[flang][Driver] Preliminary support for -ftime-report #107270
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries.
tarunprabhu
requested review from
asb,
banach-space,
tblah,
Leporacanthicus and
jeanPerier
llvmbot
added
clang
|
@llvm/pr-subscribers-clang-driver @llvm/pr-subscribers-flang-driver Author: Tarun Prabhu (tarunprabhu) ChangesThe behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries. Comments for reviewers: Support for detailed information about the LLVM passes has been commented out because it results in a duplicate registration of the FWIW, That being said, the lines that would enable this support are the ones below ( This essentially sets some global variables in LLVM (which are also set when the Patch is 28.89 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/107270.diff 15 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 1b9b3f2c6600a3..6a0a2b40cd192e 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -4012,7 +4012,7 @@ defm threadsafe_statics : BoolFOption<"threadsafe-statics",
"Do not emit code to make initialization of local statics thread safe">,
PosFlag<SetTrue>>;
def ftime_report : Flag<["-"], "ftime-report">, Group<f_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
MarshallingInfoFlag<CodeGenOpts<"TimePasses">>;
def ftime_report_EQ: Joined<["-"], "ftime-report=">, Group<f_Group>,
Visibility<[ClangOption, CC1Option]>, Values<"per-pass,per-pass-run">,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 6ce79d27e98c48..c1e724d7761dc0 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -150,7 +150,9 @@ void Flang::addCodegenOptions(const ArgList &Args,
options::OPT_flang_deprecated_no_hlfir,
options::OPT_flang_experimental_integer_overflow,
options::OPT_fno_ppc_native_vec_elem_order,
- options::OPT_fppc_native_vec_elem_order});
+ options::OPT_fppc_native_vec_elem_order,
+ options::OPT_ftime_report,
+ options::OPT_ftime_report_EQ});
}
void Flang::addPicOptions(const ArgList &Args, ArgStringList &CmdArgs) const {
diff --git a/flang/include/flang/Frontend/CompilerInstance.h b/flang/include/flang/Frontend/CompilerInstance.h
index 4fcc59f7cf577b..2844900063a9e5 100644
--- a/flang/include/flang/Frontend/CompilerInstance.h
+++ b/flang/include/flang/Frontend/CompilerInstance.h
@@ -20,6 +20,7 @@
#include "flang/Parser/provenance.h"
#include "flang/Semantics/runtime-type-info.h"
#include "flang/Semantics/semantics.h"
+#include "flang/Support/StringOstream.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
@@ -85,6 +86,27 @@ class CompilerInstance {
/// facilitate this. It is optional and will normally be just a nullptr.
std::unique_ptr<llvm::raw_pwrite_stream> outputStream;
+ /// @name Timing
+ /// Objects needed when timing is enabled.
+ /// @{
+ /// The timing manager.
+ mlir::DefaultTimingManager timingMgr;
+
+ /// The root of the timingScope. This will be reset in @ref executeAction if
+ /// timers have been enabled.
+ mlir::TimingScope timingScopeRoot;
+
+ /// @name Timing stream
+ /// The output streams to capture the timing. Three different streams are
+ /// needed because the timing classes all work slightly differently. We create
+ /// these streams so we have control over when and how the timing is
+ /// displayed. Otherwise, the timing is only displayed when the corresponding
+ /// managers/timers go out of scope.
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamMLIR;
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamLLVM;
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamCodeGen;
+ /// @}
+
public:
explicit CompilerInstance();
@@ -254,6 +276,43 @@ class CompilerInstance {
/// Produces the string which represents target feature
std::string getTargetFeatures();
+ /// {
+ /// @name Timing
+ /// @{
+ bool isTimingEnabled() { return timingMgr.isEnabled(); }
+ bool isTimingEnabled() const { return timingMgr.isEnabled(); }
+
+ mlir::DefaultTimingManager &getTimingManager() { return timingMgr; }
+ const mlir::DefaultTimingManager &getTimingManager() const {
+ return timingMgr;
+ }
+
+ mlir::TimingScope &getTimingScopeRoot() { return timingScopeRoot; }
+ const mlir::TimingScope &getTimingScopeRoot() const {
+ return timingScopeRoot;
+ }
+
+ /// Get the timing stream for the MLIR pass manager.
+ llvm::raw_ostream &getTimingStreamMLIR() {
+ assert(timingStreamMLIR && "Timing stream for MLIR was not set");
+ return *timingStreamMLIR;
+ }
+
+ /// Get the timing stream for the new LLVM pass manager.
+ llvm::raw_ostream &getTimingStreamLLVM() {
+ assert(timingStreamLLVM && "Timing stream for LLVM was not set");
+ return *timingStreamLLVM;
+ }
+
+ /// Get the timing stream fro the legacy LLVM pass manager.
+ /// NOTE: If the codegen is updated to use the new pass manager, this should
+ /// no longer be needed.
+ llvm::raw_ostream &getTimingStreamCodeGen() {
+ assert(timingStreamCodeGen && "Timing stream for codegen was not set");
+ return *timingStreamCodeGen;
+ }
+ /// @}
+
private:
/// Create a new output file
///
diff --git a/flang/include/flang/Frontend/CompilerInvocation.h b/flang/include/flang/Frontend/CompilerInvocation.h
index d1646f585cf850..4f848989aa1c80 100644
--- a/flang/include/flang/Frontend/CompilerInvocation.h
+++ b/flang/include/flang/Frontend/CompilerInvocation.h
@@ -21,6 +21,7 @@
#include "flang/Lower/LoweringOptions.h"
#include "flang/Parser/parsing.h"
#include "flang/Semantics/semantics.h"
+#include "mlir/Support/Timing.h"
#include "clang/Basic/Diagnostic.h"
#include "clang/Basic/DiagnosticOptions.h"
#include "llvm/Option/ArgList.h"
@@ -143,6 +144,14 @@ class CompilerInvocation : public CompilerInvocationBase {
},
};
+ /// Whether to time the invocation. Set when -ftime-report or -ftime-report=
+ /// is enabled.
