[flang] Implement -mcmodel flag (#95411)

This patch implements the -mcmodel flag from clang, allowing the Code
Model to be changed for the LLVM module. The same set of mcmodel
flags are accepted as in clang and the same Code Model attributes are
added to the LLVM module for those flags.

Also add `-mlarge-data-threshold` for x86-64, which is automatically set
by the shared command-line code (see below). This is also added as an 
attribute into the LLVM module and on the target machine.

A function is created for `addMCModel` that is copied out of clang's
argument handling so that it can be shared with flang.

---------

Co-authored-by: Mats Petersson <[email protected]>

Author: DavidTruby

clang/include/clang/Driver/Options.td

@@ -4612,10 +4612,10 @@ def inline_asm_EQ : Joined<["-"], "inline-asm=">, Group<m_Group>,
   NormalizedValuesScope<"CodeGenOptions">, NormalizedValues<["IAD_ATT", "IAD_Intel"]>,
   MarshallingInfoEnum<CodeGenOpts<"InlineAsmDialect">, "IAD_ATT">;
 def mcmodel_EQ : Joined<["-"], "mcmodel=">, Group<m_Group>,
-  Visibility<[ClangOption, CC1Option]>,
+  Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
   MarshallingInfoString<TargetOpts<"CodeModel">, [{"default"}]>;
 def mlarge_data_threshold_EQ : Joined<["-"], "mlarge-data-threshold=">, Group<m_Group>,
-  Flags<[TargetSpecific]>, Visibility<[ClangOption, CC1Option]>,
+  Flags<[TargetSpecific]>, Visibility<[ClangOption, CC1Option,FlangOption, FC1Option]>,
   MarshallingInfoInt<TargetOpts<"LargeDataThreshold">, "0">;
 def mtls_size_EQ : Joined<["-"], "mtls-size=">, Group<m_Group>,
   Visibility<[ClangOption, CC1Option]>,

clang/lib/Driver/ToolChains/Clang.cpp

@@ -5909,81 +5909,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
 
   TC.addClangTargetOptions(Args, CmdArgs, JA.getOffloadingDeviceKind());
 
-  if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ)) {
-    StringRef CM = A->getValue();
-    bool Ok = false;
-    if (Triple.isOSAIX() && CM == "medium")
-      CM = "large";
-    if (Triple.isAArch64(64)) {
-      Ok = CM == "tiny" || CM == "small" || CM == "large";
-      if (CM == "large" && !Triple.isOSBinFormatMachO() &&
-          RelocationModel != llvm::Reloc::Static)
-        D.Diag(diag::err_drv_argument_only_allowed_with)
-            << A->getAsString(Args) << "-fno-pic";
-    } else if (Triple.isLoongArch()) {
-      if (CM == "extreme" &&
-          Args.hasFlagNoClaim(options::OPT_fplt, options::OPT_fno_plt, false))
-        D.Diag(diag::err_drv_argument_not_allowed_with)
-            << A->getAsString(Args) << "-fplt";
-      Ok = CM == "normal" || CM == "medium" || CM == "extreme";
-      // Convert to LLVM recognizable names.
-      if (Ok)
-        CM = llvm::StringSwitch<StringRef>(CM)
-                 .Case("normal", "small")
-                 .Case("extreme", "large")
-                 .Default(CM);
-    } else if (Triple.isPPC64() || Triple.isOSAIX()) {
-      Ok = CM == "small" || CM == "medium" || CM == "large";
-    } else if (Triple.isRISCV()) {
-      if (CM == "medlow")
-        CM = "small";
-      else if (CM == "medany")
-        CM = "medium";
-      Ok = CM == "small" || CM == "medium";
-    } else if (Triple.getArch() == llvm::Triple::x86_64) {
-      Ok = llvm::is_contained({"small", "kernel", "medium", "large", "tiny"},
-                              CM);
-    } else if (Triple.isNVPTX() || Triple.isAMDGPU() || Triple.isSPIRV()) {
-      // NVPTX/AMDGPU/SPIRV does not care about the code model and will accept
-      // whatever works for the host.
-      Ok = true;
-    } else if (Triple.isSPARC64()) {
-      if (CM == "medlow")
-        CM = "small";
-      else if (CM == "medmid")
-        CM = "medium";
-      else if (CM == "medany")
-        CM = "large";
-      Ok = CM == "small" || CM == "medium" || CM == "large";
-    }
-    if (Ok) {
-      CmdArgs.push_back(Args.MakeArgString("-mcmodel=" + CM));
-    } else {
-      D.Diag(diag::err_drv_unsupported_option_argument_for_target)
-          << A->getSpelling() << CM << TripleStr;
-    }
-  }
-
-  if (Triple.getArch() == llvm::Triple::x86_64) {
-    bool IsMediumCM = false;
-    bool IsLargeCM = false;
-    if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ)) {
-      IsMediumCM = StringRef(A->getValue()) == "medium";
-      IsLargeCM = StringRef(A->getValue()) == "large";
-    }
-    if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
-      if (!IsMediumCM && !IsLargeCM) {
-        D.Diag(diag::warn_drv_large_data_threshold_invalid_code_model)
-            << A->getOption().getRenderName();
-      } else {
-        A->render(Args, CmdArgs);
-      }
-    } else if (IsMediumCM) {
-      CmdArgs.push_back("-mlarge-data-threshold=65536");
-    } else if (IsLargeCM) {
-      CmdArgs.push_back("-mlarge-data-threshold=0");
-    }
-  }
+  addMCModel(D, Args, Triple, RelocationModel, CmdArgs);
 
