[CIR][CIRGen] Add time trace to several CIRGen pieces (#898)

Then we can observe the time consumed in different part of CIR. This
patch is not complete. But I think it is fine given we can always add
them easily.

Author: ChuanqiXu9

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -75,6 +75,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/TimeProfiler.h"
 
 #include <iterator>
 #include <numeric>
@@ -462,6 +463,19 @@ void CIRGenModule::setDSOLocal(CIRGlobalValueInterface GV) const {
 }
 
 void CIRGenModule::buildGlobal(GlobalDecl GD) {
+  llvm::TimeTraceScope scope("build CIR Global", [&]() -> std::string {
+    auto *ND = dyn_cast<NamedDecl>(GD.getDecl());
+    if (!ND)
+      // TODO: How to print decls which is not named decl?
+      return "Unnamed decl";
+
+    std::string Name;
+    llvm::raw_string_ostream OS(Name);
+    ND->getNameForDiagnostic(OS, getASTContext().getPrintingPolicy(),
+                             /*Qualified=*/true);
+    return Name;
+  });
+
   const auto *Global = cast<ValueDecl>(GD.getDecl());
 
   assert(!Global->hasAttr<IFuncAttr>() && "NYI");

clang/lib/CIR/CodeGen/CIRPasses.cpp

@@ -19,6 +19,8 @@
 #include "mlir/Support/LogicalResult.h"
 #include "mlir/Transforms/Passes.h"
 
+#include "llvm/Support/TimeProfiler.h"
+
 namespace cir {
 mlir::LogicalResult runCIRToCIRPasses(
     mlir::ModuleOp theModule, mlir::MLIRContext *mlirCtx,
@@ -29,6 +31,8 @@ mlir::LogicalResult runCIRToCIRPasses(
     bool enableCIRSimplify, bool flattenCIR, bool emitMLIR,
     bool enableCallConvLowering, bool enableMem2Reg) {
 
+  llvm::TimeTraceScope scope("CIR To CIR Passes");
+
   mlir::PassManager pm(mlirCtx);
   pm.addPass(mlir::createCIRCanonicalizePass());
 

clang/lib/CIR/Dialect/Transforms/CallConvLowering.cpp

@@ -18,6 +18,8 @@
 #define GEN_PASS_DEF_CALLCONVLOWERING
 #include "clang/CIR/Dialect/Passes.h.inc"
 
+#include "llvm/Support/TimeProfiler.h"
+
 namespace mlir {
 namespace cir {
 
@@ -30,6 +32,8 @@ struct CallConvLoweringPattern : public OpRewritePattern<FuncOp> {
 
   LogicalResult matchAndRewrite(FuncOp op,
                                 PatternRewriter &rewriter) const final {
+    llvm::TimeTraceScope scope("Call Conv Lowering Pass", op.getSymName().str());
+
     const auto module = op->getParentOfType<mlir::ModuleOp>();
 
     if (!op.getAst())

clang/lib/CIR/Dialect/Transforms/GotoSolver.cpp

@@ -7,6 +7,8 @@
 #include "clang/CIR/Dialect/IR/CIRDialect.h"
 #include "clang/CIR/Dialect/Passes.h"
 
+#include "llvm/Support/TimeProfiler.h"
+
 using namespace mlir;
 using namespace mlir::cir;
 
@@ -43,6 +45,7 @@ static void process(mlir::cir::FuncOp func) {
 }
 
 void GotoSolverPass::runOnOperation() {
+  llvm::TimeTraceScope scope("Goto Solver");
   SmallVector<Operation *, 16> ops;
   getOperation()->walk([&](mlir::cir::FuncOp op) { process(op); });
 }

clang/lib/CIR/FrontendAction/CIRGenAction.cpp

@@ -169,6 +169,8 @@ class CIRGenConsumer : public clang::ASTConsumer {
   }
 
   void HandleTranslationUnit(ASTContext &C) override {
+    llvm::TimeTraceScope scope("CIR Gen");
+
     // Note that this method is called after `HandleTopLevelDecl` has already
     // ran all over the top level decls. Here clang mostly wraps defered and
     // global codegen, followed by running CIR passes.

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

@@ -64,6 +64,7 @@
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TimeProfiler.h"
 #include <cstdint>
 #include <deque>
 #include <optional>
@@ -4322,6 +4323,8 @@ void ConvertCIRToLLVMPass::buildGlobalAnnotationsVar() {
 }
 
 void ConvertCIRToLLVMPass::runOnOperation() {
+  llvm::TimeTraceScope scope("Convert CIR to LLVM Pass");
+
   auto module = getOperation();
   mlir::DataLayout dataLayout(module);
   mlir::LLVMTypeConverter converter(&getContext());
@@ -4404,6 +4407,8 @@ extern void registerCIRDialectTranslation(mlir::MLIRContext &context);
 std::unique_ptr<llvm::Module>
 lowerDirectlyFromCIRToLLVMIR(mlir::ModuleOp theModule, LLVMContext &llvmCtx,
                              bool disableVerifier) {
+  llvm::TimeTraceScope scope("lower from CIR to LLVM directly");
+
   mlir::MLIRContext *mlirCtx = theModule.getContext();
   mlir::PassManager pm(mlirCtx);
   populateCIRToLLVMPasses(pm);
@@ -4435,6 +4440,8 @@ lowerDirectlyFromCIRToLLVMIR(mlir::ModuleOp theModule, LLVMContext &llvmCtx,
   mlir::registerOpenMPDialectTranslation(*mlirCtx);
   registerCIRDialectTranslation(*mlirCtx);
 
+  llvm::TimeTraceScope __scope("translateModuleToLLVMIR");
+
   auto ModuleName = theModule.getName();
   auto llvmModule = mlir::translateModuleToLLVMIR(
       theModule, llvmCtx, ModuleName ? *ModuleName : "CIRToLLVMModule");

clang/lib/CIR/Lowering/ThroughMLIR/LowerCIRToMLIR.cpp

@@ -53,6 +53,7 @@
 #include "llvm/ADT/Sequence.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/TypeSwitch.h"
+#include "llvm/Support/TimeProfiler.h"
 
 using namespace cir;
 using namespace llvm;
@@ -1419,6 +1420,8 @@ std::unique_ptr<llvm::Module>
 lowerFromCIRToMLIRToLLVMIR(mlir::ModuleOp theModule,
                            std::unique_ptr<mlir::MLIRContext> mlirCtx,
                            LLVMContext &llvmCtx) {
+  llvm::TimeTraceScope scope("Lower from CIR to MLIR To LLVM");
+
   mlir::PassManager pm(mlirCtx.get());
 
   pm.addPass(createConvertCIRToMLIRPass());
@@ -1451,6 +1454,8 @@ std::unique_ptr<mlir::Pass> createConvertCIRToMLIRPass() {
 
 mlir::ModuleOp lowerFromCIRToMLIR(mlir::ModuleOp theModule,
                                   mlir::MLIRContext *mlirCtx) {
+  llvm::TimeTraceScope scope("Lower CIR To MLIR");
+
   mlir::PassManager pm(mlirCtx);
 
   pm.addPass(createConvertCIRToMLIRPass());