {
+/// 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) {
+ TimeTraceScope FunctionScope("CollectGlobalDebugInfo");
+
+ DebugInfoFinder DIFinder;
+ DISubprogram *SPClonedWithinModule = CollectDebugInfoForCloning(
+ F, CloneFunctionChangeType::LocalChangesOnly, DIFinder);
+
+ FindDebugInfoToIdentityMap(GlobalDebugInfo,
+ CloneFunctionChangeType::LocalChangesOnly,
+ DIFinder, SPClonedWithinModule);
+}
+
/// A little helper class for building
class CoroCloner {
public:
@@ -120,21 +138,26 @@ class CoroCloner {
TargetTransformInfo &TTI;
+ const MetadataSetTy &GlobalDebugInfo;
+
/// Create a cloner for a switch lowering.
CoroCloner(Function &OrigF, const Twine &Suffix, coro::Shape &Shape,
- Kind FKind, TargetTransformInfo &TTI)
+ Kind FKind, TargetTransformInfo &TTI,
+ const MetadataSetTy &GlobalDebugInfo)
: OrigF(OrigF), NewF(nullptr), Suffix(Suffix), Shape(Shape), FKind(FKind),
- Builder(OrigF.getContext()), TTI(TTI) {
+ Builder(OrigF.getContext()), TTI(TTI),
+ GlobalDebugInfo(GlobalDebugInfo) {
assert(Shape.ABI == coro::ABI::Switch);
}
/// Create a cloner for a continuation lowering.
CoroCloner(Function &OrigF, const Twine &Suffix, coro::Shape &Shape,
Function *NewF, AnyCoroSuspendInst *ActiveSuspend,
- TargetTransformInfo &TTI)
+ TargetTransformInfo &TTI, const MetadataSetTy &GlobalDebugInfo)
: OrigF(OrigF), NewF(NewF), Suffix(Suffix), Shape(Shape),
FKind(Shape.ABI == coro::ABI::Async ? Kind::Async : Kind::Continuation),
- Builder(OrigF.getContext()), ActiveSuspend(ActiveSuspend), TTI(TTI) {
+ Builder(OrigF.getContext()), ActiveSuspend(ActiveSuspend), TTI(TTI),
+ GlobalDebugInfo(GlobalDebugInfo) {
assert(Shape.ABI == coro::ABI::Retcon ||
Shape.ABI == coro::ABI::RetconOnce || Shape.ABI == coro::ABI::Async);
assert(NewF && "need existing function for continuation");
@@ -145,10 +168,11 @@ class CoroCloner {
/// Create a clone for a switch lowering.
static Function *createClone(Function &OrigF, const Twine &Suffix,
coro::Shape &Shape, Kind FKind,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const MetadataSetTy &GlobalDebugInfo) {
TimeTraceScope FunctionScope("CoroCloner");
- CoroCloner Cloner(OrigF, Suffix, Shape, FKind, TTI);
+ CoroCloner Cloner(OrigF, Suffix, Shape, FKind, TTI, GlobalDebugInfo);
Cloner.create();
return Cloner.getFunction();
}
@@ -157,10 +181,12 @@ class CoroCloner {
static Function *createClone(Function &OrigF, const Twine &Suffix,
coro::Shape &Shape, Function *NewF,
AnyCoroSuspendInst *ActiveSuspend,
- TargetTransformInfo &TTI) {
+ TargetTransformInfo &TTI,
+ const MetadataSetTy &GlobalDebugInfo) {
TimeTraceScope FunctionScope("CoroCloner");
- CoroCloner Cloner(OrigF, Suffix, Shape, NewF, ActiveSuspend, TTI);
+ CoroCloner Cloner(OrigF, Suffix, Shape, NewF, ActiveSuspend, TTI,
+ GlobalDebugInfo);
Cloner.create();
return Cloner.getFunction();
}
@@ -1016,8 +1042,11 @@ void CoroCloner::create() {
auto savedLinkage = NewF->getLinkage();
NewF->setLinkage(llvm::GlobalValue::ExternalLinkage);
- CloneFunctionInto(NewF, &OrigF, VMap,
- CloneFunctionChangeType::LocalChangesOnly, Returns);
+ CloneFunctionAttributesInto(NewF, &OrigF, VMap, false);
+ CloneFunctionMetadataInto(NewF, &OrigF, VMap, RF_None, nullptr, nullptr,
+ &GlobalDebugInfo);
+ CloneFunctionBodyInto(NewF, &OrigF, VMap, RF_None, Returns, "", nullptr,
+ nullptr, nullptr, &GlobalDebugInfo);
auto &Context = NewF->getContext();
@@ -1491,16 +1520,22 @@ struct SwitchCoroutineSplitter {
TargetTransformInfo &TTI) {
assert(Shape.ABI == coro::ABI::Switch);
+ MetadataSetTy GlobalDebugInfo;
+ collectGlobalDebugInfo(F, GlobalDebugInfo);
+
// Create a resume clone by cloning the body of the original function,
// setting new entry block and replacing coro.suspend an appropriate value
// to force resume or cleanup pass for every suspend point.
