[clang] [clang-repl] : Fix clang-repl crash with --cuda flag (PR #136404)
Anutosh Bhat via cfe-commits
cfe-commits at lists.llvm.org
Thu Apr 24 08:26:53 PDT 2025
https://github.com/anutosh491 updated https://github.com/llvm/llvm-project/pull/136404
>From 6c64e64c4a3b56f50808cae244b29da1231525f1 Mon Sep 17 00:00:00 2001
From: anutosh491 <andersonbhat491 at gmail.com>
Date: Fri, 18 Apr 2025 18:45:00 +0530
Subject: [PATCH 1/3] Fix cuda flag with clang-repl
---
clang/include/clang/Interpreter/Interpreter.h | 13 ++--
clang/lib/Interpreter/DeviceOffload.cpp | 39 ++++++------
clang/lib/Interpreter/DeviceOffload.h | 4 +-
clang/lib/Interpreter/Interpreter.cpp | 59 ++++++++++++++-----
4 files changed, 72 insertions(+), 43 deletions(-)
diff --git a/clang/include/clang/Interpreter/Interpreter.h b/clang/include/clang/Interpreter/Interpreter.h
index b1b63aedf86ab..56213f88b9e30 100644
--- a/clang/include/clang/Interpreter/Interpreter.h
+++ b/clang/include/clang/Interpreter/Interpreter.h
@@ -41,6 +41,7 @@ class CXXRecordDecl;
class Decl;
class IncrementalExecutor;
class IncrementalParser;
+class IncrementalCUDADeviceParser;
/// Create a pre-configured \c CompilerInstance for incremental processing.
class IncrementalCompilerBuilder {
@@ -93,7 +94,10 @@ class Interpreter {
std::unique_ptr<IncrementalExecutor> IncrExecutor;
// An optional parser for CUDA offloading
- std::unique_ptr<IncrementalParser> DeviceParser;
+ std::unique_ptr<IncrementalCUDADeviceParser> DeviceParser;
+
+ // An optional action for CUDA offloading
+ std::unique_ptr<IncrementalAction> DeviceAct;
/// List containing information about each incrementally parsed piece of code.
std::list<PartialTranslationUnit> PTUs;
@@ -175,10 +179,11 @@ class Interpreter {
llvm::Expected<Expr *> ExtractValueFromExpr(Expr *E);
llvm::Expected<llvm::orc::ExecutorAddr> CompileDtorCall(CXXRecordDecl *CXXRD);
- CodeGenerator *getCodeGen() const;
- std::unique_ptr<llvm::Module> GenModule();
+ CodeGenerator *getCodeGen(IncrementalAction *Action = nullptr) const;
+ std::unique_ptr<llvm::Module> GenModule(IncrementalAction *Action = nullptr);
PartialTranslationUnit &RegisterPTU(TranslationUnitDecl *TU,
- std::unique_ptr<llvm::Module> M = {});
+ std::unique_ptr<llvm::Module> M = {},
+ IncrementalAction *Action = nullptr);
// A cache for the compiled destructors used to for de-allocation of managed
// clang::Values.
diff --git a/clang/lib/Interpreter/DeviceOffload.cpp b/clang/lib/Interpreter/DeviceOffload.cpp
index 1999d63d1aa04..d9b00787f038d 100644
--- a/clang/lib/Interpreter/DeviceOffload.cpp
+++ b/clang/lib/Interpreter/DeviceOffload.cpp
@@ -28,20 +28,21 @@ IncrementalCUDADeviceParser::IncrementalCUDADeviceParser(
std::unique_ptr<CompilerInstance> DeviceInstance,
CompilerInstance &HostInstance,
llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> FS,
- llvm::Error &Err, const std::list<PartialTranslationUnit> &PTUs)
+ llvm::Error &Err, std::list<PartialTranslationUnit> &PTUs)
: IncrementalParser(*DeviceInstance, Err), PTUs(PTUs), VFS(FS),
CodeGenOpts(HostInstance.getCodeGenOpts()),
- TargetOpts(HostInstance.getTargetOpts()) {
+ TargetOpts(DeviceInstance->getTargetOpts()) {
if (Err)
return;
- DeviceCI = std::move(DeviceInstance);
StringRef Arch = TargetOpts.CPU;
if (!Arch.starts_with("sm_") || Arch.substr(3).getAsInteger(10, SMVersion)) {
+ DeviceInstance.release();
Err = llvm::joinErrors(std::move(Err), llvm::make_error<llvm::StringError>(