+ bool enableTimers;
+
+ /// Whether to report the timing of each run of an LLVM pass. Set when
+ /// -ftime-report=per-pass-run is enabled.
+ bool timeLLVMPassesPerRun;
+
public:
CompilerInvocation() = default;
@@ -222,6 +231,12 @@ class CompilerInvocation : public CompilerInvocationBase {
return defaultKinds;
}
+ bool getEnableTimers() { return enableTimers; }
+ bool getEnableTimers() const { return enableTimers; }
+
+ bool getTimeLLVMPassesPerRun() { return timeLLVMPassesPerRun; }
+ bool getTimeLLVMPassesPerRun() const { return timeLLVMPassesPerRun; }
+
/// Create a compiler invocation from a list of input options.
/// \returns true on success.
/// \returns false if an error was encountered while parsing the arguments
diff --git a/flang/include/flang/Support/StringOstream.h b/flang/include/flang/Support/StringOstream.h
new file mode 100644
index 00000000000000..2e5c87eae058c6
--- /dev/null
+++ b/flang/include/flang/Support/StringOstream.h
@@ -0,0 +1,32 @@
+//===-- CompilerInstance.h - Flang Compiler Instance ------------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Coding style: https://mlir.llvm.org/getting_started/DeveloperGuide/
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef FORTRAN_SUPPORT_STRINGOSTREAM_H
+#define FORTRAN_SUPPORT_STRINGOSTREAM_H
+
+#include <llvm/Support/raw_ostream.h>
+
+namespace Fortran::support {
+
+/// Helper class to maintain both the an llvm::raw_string_ostream object and
+/// its associated buffer.
+class string_ostream : public llvm::raw_string_ostream {
+private:
+ std::string buf;
+
+public:
+ string_ostream() : llvm::raw_string_ostream(buf) {}
+};
+
+} // namespace Fortran::support
+
+#endif // FORTRAN_SUPPORT_STRINGOSTREAM_H
diff --git a/flang/include/flang/Support/Timing.h b/flang/include/flang/Support/Timing.h
new file mode 100644
index 00000000000000..75ba2a8d85f39f
--- /dev/null
+++ b/flang/include/flang/Support/Timing.h
@@ -0,0 +1,27 @@
+//===- Timing.h - Execution time measurement facilities ---------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Facilities to measure and provide statistics on execution time.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef FORTRAN_SUPPORT_TIMING_H
+#define FORTRAN_SUPPORT_TIMING_H
+
+#include "mlir/Support/Timing.h"
+
+namespace Fortran::support {
+
+/// Create a strategy to render the captured times in plain text. This is
+/// intended to be passed to a TimingManager.
+std::unique_ptr<mlir::OutputStrategy> createTimingFormatterText(
+ llvm::raw_ostream &os);
+
+} // namespace Fortran::support
+
+#endif // FORTRAN_SUPPORT_TIMING_H
diff --git a/flang/lib/CMakeLists.txt b/flang/lib/CMakeLists.txt
index f41d4df1f07e3c..2182e845b6a793 100644
--- a/flang/lib/CMakeLists.txt
+++ b/flang/lib/CMakeLists.txt
@@ -4,6 +4,7 @@ add_subdirectory(Decimal)
add_subdirectory(Lower)
add_subdirectory(Parser)
add_subdirectory(Semantics)
+add_subdirectory(Support)
add_subdirectory(Frontend)
add_subdirectory(FrontendTool)
diff --git a/flang/lib/Frontend/CMakeLists.txt b/flang/lib/Frontend/CMakeLists.txt
index ecdcc73d61ec1f..f679f23b00d2c8 100644
--- a/flang/lib/Frontend/CMakeLists.txt
+++ b/flang/lib/Frontend/CMakeLists.txt
@@ -30,6 +30,7 @@ add_flang_library(flangFrontend
FortranEvaluate
FortranCommon
FortranLower
+ FortranSupport
FIRDialect
FIRDialectSupport
FIRSupport
diff --git a/flang/lib/Frontend/CompilerInstance.cpp b/flang/lib/Frontend/CompilerInstance.cpp
index d37430e0e5773e..7cc9cf63aeee1f 100644
--- a/flang/lib/Frontend/CompilerInstance.cpp
+++ b/flang/lib/Frontend/CompilerInstance.cpp
@@ -17,9 +17,12 @@
#include "flang/Parser/parsing.h"
#include "flang/Parser/provenance.h"
#include "flang/Semantics/semantics.h"
+#include "flang/Support/Timing.h"
+#include "mlir/Support/RawOstreamExtras.h"
#include "clang/Basic/DiagnosticFrontend.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/MC/TargetRegistry.h"
+#include "llvm/Pass.h"
#include "llvm/Support/Errc.h"
#include "llvm/Support/Error.h"
#include "llvm/Support/FileSystem.h"
@@ -147,7 +150,7 @@ void CompilerInstance::clearOutputFiles(bool eraseFiles) {
}
bool CompilerInstance::executeAction(FrontendAction &act) {
- auto &invoc = this->getInvocation();
+ CompilerInvocation &invoc = this->getInvocation();
llvm::Triple targetTriple{llvm::Triple(invoc.getTargetOpts().triple)};
if (targetTriple.getArch() == llvm::Triple::ArchType::x86_64) {
@@ -167,6 +170,32 @@ bool CompilerInstance::executeAction(FrontendAction &act) {
// Set options controlling lowering to FIR.
invoc.setLoweringOptions();
+ if (invoc.getEnableTimers()) {
+ // FIXME: Currently, enabling these results in a duplicate registration
+ // error of the "sort-timers" command line option. It is not clear why that
+ // is occurring. Without setting these, we cannot get detailed information
+ // about the runtime of the LLVM IR optimization and code generation passes.