createResumeEntryBlock(F, Shape);
- auto *ResumeClone = CoroCloner::createClone(
- F, ".resume", Shape, CoroCloner::Kind::SwitchResume, TTI);
- auto *DestroyClone = CoroCloner::createClone(
- F, ".destroy", Shape, CoroCloner::Kind::SwitchUnwind, TTI);
+ auto *ResumeClone = CoroCloner::createClone(F, ".resume", Shape,
+ CoroCloner::Kind::SwitchResume,
+ TTI, GlobalDebugInfo);
+ auto *DestroyClone = CoroCloner::createClone(F, ".destroy", Shape,
+ CoroCloner::Kind::SwitchUnwind,
+ TTI, GlobalDebugInfo);
auto *CleanupClone = CoroCloner::createClone(
- F, ".cleanup", Shape, CoroCloner::Kind::SwitchCleanup, TTI);
+ F, ".cleanup", Shape, CoroCloner::Kind::SwitchCleanup, TTI,
+ GlobalDebugInfo);
postSplitCleanup(*ResumeClone);
postSplitCleanup(*DestroyClone);
@@ -1885,12 +1920,16 @@ void coro::AsyncABI::splitCoroutine(Function &F, coro::Shape &Shape,
}
assert(Clones.size() == Shape.CoroSuspends.size());
+
+ MetadataSetTy GlobalDebugInfo;
+ collectGlobalDebugInfo(F, GlobalDebugInfo);
+
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto *Suspend = CS;
auto *Clone = Clones[Idx];
CoroCloner::createClone(F, "resume." + Twine(Idx), Shape, Clone, Suspend,
- TTI);
+ TTI, GlobalDebugInfo);
}
}
@@ -2016,12 +2055,16 @@ void coro::AnyRetconABI::splitCoroutine(Function &F, coro::Shape &Shape,
}
assert(Clones.size() == Shape.CoroSuspends.size());
+
+ MetadataSetTy GlobalDebugInfo;
+ collectGlobalDebugInfo(F, GlobalDebugInfo);
+
for (auto [Idx, CS] : llvm::enumerate(Shape.CoroSuspends)) {
auto Suspend = CS;
auto Clone = Clones[Idx];
CoroCloner::createClone(F, "resume." + Twine(Idx), Shape, Clone, Suspend,
- TTI);
+ TTI, GlobalDebugInfo);
}
}
>From 3982c343138b246a2e4cedf7edef5fbf0c3558d4 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Sun, 15 Sep 2024 10:51:38 -0700
Subject: [PATCH 10/11] [Analysis] Add DebugInfoCache analysis
Summary:
The analysis simply primes and caches DebugInfoFinders for each DICompileUnit in a module. This
allows (future) callers like CoroSplitPass to compute global debug info metadata (required for
coroutine function cloning) much faster. Specifically, pay the price of DICompileUnit processing
only once per compile unit, rather than once per coroutine.