"Invalid CUDA architecture",
llvm::inconvertibleErrorCode()));
return;
}
+ DeviceCI = std::move(DeviceInstance);
}
llvm::Expected<TranslationUnitDecl *>
@@ -50,25 +51,6 @@ IncrementalCUDADeviceParser::Parse(llvm::StringRef Input) {
if (!PTU)
return PTU.takeError();
- auto PTX = GeneratePTX();
- if (!PTX)
- return PTX.takeError();
-
- auto Err = GenerateFatbinary();
- if (Err)
- return std::move(Err);
-
- std::string FatbinFileName =
- "/incr_module_" + std::to_string(PTUs.size()) + ".fatbin";
- VFS->addFile(FatbinFileName, 0,
- llvm::MemoryBuffer::getMemBuffer(
- llvm::StringRef(FatbinContent.data(), FatbinContent.size()),
- "", false));
-
- CodeGenOpts.CudaGpuBinaryFileName = FatbinFileName;
-
- FatbinContent.clear();
-
return PTU;
}
@@ -172,6 +154,19 @@ llvm::Error IncrementalCUDADeviceParser::GenerateFatbinary() {
FatbinContent.append(PTXCode.begin(), PTXCode.end());
+ auto &PTU = PTUs.back();
+
+ std::string FatbinFileName = "/" + PTU.TheModule->getName().str() + ".fatbin";
+
+ VFS->addFile(FatbinFileName, 0,
+ llvm::MemoryBuffer::getMemBuffer(
+ llvm::StringRef(FatbinContent.data(), FatbinContent.size()),
+ "", false));
+
+ CodeGenOpts.CudaGpuBinaryFileName = FatbinFileName;
+
+ FatbinContent.clear();
+
return llvm::Error::success();
}
diff --git a/clang/lib/Interpreter/DeviceOffload.h b/clang/lib/Interpreter/DeviceOffload.h
index b9a1acab004c3..23d89046c09e1 100644
--- a/clang/lib/Interpreter/DeviceOffload.h
+++ b/clang/lib/Interpreter/DeviceOffload.h
@@ -24,14 +24,14 @@ class CodeGenOptions;
class TargetOptions;
class IncrementalCUDADeviceParser : public IncrementalParser {
- const std::list<PartialTranslationUnit> &PTUs;
+ std::list<PartialTranslationUnit> &PTUs;
public:
IncrementalCUDADeviceParser(
std::unique_ptr<CompilerInstance> DeviceInstance,
CompilerInstance &HostInstance,
llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> VFS,
- llvm::Error &Err, const std::list<PartialTranslationUnit> &PTUs);
+ llvm::Error &Err, std::list<PartialTranslationUnit> &PTUs);
llvm::Expected<TranslationUnitDecl *> Parse(llvm::StringRef Input) override;
diff --git a/clang/lib/Interpreter/Interpreter.cpp b/clang/lib/Interpreter/Interpreter.cpp
index f8c8d0a425659..f91563dd0378c 100644
--- a/clang/lib/Interpreter/Interpreter.cpp
+++ b/clang/lib/Interpreter/Interpreter.cpp
@@ -481,20 +481,34 @@ Interpreter::createWithCUDA(std::unique_ptr<CompilerInstance> CI,
OverlayVFS->pushOverlay(IMVFS);
CI->createFileManager(OverlayVFS);
- auto Interp = Interpreter::create(std::move(CI));
- if (auto E = Interp.takeError())
- return std::move(E);
+ llvm::Expected<std::unique_ptr<Interpreter>> InterpOrErr =
+ Interpreter::create(std::move(CI));
+ if (!InterpOrErr)
+ return InterpOrErr;
+
+ std::unique_ptr<Interpreter> Interp = std::move(*InterpOrErr);
llvm::Error Err = llvm::Error::success();
- auto DeviceParser = std::make_unique<IncrementalCUDADeviceParser>(
- std::move(DCI), *(*Interp)->getCompilerInstance(), IMVFS, Err,
- (*Interp)->PTUs);
+ llvm::LLVMContext &LLVMCtx = *Interp->TSCtx->getContext();
+
+ auto DeviceAct =
+ std::make_unique<IncrementalAction>(*DCI, LLVMCtx, Err, *Interp);
+
if (Err)
return std::move(Err);
- (*Interp)->DeviceParser = std::move(DeviceParser);
+ Interp->DeviceAct = std::move(DeviceAct);
+
+ DCI->ExecuteAction(*Interp->DeviceAct);
+
+ auto DeviceParser = std::make_unique<IncrementalCUDADeviceParser>(