+ // Once the root cause of this is determined, we should enable this to have
+ // behavior that is comparable to clang.
+ // llvm::TimePassesIsEnabled = true;
+ // llvm::TimePassesPerRun = invoc.getTimeLLVMPassesPerRun();
+
+ timingStreamMLIR = std::make_unique<Fortran::support::string_ostream>();
+ timingStreamLLVM = std::make_unique<Fortran::support::string_ostream>();
+ timingStreamCodeGen = std::make_unique<Fortran::support::string_ostream>();
+
+ timingMgr.setEnabled(true);
+ timingMgr.setDisplayMode(mlir::DefaultTimingManager::DisplayMode::Tree);
+ timingMgr.setOutput(
+ Fortran::support::createTimingFormatterText(*timingStreamMLIR));
+
+ // Creating a new TimingScope will automatically start the timer. Since this
+ // is the top-level timer, this is ok because it will end up capturing the
+ // time for all the bookkeeping and other tasks that take place between
+ // parsing, lowering etc. for which finer-grained timers will be created.
+ timingScopeRoot = timingMgr.getRootScope();
+ }
+
// Run the frontend action `act` for every input file.
for (const FrontendInputFile &fif : getFrontendOpts().inputs) {
if (act.beginSourceFile(*this, fif)) {
@@ -176,6 +205,26 @@ bool CompilerInstance::executeAction(FrontendAction &act) {
act.endSourceFile();
}
}
+
+ if (timingMgr.isEnabled()) {
+ timingScopeRoot.stop();
+
+ // Write the timings to the associated output stream and clear all timers.
+ // We need to provide another stream because the TimingManager will attempt
+ // to print in its destructor even if it has been cleared. By the time that
+ // destructor runs, the output streams will have been destroyed, so give it
+ // a null stream.
+ timingMgr.print();
+ timingMgr.setOutput(
+ Fortran::support::createTimingFormatterText(mlir::thread_safe_nulls()));
+
+ // This is deliberately done in "reverse" order and does not match the
+ // behavior of clang.
+ llvm::errs() << timingStreamCodeGen->str() << "\n";
+ llvm::errs() << timingStreamLLVM->str() << "\n";
+ llvm::errs() << timingStreamMLIR->str() << "\n";
+ }
+
return !getDiagnostics().getClient()->getNumErrors();
}
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index 1d73397d330178..8607f6316d3167 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -1358,6 +1358,24 @@ bool CompilerInvocation::createFromArgs(
}
}
+ // Process the timing-related options.
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_ftime_report,
+ clang::driver::options::OPT_ftime_report_EQ)) {
+ invoc.enableTimers = true;
+ if (a->getOption().getID() == clang::driver::options::OPT_ftime_report_EQ) {
+ llvm::StringRef val = a->getValue();
+ if (val == "per-pass") {
+ invoc.timeLLVMPassesPerRun = false;
+ } else if (val == "per-pass-run") {
+ invoc.timeLLVMPassesPerRun = true;
+ } else {
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ }
+ }
+ }
+
invoc.setArgv0(argv0);
return success;
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 5c86bd947ce73f..4b67c17f761686 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -81,6 +81,16 @@
using namespace Fortran::frontend;
+static constexpr llvm::StringLiteral timingIdParse = "Parse";
+static constexpr llvm::StringLiteral timingIdMLIRGen = "MLIR generation";
+static constexpr llvm::StringLiteral timingIdMLIRPasses =
+ "MLIR translation/optimization";
+static constexpr llvm::StringLiteral timingIdLLVMIRGen = "LLVM IR generation";
+static constexpr llvm::StringLiteral timingIdLLVMIRPasses =
+ "LLVM IR optimizations";
+static constexpr llvm::StringLiteral timingIdBackend =
+ "Assembly/Object code generation";
+
// Declare plugin extension function declarations.
#define HANDLE_EXTENSION(Ext) \
llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
@@ -224,6 +234,14 @@ static void addAMDGPUSpecificMLIRItems(mlir::ModuleOp &mlirModule,
bool CodeGenAction::beginSourceFileAction() {
llvmCtx = std::make_unique<llvm::LLVMContext>();
CompilerInstance &ci = this->getInstance();
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
+
+ // This will provide timing information even when the input is an LLVM IR or
+ // MLIR file. That is fine because those do have to be parsed, so the label
+ // is still accurate.
+ mlir::TimingScope timingScopeParse = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdParse, timingMgr));
// If the input is an LLVM file, just parse it and return.
if (this->getCurrentInput().getKind().getLanguage() == Language::LLVM_IR) {
@@ -285,6 +303,10 @@ bool CodeGenAction::beginSourceFileAction() {
if (!res)
return res;
+ timingScopeParse.stop();
+ mlir::TimingScope timingScopeMLIRGen = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRGen, timingMgr));
+
// Create a LoweringBridge
const common::IntrinsicTypeDefaultKinds &defKinds =
ci.getSemanticsContext().defaultKinds();
@@ -319,6 +341,7 @@ bool CodeGenAction::beginSourceFileAction() {
// constants etc.
addDependentLibs(*mlirModule, ci);
addAMDGPUSpecificMLIRItems(*mlirModule, ci);
+ timingScopeMLIRGen.stop();
// run the default passes.
mlir::PassManager pm((*mlirModule)->getName(),
@@ -340,6 +363,7 @@ bool CodeGenAction::beginSourceFileAction() {
pm.enableVerifier(/*verifyPasses=*/true);
pm.addPass(std::make_unique<Fortran::lower::VerifierPass>());
+ pm.enableTiming(timingScopeMLIRGen);
if (mlir::failed(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
@@ -348,6 +372,7 @@ bool CodeGenAction::beginSourceFileAction() {
ci.getDiagnostics().Report(diagID);
return false;
}
+ timingScopeMLIRGen.stop();
// Print initial full MLIR module, before lowering or transformations, if
// -save-temps has been specified.