Test Plan:
Added a smoke test for the new analysis
ninja check-llvm-unit check-llvm
---
llvm/include/llvm/Analysis/DebugInfoCache.h | 50 +++++
llvm/include/llvm/IR/DebugInfo.h | 4 +-
llvm/lib/Analysis/CMakeLists.txt | 1 +
llvm/lib/Analysis/DebugInfoCache.cpp | 47 ++++
llvm/lib/Passes/PassBuilder.cpp | 1 +
llvm/lib/Passes/PassRegistry.def | 1 +
llvm/unittests/Analysis/CMakeLists.txt | 1 +
.../unittests/Analysis/DebugInfoCacheTest.cpp | 211 ++++++++++++++++++
8 files changed, 315 insertions(+), 1 deletion(-)
create mode 100644 llvm/include/llvm/Analysis/DebugInfoCache.h
create mode 100644 llvm/lib/Analysis/DebugInfoCache.cpp
create mode 100644 llvm/unittests/Analysis/DebugInfoCacheTest.cpp
diff --git a/llvm/include/llvm/Analysis/DebugInfoCache.h b/llvm/include/llvm/Analysis/DebugInfoCache.h
new file mode 100644
index 00000000000000..dbd6802c99ea01
--- /dev/null
+++ b/llvm/include/llvm/Analysis/DebugInfoCache.h
@@ -0,0 +1,50 @@
+//===- llvm/Analysis/DebugInfoCache.h - debug info cache --------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains an analysis that builds a cache of debug info for each
+// DICompileUnit in a module.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_ANALYSIS_DEBUGINFOCACHE_H
+#define LLVM_ANALYSIS_DEBUGINFOCACHE_H
+
+#include "llvm/IR/DebugInfo.h"
+#include "llvm/IR/PassManager.h"
+
+namespace llvm {
+
+/// Processes and caches debug info for each DICompileUnit in a module.
+///
+/// The result of the analysis is a set of DebugInfoFinders primed on their
+/// respective DICompileUnit. Such DebugInfoFinders can be used to speed up
+/// function cloning which otherwise requires an expensive traversal of
+/// DICompileUnit-level debug info. See an example usage in CoroSplit.
+class DebugInfoCache {
+public:
+ using DIFinderCache = SmallDenseMap<const DICompileUnit *, DebugInfoFinder>;
+ DIFinderCache Result;
+
+ DebugInfoCache(const Module &M);
+
+ bool invalidate(Module &, const PreservedAnalyses &,
+ ModuleAnalysisManager::Invalidator &);
+};
+
+class DebugInfoCacheAnalysis
+ : public AnalysisInfoMixin<DebugInfoCacheAnalysis> {
+ friend AnalysisInfoMixin<DebugInfoCacheAnalysis>;
+ static AnalysisKey Key;
+
+public:
+ using Result = DebugInfoCache;
+ Result run(Module &M, ModuleAnalysisManager &);
+};
+} // namespace llvm
+
+#endif
diff --git a/llvm/include/llvm/IR/DebugInfo.h b/llvm/include/llvm/IR/DebugInfo.h
index 73f45c3769be44..11907fbb7f20b3 100644
--- a/llvm/include/llvm/IR/DebugInfo.h
+++ b/llvm/include/llvm/IR/DebugInfo.h
@@ -120,11 +120,13 @@ class DebugInfoFinder {
/// Process subprogram.
void processSubprogram(DISubprogram *SP);
+ /// Process a compile unit.
+ void processCompileUnit(DICompileUnit *CU);
+
/// Clear all lists.
void reset();
private:
- void processCompileUnit(DICompileUnit *CU);
void processScope(DIScope *Scope);
void processType(DIType *DT);
bool addCompileUnit(DICompileUnit *CU);
diff --git a/llvm/lib/Analysis/CMakeLists.txt b/llvm/lib/Analysis/CMakeLists.txt
index 0db5b80f336cb5..db9a569e301563 100644
--- a/llvm/lib/Analysis/CMakeLists.txt
+++ b/llvm/lib/Analysis/CMakeLists.txt
@@ -52,6 +52,7 @@ add_llvm_component_library(LLVMAnalysis
DDGPrinter.cpp
ConstraintSystem.cpp
Delinearization.cpp
+ DebugInfoCache.cpp
DemandedBits.cpp
DependenceAnalysis.cpp
DependenceGraphBuilder.cpp
diff --git a/llvm/lib/Analysis/DebugInfoCache.cpp b/llvm/lib/Analysis/DebugInfoCache.cpp
new file mode 100644
index 00000000000000..c1a3e89f0a6ccf
--- /dev/null
+++ b/llvm/lib/Analysis/DebugInfoCache.cpp
@@ -0,0 +1,47 @@
+//===- llvm/Analysis/DebugInfoCache.cpp - debug info cache ----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains an analysis that builds a cache of debug info for each
+// DICompileUnit in a module.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/DebugInfoCache.h"
+#include "llvm/IR/Module.h"
+
+using namespace llvm;
+
+namespace {
+DebugInfoFinder processCompileUnit(DICompileUnit *CU) {
+ DebugInfoFinder DIFinder;
+ DIFinder.processCompileUnit(CU);
+
+ return DIFinder;
+}
+} // namespace
+
+DebugInfoCache::DebugInfoCache(const Module &M) {
+ for (const auto CU : M.debug_compile_units()) {
+ auto DIFinder = processCompileUnit(CU);
+ Result[CU] = std::move(DIFinder);
+ }
+}
+
+bool DebugInfoCache::invalidate(Module &M, const PreservedAnalyses &PA,
+ ModuleAnalysisManager::Invalidator &) {
+ // Check whether the analysis has been explicitly invalidated. Otherwise, it's
+ // stateless and remains preserved.