   if (Arg *A = Args.getLastArg(options::OPT_mtls_size_EQ)) {
     StringRef Value = A->getValue();

clang/lib/Driver/ToolChains/CommonArgs.cpp

@@ -2825,3 +2825,84 @@ void tools::addOffloadCompressArgs(const llvm::opt::ArgList &TCArgs,
     CmdArgs.push_back(
         TCArgs.MakeArgString(Twine("-compression-level=") + Arg->getValue()));
 }
+
+void tools::addMCModel(const Driver &D, const llvm::opt::ArgList &Args,
+                       const llvm::Triple &Triple,
+                       const llvm::Reloc::Model &RelocationModel,
+                       llvm::opt::ArgStringList &CmdArgs) {
+  if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ)) {
+    StringRef CM = A->getValue();
+    bool Ok = false;
+    if (Triple.isOSAIX() && CM == "medium")
+      CM = "large";
+    if (Triple.isAArch64(64)) {
+      Ok = CM == "tiny" || CM == "small" || CM == "large";
+      if (CM == "large" && !Triple.isOSBinFormatMachO() &&
+          RelocationModel != llvm::Reloc::Static)
+        D.Diag(diag::err_drv_argument_only_allowed_with)
+            << A->getAsString(Args) << "-fno-pic";
+    } else if (Triple.isLoongArch()) {
+      if (CM == "extreme" &&
+          Args.hasFlagNoClaim(options::OPT_fplt, options::OPT_fno_plt, false))
+        D.Diag(diag::err_drv_argument_not_allowed_with)
+            << A->getAsString(Args) << "-fplt";
+      Ok = CM == "normal" || CM == "medium" || CM == "extreme";
+      // Convert to LLVM recognizable names.
+      if (Ok)
+        CM = llvm::StringSwitch<StringRef>(CM)
+                 .Case("normal", "small")
+                 .Case("extreme", "large")
+                 .Default(CM);
+    } else if (Triple.isPPC64() || Triple.isOSAIX()) {
+      Ok = CM == "small" || CM == "medium" || CM == "large";
+    } else if (Triple.isRISCV()) {
+      if (CM == "medlow")
+        CM = "small";
+      else if (CM == "medany")
+        CM = "medium";
+      Ok = CM == "small" || CM == "medium";
+    } else if (Triple.getArch() == llvm::Triple::x86_64) {
+      Ok = llvm::is_contained({"small", "kernel", "medium", "large", "tiny"},
+                              CM);
+    } else if (Triple.isNVPTX() || Triple.isAMDGPU() || Triple.isSPIRV()) {
+      // NVPTX/AMDGPU/SPIRV does not care about the code model and will accept
+      // whatever works for the host.
+      Ok = true;
+    } else if (Triple.isSPARC64()) {
+      if (CM == "medlow")
+        CM = "small";
+      else if (CM == "medmid")
+        CM = "medium";
+      else if (CM == "medany")
+        CM = "large";
+      Ok = CM == "small" || CM == "medium" || CM == "large";
+    }
+    if (Ok) {
+      CmdArgs.push_back(Args.MakeArgString("-mcmodel=" + CM));
+    } else {
+      D.Diag(diag::err_drv_unsupported_option_argument_for_target)
+          << A->getSpelling() << CM << Triple.getTriple();
+    }
+  }
+
+  if (Triple.getArch() == llvm::Triple::x86_64) {
+    bool IsMediumCM = false;
+    bool IsLargeCM = false;
+    if (Arg *A = Args.getLastArg(options::OPT_mcmodel_EQ)) {
+      IsMediumCM = StringRef(A->getValue()) == "medium";
+      IsLargeCM = StringRef(A->getValue()) == "large";
+    }
+    if (Arg *A = Args.getLastArg(options::OPT_mlarge_data_threshold_EQ)) {
+      if (!IsMediumCM && !IsLargeCM) {
+        D.Diag(diag::warn_drv_large_data_threshold_invalid_code_model)
+            << A->getOption().getRenderName();
+      } else {
+        A->render(Args, CmdArgs);
+      }
+    } else if (IsMediumCM) {
+      CmdArgs.push_back("-mlarge-data-threshold=65536");
+    } else if (IsLargeCM) {
+      CmdArgs.push_back("-mlarge-data-threshold=0");
+    }
+  }
+}