@@ -698,8 +723,10 @@ void CodeGenAction::lowerHLFIRToFIR() {
assert(mlirModule && "The MLIR module has not been generated yet.");
CompilerInstance &ci = this->getInstance();
- auto opts = ci.getInvocation().getCodeGenOpts();
+ const CodeGenOptions &opts = ci.getInvocation().getCodeGenOpts();
llvm::OptimizationLevel level = mapToLevel(opts);
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
fir::support::loadDialects(*mlirCtx);
@@ -714,6 +741,9 @@ void CodeGenAction::lowerHLFIRToFIR() {
fir::createHLFIRToFIRPassPipeline(pm, level);
(void)mlir::applyPassManagerCLOptions(pm);
+ mlir::TimingScope timingScopeMLIRPasses = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRPasses, timingMgr));
+ pm.enableTiming(timingScopeMLIRPasses);
if (!mlir::succeeded(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
clang::DiagnosticsEngine::Error, "Lowering to FIR failed");
@@ -798,9 +828,12 @@ void CodeGenAction::generateLLVMIR() {
assert(mlirModule && "The MLIR module has not been generated yet.");
CompilerInstance &ci = this->getInstance();
- auto opts = ci.getInvocation().getCodeGenOpts();
- auto mathOpts = ci.getInvocation().getLoweringOpts().getMathOptions();
+ CompilerInvocation &invoc = ci.getInvocation();
+ const CodeGenOptions &opts = invoc.getCodeGenOpts();
+ const auto &mathOpts = invoc.getLoweringOpts().getMathOptions();
llvm::OptimizationLevel level = mapToLevel(opts);
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
fir::support::loadDialects(*mlirCtx);
mlir::DialectRegistry registry;
@@ -832,11 +865,15 @@ void CodeGenAction::generateLLVMIR() {
(void)mlir::applyPassManagerCLOptions(pm);
// run the pass manager
+ mlir::TimingScope timingScopeMLIRPasses = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRPasses, timingMgr));
+ pm.enableTiming(timingScopeMLIRPasses);
if (!mlir::succeeded(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
clang::DiagnosticsEngine::Error, "Lowering to LLVM IR failed");
ci.getDiagnostics().Report(diagID);
}
+ timingScopeMLIRPasses.stop();
// Print final MLIR module, just before translation into LLVM IR, if
// -save-temps has been specified.
@@ -849,6 +886,8 @@ void CodeGenAction::generateLLVMIR() {
}
// Translate to LLVM IR
+ mlir::TimingScope timingScopeLLVMIRGen = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdLLVMIRGen, timingMgr));
std::optional<llvm::StringRef> moduleName = mlirModule->getName();
llvmModule = mlir::translateModuleToLLVMIR(
*mlirModule, *llvmCtx, moduleName ? *moduleName : "FIRModule");
@@ -952,11 +991,12 @@ static void generateMachineCodeOrAssemblyImpl(clang::DiagnosticsEngine &diags,
}
void CodeGenAction::runOptimizati...
[truncated]
|
|
@llvm/pr-subscribers-clang Author: Tarun Prabhu (tarunprabhu) ChangesThe behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries. Comments for reviewers: Support for detailed information about the LLVM passes has been commented out because it results in a duplicate registration of the FWIW, That being said, the lines that would enable this support are the ones below ( This essentially sets some global variables in LLVM (which are also set when the Patch is 28.89 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/107270.diff 15 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 1b9b3f2c6600a3..6a0a2b40cd192e 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -4012,7 +4012,7 @@ defm threadsafe_statics : BoolFOption<"threadsafe-statics",
"Do not emit code to make initialization of local statics thread safe">,
PosFlag<SetTrue>>;
def ftime_report : Flag<["-"], "ftime-report">, Group<f_Group>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
MarshallingInfoFlag<CodeGenOpts<"TimePasses">>;
def ftime_report_EQ: Joined<["-"], "ftime-report=">, Group<f_Group>,
Visibility<[ClangOption, CC1Option]>, Values<"per-pass,per-pass-run">,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 6ce79d27e98c48..c1e724d7761dc0 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -150,7 +150,9 @@ void Flang::addCodegenOptions(const ArgList &Args,
options::OPT_flang_deprecated_no_hlfir,
options::OPT_flang_experimental_integer_overflow,
options::OPT_fno_ppc_native_vec_elem_order,
- options::OPT_fppc_native_vec_elem_order});
+ options::OPT_fppc_native_vec_elem_order,
+ options::OPT_ftime_report,
+ options::OPT_ftime_report_EQ});
}
void Flang::addPicOptions(const ArgList &Args, ArgStringList &CmdArgs) const {
diff --git a/flang/include/flang/Frontend/CompilerInstance.h b/flang/include/flang/Frontend/CompilerInstance.h
index 4fcc59f7cf577b..2844900063a9e5 100644
--- a/flang/include/flang/Frontend/CompilerInstance.h
+++ b/flang/include/flang/Frontend/CompilerInstance.h
@@ -20,6 +20,7 @@
#include "flang/Parser/provenance.h"
#include "flang/Semantics/runtime-type-info.h"
#include "flang/Semantics/semantics.h"
+#include "flang/Support/StringOstream.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
@@ -85,6 +86,27 @@ class CompilerInstance {
/// facilitate this. It is optional and will normally be just a nullptr.
std::unique_ptr<llvm::raw_pwrite_stream> outputStream;
+ /// @name Timing
+ /// Objects needed when timing is enabled.