+ auto PAC = PA.getChecker<DebugInfoCacheAnalysis>();
+ return !PAC.preservedWhenStateless();
+}
+
+AnalysisKey DebugInfoCacheAnalysis::Key;
+
+DebugInfoCache DebugInfoCacheAnalysis::run(Module &M, ModuleAnalysisManager &) {
+ return DebugInfoCache(M);
+}
diff --git a/llvm/lib/Passes/PassBuilder.cpp b/llvm/lib/Passes/PassBuilder.cpp
index cf7ceed63607a6..3a970315f2520a 100644
--- a/llvm/lib/Passes/PassBuilder.cpp
+++ b/llvm/lib/Passes/PassBuilder.cpp
@@ -34,6 +34,7 @@
#include "llvm/Analysis/DDGPrinter.h"
#include "llvm/Analysis/DXILMetadataAnalysis.h"
#include "llvm/Analysis/DXILResource.h"
+#include "llvm/Analysis/DebugInfoCache.h"
#include "llvm/Analysis/Delinearization.h"
#include "llvm/Analysis/DemandedBits.h"
#include "llvm/Analysis/DependenceAnalysis.h"
diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def
index 7c3798f6462a46..d697c7c8ab83b8 100644
--- a/llvm/lib/Passes/PassRegistry.def
+++ b/llvm/lib/Passes/PassRegistry.def
@@ -21,6 +21,7 @@
MODULE_ANALYSIS("callgraph", CallGraphAnalysis())
MODULE_ANALYSIS("collector-metadata", CollectorMetadataAnalysis())
MODULE_ANALYSIS("ctx-prof-analysis", CtxProfAnalysis())
+MODULE_ANALYSIS("debug-info-cache", DebugInfoCacheAnalysis())
MODULE_ANALYSIS("dxil-metadata", DXILMetadataAnalysis())
MODULE_ANALYSIS("dxil-resource", DXILResourceAnalysis())
MODULE_ANALYSIS("inline-advisor", InlineAdvisorAnalysis())
diff --git a/llvm/unittests/Analysis/CMakeLists.txt b/llvm/unittests/Analysis/CMakeLists.txt
index 76d16513d93417..73694a1d6ba297 100644
--- a/llvm/unittests/Analysis/CMakeLists.txt
+++ b/llvm/unittests/Analysis/CMakeLists.txt
@@ -25,6 +25,7 @@ set(ANALYSIS_TEST_SOURCES
ConstraintSystemTest.cpp
CtxProfAnalysisTest.cpp
DDGTest.cpp
+ DebugInfoCacheTest.cpp
DomTreeUpdaterTest.cpp
DXILResourceTest.cpp
GraphWriterTest.cpp
diff --git a/llvm/unittests/Analysis/DebugInfoCacheTest.cpp b/llvm/unittests/Analysis/DebugInfoCacheTest.cpp
new file mode 100644
index 00000000000000..b32e4cb543158a
--- /dev/null
+++ b/llvm/unittests/Analysis/DebugInfoCacheTest.cpp
@@ -0,0 +1,211 @@
+//===- DebugInfoCacheTest.cpp - DebugInfoCache unit tests -----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/DebugInfoCache.h"
+#include "llvm/AsmParser/Parser.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Verifier.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/raw_ostream.h"
+#include "gtest/gtest.h"
+
+namespace llvm {
+namespace {
+
+// Forward declare the assembly
+extern StringRef MultiCUModule;
+
+const DICompileUnit *findCU(const Module &M, StringRef FileName) {
+ for (const auto CU : M.debug_compile_units()) {
+ if (CU->getFilename() == FileName)
+ return CU;
+ }
+
+ return nullptr;
+}
+
+class DebugInfoCacheTest : public testing::Test {
+protected:
+ LLVMContext C;
+
+ std::unique_ptr<Module> makeModule(StringRef Assembly) {
+ SMDiagnostic Err;
+ auto M = parseAssemblyString(Assembly, Err, C);
+ if (!M)
+ Err.print("DebugInfoCacheTest", errs());
+
+ verifyModule(*M, &errs());
+ return M;
+ }
+};
+
+TEST_F(DebugInfoCacheTest, TestEmpty) {
+ auto M = makeModule("");
+ DebugInfoCache DIC{*M};
+ EXPECT_EQ(DIC.Result.size(), 0u);
+}
+
+TEST_F(DebugInfoCacheTest, TestMultiCU) {
+ auto M = makeModule(MultiCUModule);
+ DebugInfoCache DIC{*M};
