AMDGPU: Port AMDGPUAttributor to new pass manager (#71349) (d34a10a4) · Commits · llvm-doe / llvm-project

llvm/lib/Target/AMDGPU/AMDGPU.h

+12 −2

Original line number	Diff line number	Diff line
		@@ -17,6 +17,7 @@
		namespace llvm {

		class AMDGPUTargetMachine;
		class GCNTargetMachine;
		class TargetMachine;

		// GlobalISel passes
		@@ -86,8 +87,8 @@ extern char &AMDGPUMachineCFGStructurizerID;
		void initializeAMDGPUAlwaysInlinePass(PassRegistry&);

		Pass *createAMDGPUAnnotateKernelFeaturesPass();
		Pass *createAMDGPUAttributorPass();
		void initializeAMDGPUAttributorPass(PassRegistry &);
		Pass *createAMDGPUAttributorLegacyPass();
		void initializeAMDGPUAttributorLegacyPass(PassRegistry &);
		void initializeAMDGPUAnnotateKernelFeaturesPass(PassRegistry &);
		extern char &AMDGPUAnnotateKernelFeaturesID;

		@@ -262,6 +263,15 @@ public:
		PreservedAnalyses run(Function &, FunctionAnalysisManager &);
		};

		class AMDGPUAttributorPass : public PassInfoMixin<AMDGPUAttributorPass> {
		private:
		TargetMachine &TM;

		public:
		AMDGPUAttributorPass(TargetMachine &TM) : TM(TM){};
		PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
		};

		FunctionPass *createAMDGPUAnnotateUniformValues();

		ModulePass *createAMDGPUPrintfRuntimeBinding();

llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp

+67 −49

Original line number	Diff line number	Diff line
		@@ -933,24 +933,8 @@ static void addPreloadKernArgHint(Function &F, TargetMachine &TM) {
		}
		}

		class AMDGPUAttributor : public ModulePass {
		public:
		AMDGPUAttributor() : ModulePass(ID) {}

		/// doInitialization - Virtual method overridden by subclasses to do
		/// any necessary initialization before any pass is run.
		bool doInitialization(Module &) override {
		auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
		if (!TPC)
		report_fatal_error("TargetMachine is required");

		TM = &TPC->getTM<TargetMachine>();
		return false;
		}

		bool runOnModule(Module &M) override {
		static bool runImpl(Module &M, AnalysisGetter &AG, TargetMachine &TM) {
		SetVector<Function *> Functions;
		AnalysisGetter AG(this);
		for (Function &F : M) {
		if (!F.isIntrinsic())
		Functions.insert(&F);
		@@ -958,7 +942,7 @@ public:

		CallGraphUpdater CGUpdater;
		BumpPtrAllocator Allocator;
		AMDGPUInformationCache InfoCache(M, AG, Allocator, nullptr, *TM);
		AMDGPUInformationCache InfoCache(M, AG, Allocator, nullptr, TM);
		DenseSet<const char *> Allowed(
		{&AAAMDAttributes::ID, &AAUniformWorkGroupSize::ID,
		&AAPotentialValues::ID, &AAAMDFlatWorkGroupSize::ID,
		@@ -984,7 +968,7 @@ public:
		A.getOrCreateAAFor<AAAMDFlatWorkGroupSize>(IRPosition::function(F));
		A.getOrCreateAAFor<AAAMDWavesPerEU>(IRPosition::function(F));
		} else if (CC == CallingConv::AMDGPU_KERNEL) {
		addPreloadKernArgHint(F, *TM);
		addPreloadKernArgHint(F, TM);
		}
		}
		}
		@@ -993,6 +977,26 @@ public:
		return Change == ChangeStatus::CHANGED;
		}

		class AMDGPUAttributorLegacy : public ModulePass {
		public:
		AMDGPUAttributorLegacy() : ModulePass(ID) {}

		/// doInitialization - Virtual method overridden by subclasses to do
		/// any necessary initialization before any pass is run.
		bool doInitialization(Module &) override {
		auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
		if (!TPC)
		report_fatal_error("TargetMachine is required");