+ /// @{
+ /// The timing manager.
+ mlir::DefaultTimingManager timingMgr;
+
+ /// The root of the timingScope. This will be reset in @ref executeAction if
+ /// timers have been enabled.
+ mlir::TimingScope timingScopeRoot;
+
+ /// @name Timing stream
+ /// The output streams to capture the timing. Three different streams are
+ /// needed because the timing classes all work slightly differently. We create
+ /// these streams so we have control over when and how the timing is
+ /// displayed. Otherwise, the timing is only displayed when the corresponding
+ /// managers/timers go out of scope.
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamMLIR;
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamLLVM;
+ std::unique_ptr<Fortran::support::string_ostream> timingStreamCodeGen;
+ /// @}
+
public:
explicit CompilerInstance();
@@ -254,6 +276,43 @@ class CompilerInstance {
/// Produces the string which represents target feature
std::string getTargetFeatures();
+ /// {
+ /// @name Timing
+ /// @{
+ bool isTimingEnabled() { return timingMgr.isEnabled(); }
+ bool isTimingEnabled() const { return timingMgr.isEnabled(); }
+
+ mlir::DefaultTimingManager &getTimingManager() { return timingMgr; }
+ const mlir::DefaultTimingManager &getTimingManager() const {
+ return timingMgr;
+ }
+
+ mlir::TimingScope &getTimingScopeRoot() { return timingScopeRoot; }
+ const mlir::TimingScope &getTimingScopeRoot() const {
+ return timingScopeRoot;
+ }
+
+ /// Get the timing stream for the MLIR pass manager.
+ llvm::raw_ostream &getTimingStreamMLIR() {
+ assert(timingStreamMLIR && "Timing stream for MLIR was not set");
+ return *timingStreamMLIR;
+ }
+
+ /// Get the timing stream for the new LLVM pass manager.
+ llvm::raw_ostream &getTimingStreamLLVM() {
+ assert(timingStreamLLVM && "Timing stream for LLVM was not set");
+ return *timingStreamLLVM;
+ }
+
+ /// Get the timing stream fro the legacy LLVM pass manager.
+ /// NOTE: If the codegen is updated to use the new pass manager, this should
+ /// no longer be needed.
+ llvm::raw_ostream &getTimingStreamCodeGen() {
+ assert(timingStreamCodeGen && "Timing stream for codegen was not set");
+ return *timingStreamCodeGen;
+ }
+ /// @}
+
private:
/// Create a new output file
///
diff --git a/flang/include/flang/Frontend/CompilerInvocation.h b/flang/include/flang/Frontend/CompilerInvocation.h
index d1646f585cf850..4f848989aa1c80 100644
--- a/flang/include/flang/Frontend/CompilerInvocation.h
+++ b/flang/include/flang/Frontend/CompilerInvocation.h
@@ -21,6 +21,7 @@
#include "flang/Lower/LoweringOptions.h"
#include "flang/Parser/parsing.h"
#include "flang/Semantics/semantics.h"
+#include "mlir/Support/Timing.h"
#include "clang/Basic/Diagnostic.h"
#include "clang/Basic/DiagnosticOptions.h"
#include "llvm/Option/ArgList.h"
@@ -143,6 +144,14 @@ class CompilerInvocation : public CompilerInvocationBase {
},
};
+ /// Whether to time the invocation. Set when -ftime-report or -ftime-report=
+ /// is enabled.
+ bool enableTimers;
+
+ /// Whether to report the timing of each run of an LLVM pass. Set when
+ /// -ftime-report=per-pass-run is enabled.
+ bool timeLLVMPassesPerRun;
+
public:
CompilerInvocation() = default;
@@ -222,6 +231,12 @@ class CompilerInvocation : public CompilerInvocationBase {
return defaultKinds;
}
+ bool getEnableTimers() { return enableTimers; }
+ bool getEnableTimers() const { return enableTimers; }
+
+ bool getTimeLLVMPassesPerRun() { return timeLLVMPassesPerRun; }
+ bool getTimeLLVMPassesPerRun() const { return timeLLVMPassesPerRun; }
+
/// Create a compiler invocation from a list of input options.
/// \returns true on success.
/// \returns false if an error was encountered while parsing the arguments
diff --git a/flang/include/flang/Support/StringOstream.h b/flang/include/flang/Support/StringOstream.h
new file mode 100644
index 00000000000000..2e5c87eae058c6
--- /dev/null
+++ b/flang/include/flang/Support/StringOstream.h
@@ -0,0 +1,32 @@
+//===-- CompilerInstance.h - Flang Compiler Instance ------------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Coding style: https://mlir.llvm.org/getting_started/DeveloperGuide/
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef FORTRAN_SUPPORT_STRINGOSTREAM_H
+#define FORTRAN_SUPPORT_STRINGOSTREAM_H
+
+#include <llvm/Support/raw_ostream.h>
+
+namespace Fortran::support {
+
+/// Helper class to maintain both the an llvm::raw_string_ostream object and
+/// its associated buffer.
+class string_ostream : public llvm::raw_string_ostream {
+private:
+ std::string buf;
+
+public:
+ string_ostream() : llvm::raw_string_ostream(buf) {}
+};
+
+} // namespace Fortran::support
+
+#endif // FORTRAN_SUPPORT_STRINGOSTREAM_H
diff --git a/flang/include/flang/Support/Timing.h b/flang/include/flang/Support/Timing.h
new file mode 100644
index 00000000000000..75ba2a8d85f39f
--- /dev/null
+++ b/flang/include/flang/Support/Timing.h
@@ -0,0 +1,27 @@
+//===- Timing.h - Execution time measurement facilities ---------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Facilities to measure and provide statistics on execution time.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef FORTRAN_SUPPORT_TIMING_H
+#define FORTRAN_SUPPORT_TIMING_H
+
+#include "mlir/Support/Timing.h"
+
+namespace Fortran::support {
+
+/// Create a strategy to render the captured times in plain text. This is
+/// intended to be passed to a TimingManager.