+ EXPECT_EQ(DIC.Result.size(), 2u);
+
+ auto *File1CU = findCU(*M, "file1.cpp");
+ EXPECT_NE(File1CU, nullptr);
+
+ auto File1DIFinder = DIC.Result.find(File1CU);
+ EXPECT_NE(File1DIFinder, DIC.Result.end());
+
+ EXPECT_EQ(File1DIFinder->getSecond().compile_unit_count(), 1u);
+ EXPECT_EQ(File1DIFinder->getSecond().type_count(), 6u);
+ EXPECT_EQ(File1DIFinder->getSecond().subprogram_count(), 0u);
+ EXPECT_EQ(File1DIFinder->getSecond().scope_count(), 1u);
+
+ auto *File2CU = findCU(*M, "file2.cpp");
+ EXPECT_NE(File1CU, nullptr);
+
+ auto File2DIFinder = DIC.Result.find(File2CU);
+ EXPECT_NE(File2DIFinder, DIC.Result.end());
+
+ EXPECT_EQ(File2DIFinder->getSecond().compile_unit_count(), 1u);
+ EXPECT_EQ(File2DIFinder->getSecond().type_count(), 2u);
+ EXPECT_EQ(File2DIFinder->getSecond().subprogram_count(), 0u);
+ EXPECT_EQ(File2DIFinder->getSecond().scope_count(), 2u);
+}
+
+/* Generated roughly by
+file1.cpp:
+struct file1_extern_type1;
+struct file1_extern_type2;
+
+namespace file1 {
+typedef struct file1_type1 { int x; float y; } file1_type1;
+file1_type1 global{0, 1.};
+} // file1
+
+extern struct file1_extern_type1 *file1_extern_func1(struct
+file1_extern_type2*);
+
+file1::file1_type1 file1_func1(file1::file1_type1 x) { return x; }
+--------
+file2.cpp:
+struct file2_extern_type1;
+struct file2_extern_type2;
+
+namespace file2 {
+typedef struct file2_type1 { float x; float y; } file2_type1;
+enum class file2_type2 { opt1, opt2 };
+
+namespace inner {
+file2_type2 inner_global{file2_type2::opt2};
+} // inner
+} // file2
+
+extern struct file2_extern_type1 *file2_extern_func1(struct
+file2_extern_type2*);
+
+file2::file2_type1 file2_func1(file2::file2_type1 x, file2::file2_type2 y) {
+return x; }
+--------
+$ clang -S -emit-llvm file*.cpp
+$ llvm-link -S -o single.ll file*.ll
+*/
+StringRef MultiCUModule = R"""(
+%"struct.file1::file1_type1" = type { i32, float }
+%"struct.file2::file2_type1" = type { float, float }
+
+ at _ZN5file16globalE = dso_local global %"struct.file1::file1_type1" { i32 0, float 1.000000e+00 }, align 4, !dbg !0
+ at _ZN5file25inner12inner_globalE = dso_local global i32 1, align 4, !dbg !11
+
+define dso_local i64 @_Z11file1_func1N5file111file1_type1E(i64 %0) !dbg !33 {
+ %2 = alloca %"struct.file1::file1_type1", align 4
+ %3 = alloca %"struct.file1::file1_type1", align 4
+ store i64 %0, ptr %3, align 4
+ #dbg_declare(ptr %3, !37, !DIExpression(), !38)
+ call void @llvm.memcpy.p0.p0.i64(ptr align 4 %2, ptr align 4 %3, i64 8, i1 false), !dbg !39
+ %4 = load i64, ptr %2, align 4, !dbg !40
+ ret i64 %4, !dbg !40
+}
+
+declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg)
+
+define dso_local <2 x float> @_Z11file2_func1N5file211file2_type1ENS_11file2_type2E(<2 x float> %0, i32 noundef %1) !dbg !41 {
+ %3 = alloca %"struct.file2::file2_type1", align 4
+ %4 = alloca %"struct.file2::file2_type1", align 4
+ %5 = alloca i32, align 4
+ store <2 x float> %0, ptr %4, align 4
+ #dbg_declare(ptr %4, !49, !DIExpression(), !50)
+ store i32 %1, ptr %5, align 4
+ #dbg_declare(ptr %5, !51, !DIExpression(), !52)
+ call void @llvm.memcpy.p0.p0.i64(ptr align 4 %3, ptr align 4 %4, i64 8, i1 false), !dbg !53
+ %6 = load <2 x float>, ptr %3, align 4, !dbg !54
+ ret <2 x float> %6, !dbg !54
+}
+