+ std::move(DCI), *Interp->getCompilerInstance(), IMVFS, Err, Interp->PTUs);
+
+ if (Err)
+ return std::move(Err);
- return Interp;
+ Interp->DeviceParser = std::move(DeviceParser);
+ return std::move(Interp);
}
const CompilerInstance *Interpreter::getCompilerInstance() const {
@@ -532,15 +546,17 @@ size_t Interpreter::getEffectivePTUSize() const {
PartialTranslationUnit &
Interpreter::RegisterPTU(TranslationUnitDecl *TU,
- std::unique_ptr<llvm::Module> M /*={}*/) {
+ std::unique_ptr<llvm::Module> M /*={}*/,
+ IncrementalAction *Action) {
PTUs.emplace_back(PartialTranslationUnit());
PartialTranslationUnit &LastPTU = PTUs.back();
LastPTU.TUPart = TU;
if (!M)
- M = GenModule();
+ M = GenModule(Action);
- assert((!getCodeGen() || M) && "Must have a llvm::Module at this point");
+ assert((!getCodeGen(Action) || M) &&
+ "Must have a llvm::Module at this point");
LastPTU.TheModule = std::move(M);
LLVM_DEBUG(llvm::dbgs() << "compile-ptu " << PTUs.size() - 1
@@ -560,6 +576,16 @@ Interpreter::Parse(llvm::StringRef Code) {
llvm::Expected<TranslationUnitDecl *> DeviceTU = DeviceParser->Parse(Code);
if (auto E = DeviceTU.takeError())
return std::move(E);
+
+ RegisterPTU(*DeviceTU, nullptr, DeviceAct.get());
+
+ llvm::Expected<llvm::StringRef> PTX = DeviceParser->GeneratePTX();
+ if (!PTX)
+ return PTX.takeError();
+
+ llvm::Error Err = DeviceParser->GenerateFatbinary();
+ if (Err)
+ return std::move(Err);
}
// Tell the interpreter sliently ignore unused expressions since value
@@ -736,9 +762,10 @@ llvm::Error Interpreter::LoadDynamicLibrary(const char *name) {
return llvm::Error::success();
}
-std::unique_ptr<llvm::Module> Interpreter::GenModule() {
+std::unique_ptr<llvm::Module>
+Interpreter::GenModule(IncrementalAction *Action) {
static unsigned ID = 0;
- if (CodeGenerator *CG = getCodeGen()) {
+ if (CodeGenerator *CG = getCodeGen(Action)) {
// Clang's CodeGen is designed to work with a single llvm::Module. In many
// cases for convenience various CodeGen parts have a reference to the
// llvm::Module (TheModule or Module) which does not change when a new
@@ -760,8 +787,10 @@ std::unique_ptr<llvm::Module> Interpreter::GenModule() {
return nullptr;
}
-CodeGenerator *Interpreter::getCodeGen() const {
- FrontendAction *WrappedAct = Act->getWrapped();
+CodeGenerator *Interpreter::getCodeGen(IncrementalAction *Action) const {
+ if (!Action)
+ Action = Act.get();
+ FrontendAction *WrappedAct = Action->getWrapped();
if (!WrappedAct->hasIRSupport())
return nullptr;
return static_cast<CodeGenAction *>(WrappedAct)->getCodeGenerator();
>From a445870215b431bdd5b7675920b5d523491499b6 Mon Sep 17 00:00:00 2001
From: anutosh491 <andersonbhat491 at gmail.com>
Date: Thu, 24 Apr 2025 19:31:12 +0530
Subject: [PATCH 2/3] Address reviews
---
clang/lib/Interpreter/DeviceOffload.cpp | 5 ++---
clang/lib/Interpreter/DeviceOffload.h | 4 ++--
2 files changed, 4 insertions(+), 5 deletions(-)
diff --git a/clang/lib/Interpreter/DeviceOffload.cpp b/clang/lib/Interpreter/DeviceOffload.cpp
index d9b00787f038d..f35238f2b3734 100644
--- a/clang/lib/Interpreter/DeviceOffload.cpp
+++ b/clang/lib/Interpreter/DeviceOffload.cpp
@@ -28,7 +28,7 @@ IncrementalCUDADeviceParser::IncrementalCUDADeviceParser(
std::unique_ptr<CompilerInstance> DeviceInstance,
CompilerInstance &HostInstance,
llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> FS,