+std::unique_ptr<mlir::OutputStrategy> createTimingFormatterText(
+ llvm::raw_ostream &os);
+
+} // namespace Fortran::support
+
+#endif // FORTRAN_SUPPORT_TIMING_H
diff --git a/flang/lib/CMakeLists.txt b/flang/lib/CMakeLists.txt
index f41d4df1f07e3c..2182e845b6a793 100644
--- a/flang/lib/CMakeLists.txt
+++ b/flang/lib/CMakeLists.txt
@@ -4,6 +4,7 @@ add_subdirectory(Decimal)
add_subdirectory(Lower)
add_subdirectory(Parser)
add_subdirectory(Semantics)
+add_subdirectory(Support)
add_subdirectory(Frontend)
add_subdirectory(FrontendTool)
diff --git a/flang/lib/Frontend/CMakeLists.txt b/flang/lib/Frontend/CMakeLists.txt
index ecdcc73d61ec1f..f679f23b00d2c8 100644
--- a/flang/lib/Frontend/CMakeLists.txt
+++ b/flang/lib/Frontend/CMakeLists.txt
@@ -30,6 +30,7 @@ add_flang_library(flangFrontend
FortranEvaluate
FortranCommon
FortranLower
+ FortranSupport
FIRDialect
FIRDialectSupport
FIRSupport
diff --git a/flang/lib/Frontend/CompilerInstance.cpp b/flang/lib/Frontend/CompilerInstance.cpp
index d37430e0e5773e..7cc9cf63aeee1f 100644
--- a/flang/lib/Frontend/CompilerInstance.cpp
+++ b/flang/lib/Frontend/CompilerInstance.cpp
@@ -17,9 +17,12 @@
#include "flang/Parser/parsing.h"
#include "flang/Parser/provenance.h"
#include "flang/Semantics/semantics.h"
+#include "flang/Support/Timing.h"
+#include "mlir/Support/RawOstreamExtras.h"
#include "clang/Basic/DiagnosticFrontend.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/MC/TargetRegistry.h"
+#include "llvm/Pass.h"
#include "llvm/Support/Errc.h"
#include "llvm/Support/Error.h"
#include "llvm/Support/FileSystem.h"
@@ -147,7 +150,7 @@ void CompilerInstance::clearOutputFiles(bool eraseFiles) {
}
bool CompilerInstance::executeAction(FrontendAction &act) {
- auto &invoc = this->getInvocation();
+ CompilerInvocation &invoc = this->getInvocation();
llvm::Triple targetTriple{llvm::Triple(invoc.getTargetOpts().triple)};
if (targetTriple.getArch() == llvm::Triple::ArchType::x86_64) {
@@ -167,6 +170,32 @@ bool CompilerInstance::executeAction(FrontendAction &act) {
// Set options controlling lowering to FIR.
invoc.setLoweringOptions();
+ if (invoc.getEnableTimers()) {
+ // FIXME: Currently, enabling these results in a duplicate registration
+ // error of the "sort-timers" command line option. It is not clear why that
+ // is occurring. Without setting these, we cannot get detailed information
+ // about the runtime of the LLVM IR optimization and code generation passes.
+ // Once the root cause of this is determined, we should enable this to have
+ // behavior that is comparable to clang.
+ // llvm::TimePassesIsEnabled = true;
+ // llvm::TimePassesPerRun = invoc.getTimeLLVMPassesPerRun();
+
+ timingStreamMLIR = std::make_unique<Fortran::support::string_ostream>();
+ timingStreamLLVM = std::make_unique<Fortran::support::string_ostream>();
+ timingStreamCodeGen = std::make_unique<Fortran::support::string_ostream>();
+
+ timingMgr.setEnabled(true);
+ timingMgr.setDisplayMode(mlir::DefaultTimingManager::DisplayMode::Tree);
+ timingMgr.setOutput(
+ Fortran::support::createTimingFormatterText(*timingStreamMLIR));
+
+ // Creating a new TimingScope will automatically start the timer. Since this
+ // is the top-level timer, this is ok because it will end up capturing the
+ // time for all the bookkeeping and other tasks that take place between
+ // parsing, lowering etc. for which finer-grained timers will be created.
+ timingScopeRoot = timingMgr.getRootScope();
+ }
+
// Run the frontend action `act` for every input file.
for (const FrontendInputFile &fif : getFrontendOpts().inputs) {
if (act.beginSourceFile(*this, fif)) {
@@ -176,6 +205,26 @@ bool CompilerInstance::executeAction(FrontendAction &act) {
act.endSourceFile();
}
}
+
+ if (timingMgr.isEnabled()) {
+ timingScopeRoot.stop();
+
+ // Write the timings to the associated output stream and clear all timers.
+ // We need to provide another stream because the TimingManager will attempt
+ // to print in its destructor even if it has been cleared. By the time that
+ // destructor runs, the output streams will have been destroyed, so give it
+ // a null stream.
+ timingMgr.print();
+ timingMgr.setOutput(
+ Fortran::support::createTimingFormatterText(mlir::thread_safe_nulls()));
+
+ // This is deliberately done in "reverse" order and does not match the
+ // behavior of clang.
+ llvm::errs() << timingStreamCodeGen->str() << "\n";
+ llvm::errs() << timingStreamLLVM->str() << "\n";
+ llvm::errs() << timingStreamMLIR->str() << "\n";
+ }
+
return !getDiagnostics().getClient()->getNumErrors();
}
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index 1d73397d330178..8607f6316d3167 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -1358,6 +1358,24 @@ bool CompilerInvocation::createFromArgs(
}
}
+ // Process the timing-related options.