+!llvm.dbg.cu = !{!20, !22}
+!llvm.ident = !{!25, !25}
+!llvm.module.flags = !{!26, !27, !28, !29, !30, !31, !32}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "global", linkageName: "_ZN5file16globalE", scope: !2, file: !3, line: 6, type: !4, isLocal: false, isDefinition: true)
+!2 = !DINamespace(name: "file1", scope: null)
+!3 = !DIFile(filename: "file1.cpp", directory: "")
+!4 = !DIDerivedType(tag: DW_TAG_typedef, name: "file1_type1", scope: !2, file: !3, line: 5, baseType: !5)
+!5 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "file1_type1", scope: !2, file: !3, line: 5, size: 64, flags: DIFlagTypePassByValue, elements: !6, identifier: "_ZTSN5file111file1_type1E")
+!6 = !{!7, !9}
+!7 = !DIDerivedType(tag: DW_TAG_member, name: "x", scope: !5, file: !3, line: 5, baseType: !8, size: 32)
+!8 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!9 = !DIDerivedType(tag: DW_TAG_member, name: "y", scope: !5, file: !3, line: 5, baseType: !10, size: 32, offset: 32)
+!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
+!11 = !DIGlobalVariableExpression(var: !12, expr: !DIExpression())
+!12 = distinct !DIGlobalVariable(name: "inner_global", linkageName: "_ZN5file25inner12inner_globalE", scope: !13, file: !15, line: 9, type: !16, isLocal: false, isDefinition: true)
+!13 = !DINamespace(name: "inner", scope: !14)
+!14 = !DINamespace(name: "file2", scope: null)
+!15 = !DIFile(filename: "file2.cpp", directory: "")
+!16 = distinct !DICompositeType(tag: DW_TAG_enumeration_type, name: "file2_type2", scope: !14, file: !15, line: 6, baseType: !8, size: 32, flags: DIFlagEnumClass, elements: !17, identifier: "_ZTSN5file211file2_type2E")
+!17 = !{!18, !19}
+!18 = !DIEnumerator(name: "opt1", value: 0)
+!19 = !DIEnumerator(name: "opt2", value: 1)
+!20 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !3, isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, globals: !21, splitDebugInlining: false, nameTableKind: None)
+!21 = !{!0}
+!22 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !15, isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !23, globals: !24, splitDebugInlining: false, nameTableKind: None)
+!23 = !{!16}
+!24 = !{!11}
+!25 = !{!"clang"}
+!26 = !{i32 7, !"Dwarf Version", i32 5}
+!27 = !{i32 2, !"Debug Info Version", i32 3}
+!28 = !{i32 1, !"wchar_size", i32 4}
+!29 = !{i32 8, !"PIC Level", i32 2}
+!30 = !{i32 7, !"PIE Level", i32 2}
+!31 = !{i32 7, !"uwtable", i32 2}
+!32 = !{i32 7, !"frame-pointer", i32 2}
+!33 = distinct !DISubprogram(name: "file1_func1", linkageName: "_Z11file1_func1N5file111file1_type1E", scope: !3, file: !3, line: 11, type: !34, scopeLine: 11, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !20, retainedNodes: !36)
+!34 = !DISubroutineType(types: !35)
+!35 = !{!4, !4}
+!36 = !{}
+!37 = !DILocalVariable(name: "x", arg: 1, scope: !33, file: !3, line: 11, type: !4)
+!38 = !DILocation(line: 11, column: 51, scope: !33)
+!39 = !DILocation(line: 11, column: 63, scope: !33)
+!40 = !DILocation(line: 11, column: 56, scope: !33)
+!41 = distinct !DISubprogram(name: "file2_func1", linkageName: "_Z11file2_func1N5file211file2_type1ENS_11file2_type2E", scope: !15, file: !15, line: 15, type: !42, scopeLine: 15, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !22, retainedNodes: !36)
