[flang][OpenMP] Add flag to enable/disable GlobalFilteringPass

clang/include/clang/Driver/Options.td

@@ -6826,6 +6826,7 @@ defm xor_operator : OptInFC1FFlag<"xor-operator", "Enable .XOR. as a synonym of
 defm logical_abbreviations : OptInFC1FFlag<"logical-abbreviations", "Enable logical abbreviations">;
 defm implicit_none : OptInFC1FFlag<"implicit-none", "No implicit typing allowed unless overridden by IMPLICIT statements">;
 defm underscoring : OptInFC1FFlag<"underscoring", "Appends one trailing underscore to external names">;
+defm offload_global_filtering : OptInFC1FFlag<"offload-global-filtering", "Enable/disable OpenMP global filtering pass">;
 defm ppc_native_vec_elem_order: BoolOptionWithoutMarshalling<"f", "ppc-native-vector-element-order",
   PosFlag<SetTrue, [], [ClangOption], "Specifies PowerPC native vector element order (default)">,
   NegFlag<SetFalse, [], [ClangOption], "Specifies PowerPC non-native vector element order">>;

clang/lib/Driver/ToolChains/Flang.cpp

@@ -120,7 +120,9 @@ void Flang::addOtherOptions(const ArgList &Args, ArgStringList &CmdArgs) const {
                    options::OPT_fintrinsic_modules_path, options::OPT_pedantic,
                    options::OPT_std_EQ, options::OPT_W_Joined,
                    options::OPT_fconvert_EQ, options::OPT_fpass_plugin_EQ,
-                   options::OPT_funderscoring, options::OPT_fno_underscoring});
+                   options::OPT_funderscoring, options::OPT_fno_underscoring,
+                   options::OPT_foffload_global_filtering,
+                   options::OPT_fno_offload_global_filtering});
 
   llvm::codegenoptions::DebugInfoKind DebugInfoKind;
   if (Args.hasArg(options::OPT_gN_Group)) {

flang/include/flang/Frontend/CodeGenOptions.def

@@ -35,6 +35,7 @@ CODEGENOPT(LoopVersioning, 1, 0) ///< Enable loop versioning.
 CODEGENOPT(AliasAnalysis, 1, 0) ///< Enable alias analysis pass
 
 CODEGENOPT(Underscoring, 1, 1)
+CODEGENOPT(OffloadGlobalFiltering, 1, 1)
 ENUM_CODEGENOPT(RelocationModel, llvm::Reloc::Model, 3, llvm::Reloc::PIC_) ///< Name of the relocation model to use.
 ENUM_CODEGENOPT(DebugInfo,  llvm::codegenoptions::DebugInfoKind, 4,  llvm::codegenoptions::NoDebugInfo) ///< Level of debug info to generate
 ENUM_CODEGENOPT(VecLib, llvm::driver::VectorLibrary, 3, llvm::driver::VectorLibrary::NoLibrary) ///< Vector functions library to use

flang/include/flang/Optimizer/Passes/Pipelines.h

@@ -128,6 +128,12 @@ void createHLFIRToFIRPassPipeline(
 using DoConcurrentMappingKind =
     Fortran::frontend::CodeGenOptions::DoConcurrentMappingKind;
 
+struct OpenMPFIRPassPipelineOpts {
+  bool isTargetDevice;
+  bool enableOffloadGlobalFiltering;
+  DoConcurrentMappingKind doConcurrentMappingKind;
+};
+
 /// Create a pass pipeline for handling certain OpenMP transformations needed
 /// prior to FIR lowering.
 ///
@@ -137,9 +143,8 @@ using DoConcurrentMappingKind =
 /// \param pm - MLIR pass manager that will hold the pipeline definition.
 /// \param isTargetDevice - Whether code is being generated for a target device
 /// rather than the host device.
-void createOpenMPFIRPassPipeline(
-    mlir::PassManager &pm, bool isTargetDevice,
-    DoConcurrentMappingKind doConcurrentMappingKind);
+void createOpenMPFIRPassPipeline(mlir::PassManager &pm,
+                                 OpenMPFIRPassPipelineOpts opts);
 
 #if !defined(FLANG_EXCLUDE_CODEGEN)
 void createDebugPasses(mlir::PassManager &pm,

flang/lib/Frontend/CompilerInvocation.cpp

@@ -443,6 +443,11 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
     opts.Underscoring = 0;
   }
 
+  if (args.hasFlag(clang::driver::options::OPT_fno_offload_global_filtering,
+                   clang::driver::options::OPT_foffload_global_filtering, false)) {
+    opts.OffloadGlobalFiltering = 0;
+  }
+
   parseDoConcurrentMapping(opts, args, diags);
 }
 

flang/lib/Frontend/FrontendActions.cpp

@@ -334,12 +334,16 @@ bool CodeGenAction::beginSourceFileAction() {
       ci.getInvocation().getFrontendOpts().features.IsEnabled(
           Fortran::common::LanguageFeature::OpenMP);
 