- llvm::Error &Err, std::list<PartialTranslationUnit> &PTUs)
+ llvm::Error &Err, const std::list<PartialTranslationUnit> &PTUs)
: IncrementalParser(*DeviceInstance, Err), PTUs(PTUs), VFS(FS),
CodeGenOpts(HostInstance.getCodeGenOpts()),
TargetOpts(DeviceInstance->getTargetOpts()) {
@@ -36,7 +36,6 @@ IncrementalCUDADeviceParser::IncrementalCUDADeviceParser(
return;
StringRef Arch = TargetOpts.CPU;
if (!Arch.starts_with("sm_") || Arch.substr(3).getAsInteger(10, SMVersion)) {
- DeviceInstance.release();
Err = llvm::joinErrors(std::move(Err), llvm::make_error<llvm::StringError>(
"Invalid CUDA architecture",
llvm::inconvertibleErrorCode()));
@@ -154,7 +153,7 @@ llvm::Error IncrementalCUDADeviceParser::GenerateFatbinary() {
FatbinContent.append(PTXCode.begin(), PTXCode.end());
- auto &PTU = PTUs.back();
+ const PartialTranslationUnit &PTU = PTUs.back();
std::string FatbinFileName = "/" + PTU.TheModule->getName().str() + ".fatbin";
diff --git a/clang/lib/Interpreter/DeviceOffload.h b/clang/lib/Interpreter/DeviceOffload.h
index 23d89046c09e1..b9a1acab004c3 100644
--- a/clang/lib/Interpreter/DeviceOffload.h
+++ b/clang/lib/Interpreter/DeviceOffload.h
@@ -24,14 +24,14 @@ class CodeGenOptions;
class TargetOptions;
class IncrementalCUDADeviceParser : public IncrementalParser {
- std::list<PartialTranslationUnit> &PTUs;
+ const std::list<PartialTranslationUnit> &PTUs;
public:
IncrementalCUDADeviceParser(
std::unique_ptr<CompilerInstance> DeviceInstance,
CompilerInstance &HostInstance,
llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> VFS,
- llvm::Error &Err, std::list<PartialTranslationUnit> &PTUs);
+ llvm::Error &Err, const std::list<PartialTranslationUnit> &PTUs);
llvm::Expected<TranslationUnitDecl *> Parse(llvm::StringRef Input) override;
>From 7aed4345e7d769e859daac0f64eef9f132d507e5 Mon Sep 17 00:00:00 2001
From: anutosh491 <andersonbhat491 at gmail.com>
Date: Thu, 24 Apr 2025 20:56:36 +0530
Subject: [PATCH 3/3] use base class for parse
---
clang/lib/Interpreter/DeviceOffload.cpp | 9 ---------
clang/lib/Interpreter/DeviceOffload.h | 2 --
2 files changed, 11 deletions(-)
diff --git a/clang/lib/Interpreter/DeviceOffload.cpp b/clang/lib/Interpreter/DeviceOffload.cpp
index f35238f2b3734..7d0125403ea52 100644
--- a/clang/lib/Interpreter/DeviceOffload.cpp
+++ b/clang/lib/Interpreter/DeviceOffload.cpp
@@ -44,15 +44,6 @@ IncrementalCUDADeviceParser::IncrementalCUDADeviceParser(
DeviceCI = std::move(DeviceInstance);
}
-llvm::Expected<TranslationUnitDecl *>
-IncrementalCUDADeviceParser::Parse(llvm::StringRef Input) {
- auto PTU = IncrementalParser::Parse(Input);
- if (!PTU)
- return PTU.takeError();
-
- return PTU;
-}
-
llvm::Expected<llvm::StringRef> IncrementalCUDADeviceParser::GeneratePTX() {
auto &PTU = PTUs.back();
std::string Error;
diff --git a/clang/lib/Interpreter/DeviceOffload.h b/clang/lib/Interpreter/DeviceOffload.h
index b9a1acab004c3..43645033c4840 100644
--- a/clang/lib/Interpreter/DeviceOffload.h
+++ b/clang/lib/Interpreter/DeviceOffload.h
@@ -33,8 +33,6 @@ class IncrementalCUDADeviceParser : public IncrementalParser {
llvm::IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> VFS,
llvm::Error &Err, const std::list<PartialTranslationUnit> &PTUs);
- llvm::Expected<TranslationUnitDecl *> Parse(llvm::StringRef Input) override;
-
// Generate PTX for the last PTU.
llvm::Expected<llvm::StringRef> GeneratePTX();
More information about the cfe-commits
mailing list