+ if (const llvm::opt::Arg *a =
+ args.getLastArg(clang::driver::options::OPT_ftime_report,
+ clang::driver::options::OPT_ftime_report_EQ)) {
+ invoc.enableTimers = true;
+ if (a->getOption().getID() == clang::driver::options::OPT_ftime_report_EQ) {
+ llvm::StringRef val = a->getValue();
+ if (val == "per-pass") {
+ invoc.timeLLVMPassesPerRun = false;
+ } else if (val == "per-pass-run") {
+ invoc.timeLLVMPassesPerRun = true;
+ } else {
+ diags.Report(clang::diag::err_drv_invalid_value)
+ << a->getAsString(args) << a->getValue();
+ }
+ }
+ }
+
invoc.setArgv0(argv0);
return success;
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 5c86bd947ce73f..4b67c17f761686 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -81,6 +81,16 @@
using namespace Fortran::frontend;
+static constexpr llvm::StringLiteral timingIdParse = "Parse";
+static constexpr llvm::StringLiteral timingIdMLIRGen = "MLIR generation";
+static constexpr llvm::StringLiteral timingIdMLIRPasses =
+ "MLIR translation/optimization";
+static constexpr llvm::StringLiteral timingIdLLVMIRGen = "LLVM IR generation";
+static constexpr llvm::StringLiteral timingIdLLVMIRPasses =
+ "LLVM IR optimizations";
+static constexpr llvm::StringLiteral timingIdBackend =
+ "Assembly/Object code generation";
+
// Declare plugin extension function declarations.
#define HANDLE_EXTENSION(Ext) \
llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
@@ -224,6 +234,14 @@ static void addAMDGPUSpecificMLIRItems(mlir::ModuleOp &mlirModule,
bool CodeGenAction::beginSourceFileAction() {
llvmCtx = std::make_unique<llvm::LLVMContext>();
CompilerInstance &ci = this->getInstance();
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
+
+ // This will provide timing information even when the input is an LLVM IR or
+ // MLIR file. That is fine because those do have to be parsed, so the label
+ // is still accurate.
+ mlir::TimingScope timingScopeParse = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdParse, timingMgr));
// If the input is an LLVM file, just parse it and return.
if (this->getCurrentInput().getKind().getLanguage() == Language::LLVM_IR) {
@@ -285,6 +303,10 @@ bool CodeGenAction::beginSourceFileAction() {
if (!res)
return res;
+ timingScopeParse.stop();
+ mlir::TimingScope timingScopeMLIRGen = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRGen, timingMgr));
+
// Create a LoweringBridge
const common::IntrinsicTypeDefaultKinds &defKinds =
ci.getSemanticsContext().defaultKinds();
@@ -319,6 +341,7 @@ bool CodeGenAction::beginSourceFileAction() {
// constants etc.
addDependentLibs(*mlirModule, ci);
addAMDGPUSpecificMLIRItems(*mlirModule, ci);
+ timingScopeMLIRGen.stop();
// run the default passes.
mlir::PassManager pm((*mlirModule)->getName(),
@@ -340,6 +363,7 @@ bool CodeGenAction::beginSourceFileAction() {
pm.enableVerifier(/*verifyPasses=*/true);
pm.addPass(std::make_unique<Fortran::lower::VerifierPass>());
+ pm.enableTiming(timingScopeMLIRGen);
if (mlir::failed(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
@@ -348,6 +372,7 @@ bool CodeGenAction::beginSourceFileAction() {
ci.getDiagnostics().Report(diagID);
return false;
}
+ timingScopeMLIRGen.stop();
// Print initial full MLIR module, before lowering or transformations, if
// -save-temps has been specified.
@@ -698,8 +723,10 @@ void CodeGenAction::lowerHLFIRToFIR() {
assert(mlirModule && "The MLIR module has not been generated yet.");
CompilerInstance &ci = this->getInstance();
- auto opts = ci.getInvocation().getCodeGenOpts();
+ const CodeGenOptions &opts = ci.getInvocation().getCodeGenOpts();
llvm::OptimizationLevel level = mapToLevel(opts);
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
fir::support::loadDialects(*mlirCtx);
@@ -714,6 +741,9 @@ void CodeGenAction::lowerHLFIRToFIR() {
fir::createHLFIRToFIRPassPipeline(pm, level);
(void)mlir::applyPassManagerCLOptions(pm);
+ mlir::TimingScope timingScopeMLIRPasses = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRPasses, timingMgr));
+ pm.enableTiming(timingScopeMLIRPasses);
if (!mlir::succeeded(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
clang::DiagnosticsEngine::Error, "Lowering to FIR failed");
@@ -798,9 +828,12 @@ void CodeGenAction::generateLLVMIR() {
assert(mlirModule && "The MLIR module has not been generated yet.");
CompilerInstance &ci = this->getInstance();
- auto opts = ci.getInvocation().getCodeGenOpts();
- auto mathOpts = ci.getInvocation().getLoweringOpts().getMathOptions();
+ CompilerInvocation &invoc = ci.getInvocation();
+ const CodeGenOptions &opts = invoc.getCodeGenOpts();
+ const auto &mathOpts = invoc.getLoweringOpts().getMathOptions();
llvm::OptimizationLevel level = mapToLevel(opts);
+ mlir::DefaultTimingManager &timingMgr = ci.getTimingManager();
+ mlir::TimingScope &timingScopeRoot = ci.getTimingScopeRoot();
fir::support::loadDialects(*mlirCtx);
mlir::DialectRegistry registry;
@@ -832,11 +865,15 @@ void CodeGenAction::generateLLVMIR() {
(void)mlir::applyPassManagerCLOptions(pm);
// run the pass manager
+ mlir::TimingScope timingScopeMLIRPasses = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdMLIRPasses, timingMgr));
+ pm.enableTiming(timingScopeMLIRPasses);
if (!mlir::succeeded(pm.run(*mlirModule))) {
unsigned diagID = ci.getDiagnostics().getCustomDiagID(
clang::DiagnosticsEngine::Error, "Lowering to LLVM IR failed");
ci.getDiagnostics().Report(diagID);
}
+ timingScopeMLIRPasses.stop();
// Print final MLIR module, just before translation into LLVM IR, if
// -save-temps has been specified.