+  fir::OpenMPFIRPassPipelineOpts opts;
+
   using DoConcurrentMappingKind =
       Fortran::frontend::CodeGenOptions::DoConcurrentMappingKind;
-  DoConcurrentMappingKind doConcurrentMappingKind =
+  opts.doConcurrentMappingKind =
       ci.getInvocation().getCodeGenOpts().getDoConcurrentMapping();
+  opts.enableOffloadGlobalFiltering =
+      ci.getInvocation().getCodeGenOpts().OffloadGlobalFiltering;
 
-  if (doConcurrentMappingKind != DoConcurrentMappingKind::DCMK_None &&
+  if (opts.doConcurrentMappingKind != DoConcurrentMappingKind::DCMK_None &&
       !isOpenMPEnabled) {
     unsigned diagID = ci.getDiagnostics().getCustomDiagID(
         clang::DiagnosticsEngine::Warning,
@@ -349,15 +353,15 @@ bool CodeGenAction::beginSourceFileAction() {
   }
 
   if (isOpenMPEnabled) {
-    bool isDevice = false;
+    opts.isTargetDevice = false;
     if (auto offloadMod = llvm::dyn_cast<mlir::omp::OffloadModuleInterface>(
             mlirModule->getOperation()))
-      isDevice = offloadMod.getIsTargetDevice();
+      opts.isTargetDevice = offloadMod.getIsTargetDevice();
 
     // WARNING: This pipeline must be run immediately after the lowering to
     // ensure that the FIR is correct with respect to OpenMP operations/
     // attributes.
-    fir::createOpenMPFIRPassPipeline(pm, isDevice, doConcurrentMappingKind);
+    fir::createOpenMPFIRPassPipeline(pm, opts);
   }
 
   pm.enableVerifier(/*verifyPasses=*/true);
@@ -371,7 +375,6 @@ bool CodeGenAction::beginSourceFileAction() {
     return false;
   }
 
-
   // Print initial full MLIR module, before lowering or transformations, if
   // -save-temps has been specified.
   if (!saveMLIRTempFile(ci.getInvocation(), *mlirModule, getCurrentFile(),

flang/lib/Optimizer/OpenMP/GlobalFiltering.cpp

@@ -29,6 +29,8 @@ namespace flangomp {
 using namespace mlir;
 
 namespace {
+// TODO Remove this pass when AOMP moves to `clang-linker-wrapper` (instead of
+// `clang-offload-packager`).
 class GlobalFilteringPass
     : public flangomp::impl::GlobalFilteringPassBase<GlobalFilteringPass> {
 public:

flang/lib/Optimizer/Passes/Pipelines.cpp

@@ -241,18 +241,19 @@ void createHLFIRToFIRPassPipeline(mlir::PassManager &pm,
 /// \param pm - MLIR pass manager that will hold the pipeline definition.
 /// \param isTargetDevice - Whether code is being generated for a target device
 /// rather than the host device.
-void createOpenMPFIRPassPipeline(
-    mlir::PassManager &pm, bool isTargetDevice,
-    DoConcurrentMappingKind doConcurrentMappingKind) {
-  if (doConcurrentMappingKind != DoConcurrentMappingKind::DCMK_None)
+void createOpenMPFIRPassPipeline(mlir::PassManager &pm,
+                                 OpenMPFIRPassPipelineOpts opts) {
+  if (opts.doConcurrentMappingKind != DoConcurrentMappingKind::DCMK_None)
     pm.addPass(flangomp::createDoConcurrentConversionPass(
-        doConcurrentMappingKind == DoConcurrentMappingKind::DCMK_Device));
+        opts.doConcurrentMappingKind == DoConcurrentMappingKind::DCMK_Device));
 
   pm.addPass(flangomp::createMapInfoFinalizationPass());
   pm.addPass(flangomp::createMarkDeclareTargetPass());
-  if (isTargetDevice) {
+  if (opts.isTargetDevice) {
     pm.addPass(flangomp::createFunctionFilteringPass());
-    pm.addPass(flangomp::createGlobalFilteringPass());
+
+    if (opts.enableOffloadGlobalFiltering)
+      pm.addPass(flangomp::createGlobalFilteringPass());
   }
 }
 

flang/tools/bbc/bbc.cpp

@@ -295,15 +295,16 @@ static llvm::LogicalResult runOpenMPPasses(mlir::ModuleOp mlirModule) {
   using DoConcurrentMappingKind =
       Fortran::frontend::CodeGenOptions::DoConcurrentMappingKind;
 
-  auto doConcurrentMappingKind =
+  fir::OpenMPFIRPassPipelineOpts opts;
+  opts.isTargetDevice = enableOpenMPDevice;
+  opts.doConcurrentMappingKind =
       llvm::StringSwitch<DoConcurrentMappingKind>(
           enableDoConcurrentToOpenMPConversion)
           .Case("host", DoConcurrentMappingKind::DCMK_Host)
           .Case("device", DoConcurrentMappingKind::DCMK_Device)
           .Default(DoConcurrentMappingKind::DCMK_None);
 
-  fir::createOpenMPFIRPassPipeline(pm, enableOpenMPDevice,
-                                   doConcurrentMappingKind);
+  fir::createOpenMPFIRPassPipeline(pm, opts);
   (void)mlir::applyPassManagerCLOptions(pm);
   if (mlir::failed(pm.run(mlirModule))) {
     llvm::errs() << "FATAL: failed to correctly apply OpenMP pass pipeline";