+!42 = !DISubroutineType(types: !43)
+!43 = !{!44, !44, !16}
+!44 = !DIDerivedType(tag: DW_TAG_typedef, name: "file2_type1", scope: !14, file: !15, line: 5, baseType: !45)
+!45 = distinct !DICompositeType(tag: DW_TAG_structure_type, name: "file2_type1", scope: !14, file: !15, line: 5, size: 64, flags: DIFlagTypePassByValue, elements: !46, identifier: "_ZTSN5file211file2_type1E")
+!46 = !{!47, !48}
+!47 = !DIDerivedType(tag: DW_TAG_member, name: "x", scope: !45, file: !15, line: 5, baseType: !10, size: 32)
+!48 = !DIDerivedType(tag: DW_TAG_member, name: "y", scope: !45, file: !15, line: 5, baseType: !10, size: 32, offset: 32)
+!49 = !DILocalVariable(name: "x", arg: 1, scope: !41, file: !15, line: 15, type: !44)
+!50 = !DILocation(line: 15, column: 51, scope: !41)
+!51 = !DILocalVariable(name: "y", arg: 2, scope: !41, file: !15, line: 15, type: !16)
+!52 = !DILocation(line: 15, column: 73, scope: !41)
+!53 = !DILocation(line: 15, column: 85, scope: !41)
+!54 = !DILocation(line: 15, column: 78, scope: !41)
+)""";
+} // namespace
+} // namespace llvm
>From f43a8bcf45c169f617952f880a176c8debddc085 Mon Sep 17 00:00:00 2001
From: Artem Pianykh <arr at fb.com>
Date: Sun, 15 Sep 2024 11:00:00 -0700
Subject: [PATCH 11/11] [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 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.
---
llvm/include/llvm/Transforms/Coroutines/ABI.h | 13 +++--
llvm/lib/Analysis/CGSCCPassManager.cpp | 7 +++
llvm/lib/Transforms/Coroutines/CoroSplit.cpp | 56 +++++++++++++++----
llvm/test/Other/new-pass-manager.ll | 1 +
llvm/test/Other/new-pm-defaults.ll | 1 +
llvm/test/Other/new-pm-lto-defaults.ll | 1 +
llvm/test/Other/new-pm-pgo-preinline.ll | 1 +
.../Other/new-pm-thinlto-postlink-defaults.ll | 1 +
.../new-pm-thinlto-postlink-pgo-defaults.ll | 1 +
...-pm-thinlto-postlink-samplepgo-defaults.ll | 1 +
.../Other/new-pm-thinlto-prelink-defaults.ll | 1 +
.../new-pm-thinlto-prelink-pgo-defaults.ll | 1 +
...w-pm-thinlto-prelink-samplepgo-defaults.ll | 1 +
.../Analysis/CGSCCPassManagerTest.cpp | 4 +-
14 files changed, 73 insertions(+), 17 deletions(-)
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 99f0b0b830b571..9ccd4d8f0aff25 100644
--- a/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
+++ b/llvm/lib/Transforms/Coroutines/CoroSplit.cpp
@@ -31,6 +31,7 @@
#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"
@@ -85,15 +86,39 @@ using namespace llvm;
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);
@@ -1517,11 +1542,11 @@ 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
@@ -1835,7 +1860,8 @@ 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
@@ -1922,7 +1948,7 @@ 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;
@@ -1935,7 +1961,8 @@ 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());
@@ -2057,7 +2084,7 @@ 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;
@@ -2111,13 +2138,15 @@ 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;
@@ -2142,7 +2171,7 @@ 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.
@@ -2339,6 +2368,9 @@ 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");
@@ -2375,7 +2407,7 @@ 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.
More information about the llvm-commits
mailing list