@@ -849,6 +886,8 @@ void CodeGenAction::generateLLVMIR() {
}
// Translate to LLVM IR
+ mlir::TimingScope timingScopeLLVMIRGen = timingScopeRoot.nest(
+ mlir::TimingIdentifier::get(timingIdLLVMIRGen, timingMgr));
std::optional<llvm::StringRef> moduleName = mlirModule->getName();
llvmModule = mlir::translateModuleToLLVMIR(
*mlirModule, *llvmCtx, moduleName ? *moduleName : "FIRModule");
@@ -952,11 +991,12 @@ static void generateMachineCodeOrAssemblyImpl(clang::DiagnosticsEngine &diags,
}
void CodeGenAction::runOptimizati...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
…ce that is not supported right now anyway. Add checks that will only print out timing streams if there is anything in them to avoid empty lines.
tblah
reviewed
Sep 5, 2024
macurtis-amd
added a commit
that referenced
this pull request
Jan 9, 2025
This is intended to help with flang `-ftime-report` support: - #107270. With this change, I was able to cherry-pick #107270, uncomment `llvm::TimePassesIsEnabled = true;` and compile with `-ftime-report`. I also noticed that `clang/lib/Driver/OffloadBundler.cpp` was statically constructing a `TimerGroup` and changed it to lazily construct via ManagedStatic.
macurtis-amd
reviewed
Jan 9, 2025
| // about the runtime of the LLVM IR optimization and code generation passes. | ||
| // Once the root cause of this is determined, we should enable this to have | ||
| // behavior that is comparable to clang. | ||
| // llvm::TimePassesIsEnabled = true; |
There was a problem hiding this comment.
@tarunprabhu #121663 has been merged which fixes the assert you were seeing. You should be able to uncomment this line (and remove preceding comment).
| @@ -1358,6 +1358,12 @@ bool CompilerInvocation::createFromArgs( | |||
| } | |||
| } | |||
|
|
|||
| // Process the timing-related options. | |||
| if (const llvm::opt::Arg *a = | |||
There was a problem hiding this comment.
a is unused and produces a warning in my build.
| @@ -222,6 +227,9 @@ class CompilerInvocation : public CompilerInvocationBase { | |||
| return defaultKinds; | |||
| } | |||
|
|
|||
| bool getEnableTimers() { return enableTimers; } | |||
There was a problem hiding this comment.
Nit: non-const getter is not needed.
MaskRay
reviewed
Jan 10, 2025
| @@ -81,6 +81,16 @@ | |||
|
|
|||
| using namespace Fortran::frontend; | |||
|
|
|||
| static constexpr llvm::StringLiteral timingIdParse = "Parse"; | |||
There was a problem hiding this comment.
constexpr at namespace level implies static
MaskRay
reviewed
Jan 10, 2025
| ! Check that -ftime-report flag is passed as-is to fc1. The value of the flag | ||
| ! is only checked there. This behavior intentionally mirrors that of clang. | ||
| ! | ||
| ! RUN: %flang -### -c -ftime-report %s 2>&1 | FileCheck %s --check-prefix=CHECK-DRIVER |
There was a problem hiding this comment.
do you need -mllvm -sort-timers=0 to prevent reordering of timers in the TimerGroup?
macurtis-amd
added a commit
that referenced
this pull request
Jan 14, 2025
The behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries. Based on @tarunprabhu's PR #107270 with minor changes addressing latest review feedback. He's busy and we'd like to get this support in ASAP. Co-authored-by: Tarun Prabhu <tarun.prabhu@gmail.com>
github-actions bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Jan 14, 2025
…2894) The behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries. Based on @tarunprabhu's PR llvm/llvm-project#107270 with minor changes addressing latest review feedback. He's busy and we'd like to get this support in ASAP. Co-authored-by: Tarun Prabhu <tarun.prabhu@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The behavior is not entirely consistent with that of clang for the moment since detailed timing information on the LLVM IR optimization and code generation passes is not provided. The -ftime-report= option is also not enabled since that is only relevant for information about the LLVM IR passes. However, some code to handle that option has been included, to make it easier to support the option when the issues blocking it are resolved. A FortranSupport library has been created that is intended to mirror the LLVM and MLIR support libraries.
Comments for reviewers:
Support for detailed information about the LLVM passes has been commented out because it results in a duplicate registration of the
-sort-timesLLVM command-line option. I have no idea why this is happening. The only way I was able to resolve it was by removing the callInitLLVM().check-flangstill passed after the call was removed, but I only tested on Linux on x86. If anyone has suggestions on a better way to resolve this, I am happy to take them on board.FWIW,
clangdoes not seem to callInitLLVM().That being said, the lines that would enable this support are the ones below (
clangdoes precisely this as well):This essentially sets some global variables in LLVM (which are also set when the
-ftime-passesLLVM command-line option is used). In a different PR, @banach-space noted that such approaches ought to be avoided and I tend to agree. One way to resolve the issue would be to provide some hooks in LLVM that will allow the timers to be initialized independently of the command line options. I can put up a draft PR for this is anyone is interested.