		TM = &TPC->getTM<TargetMachine>();
		return false;
		}

		bool runOnModule(Module &M) override {
		AnalysisGetter AG(this);
		return runImpl(M, AG, *TM);
		}

		void getAnalysisUsage(AnalysisUsage &AU) const override {
		AU.addRequired<CycleInfoWrapperPass>();
		}
		@@ -1003,11 +1007,25 @@ public:
		};
		} // namespace

		char AMDGPUAttributor::ID = 0;
		PreservedAnalyses llvm::AMDGPUAttributorPass::run(Module &M,
		ModuleAnalysisManager &AM) {

		FunctionAnalysisManager &FAM =
		AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
		AnalysisGetter AG(FAM);

		Pass *llvm::createAMDGPUAttributorPass() { return new AMDGPUAttributor(); }
		INITIALIZE_PASS_BEGIN(AMDGPUAttributor, DEBUG_TYPE, "AMDGPU Attributor", false,
		false)
		// TODO: Probably preserves CFG
		return runImpl(M, AG, TM) ? PreservedAnalyses::none()
		: PreservedAnalyses::all();
		}

		char AMDGPUAttributorLegacy::ID = 0;

		Pass *llvm::createAMDGPUAttributorLegacyPass() {
		return new AMDGPUAttributorLegacy();
		}
		INITIALIZE_PASS_BEGIN(AMDGPUAttributorLegacy, DEBUG_TYPE, "AMDGPU Attributor",
		false, false)
		INITIALIZE_PASS_DEPENDENCY(CycleInfoWrapperPass);
		INITIALIZE_PASS_END(AMDGPUAttributor, DEBUG_TYPE, "AMDGPU Attributor", false,
		false)
		INITIALIZE_PASS_END(AMDGPUAttributorLegacy, DEBUG_TYPE, "AMDGPU Attributor",
		false, false)

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

+6 −2

Original line number	Diff line number	Diff line
		@@ -381,7 +381,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
		initializeSILoadStoreOptimizerPass(*PR);
		initializeAMDGPUCtorDtorLoweringLegacyPass(*PR);
		initializeAMDGPUAlwaysInlinePass(*PR);
		initializeAMDGPUAttributorPass(*PR);
		initializeAMDGPUAttributorLegacyPass(*PR);
		initializeAMDGPUAnnotateKernelFeaturesPass(*PR);
		initializeAMDGPUAnnotateUniformValuesPass(*PR);
		initializeAMDGPUArgumentUsageInfoPass(*PR);
		@@ -610,6 +610,10 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
		PB.registerPipelineParsingCallback(
		[this](StringRef PassName, ModulePassManager &PM,
		ArrayRef<PassBuilder::PipelineElement>) {
		if (PassName == "amdgpu-attributor") {
		PM.addPass(AMDGPUAttributorPass(*this));
		return true;
		}
		if (PassName == "amdgpu-unify-metadata") {
		PM.addPass(AMDGPUUnifyMetadataPass());
		return true;
		@@ -1021,7 +1025,7 @@ void AMDGPUPassConfig::addIRPasses() {
		// AMDGPUAttributor infers lack of llvm.amdgcn.lds.kernel.id calls, so run
		// after their introduction
		if (TM.getOptLevel() > CodeGenOptLevel::None)
		addPass(createAMDGPUAttributorPass());
		addPass(createAMDGPUAttributorLegacyPass());

		if (TM.getOptLevel() > CodeGenOptLevel::None)
		addPass(createInferAddressSpacesPass());

llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll

+1 −1

Original line number	Diff line number	Diff line
		; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
		; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -amdgpu-annotate-kernel-features < %s \| FileCheck -check-prefixes=HSA,AKF_HSA %s
		; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -amdgpu-attributor < %s \| FileCheck -check-prefixes=HSA,ATTRIBUTOR_HSA %s
		; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -passes=amdgpu-attributor < %s \| FileCheck -check-prefixes=HSA,ATTRIBUTOR_HSA %s

		declare void @llvm.memcpy.p1.p4.i32(ptr addrspace(1) nocapture, ptr addrspace(4) nocapture, i32, i1) #0

llvm/test/CodeGen/AMDGPU/annotate-existing-abi-attributes.ll

+1 −1

Original line number	Diff line number	Diff line
		; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
		; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -amdgpu-attributor %s \| FileCheck %s
		; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -passes=amdgpu-attributor %s \| FileCheck %s

		; Check handling for pre-existing attributes on function declarations

Admin message