clang/lib/Driver/ToolChains/CommonArgs.h

@@ -223,6 +223,10 @@ void addOutlineAtomicsArgs(const Driver &D, const ToolChain &TC,
                            const llvm::Triple &Triple);
 void addOffloadCompressArgs(const llvm::opt::ArgList &TCArgs,
                             llvm::opt::ArgStringList &CmdArgs);
+void addMCModel(const Driver &D, const llvm::opt::ArgList &Args,
+                const llvm::Triple &Triple,
+                const llvm::Reloc::Model &RelocationModel,
+                llvm::opt::ArgStringList &CmdArgs);
 
 } // end namespace tools
 } // end namespace driver

clang/lib/Driver/ToolChains/Flang.cpp

@@ -735,6 +735,11 @@ void Flang::ConstructJob(Compilation &C, const JobAction &JA,
   // Add target args, features, etc.
   addTargetOptions(Args, CmdArgs);
 
+  llvm::Reloc::Model RelocationModel =
+      std::get<0>(ParsePICArgs(getToolChain(), Args));
+  // Add MCModel information
+  addMCModel(D, Args, Triple, RelocationModel, CmdArgs);
+
   // Add Codegen options
   addCodegenOptions(Args, CmdArgs);
 

flang/include/flang/Frontend/CodeGenOptions.h

@@ -133,6 +133,13 @@ class CodeGenOptions : public CodeGenOptionsBase {
   /// transformation.
   OptRemark OptimizationRemarkAnalysis;
 
+  /// The code model to use (-mcmodel).
+  std::string CodeModel;
+
+  /// The code model-specific large data threshold to use
+  /// (-mlarge-data-threshold).
+  uint64_t LargeDataThreshold;
+
   // Define accessors/mutators for code generation options of enumeration type.
 #define CODEGENOPT(Name, Bits, Default)
 #define ENUM_CODEGENOPT(Name, Type, Bits, Default)                             \
@@ -143,6 +150,8 @@ class CodeGenOptions : public CodeGenOptionsBase {
   CodeGenOptions();
 };
 
+std::optional<llvm::CodeModel::Model> getCodeModel(llvm::StringRef string);
+
 } // end namespace Fortran::frontend
 
 #endif // FORTRAN_FRONTEND_CODEGENOPTIONS_H

flang/lib/Frontend/CodeGenOptions.cpp

@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "flang/Frontend/CodeGenOptions.h"
+#include <optional>
 #include <string.h>
 
 namespace Fortran::frontend {
@@ -21,4 +22,14 @@ CodeGenOptions::CodeGenOptions() {
 #include "flang/Frontend/CodeGenOptions.def"
 }
 
+std::optional<llvm::CodeModel::Model> getCodeModel(llvm::StringRef string) {
+  return llvm::StringSwitch<std::optional<llvm::CodeModel::Model>>(string)
+      .Case("tiny", llvm::CodeModel::Model::Tiny)
+      .Case("small", llvm::CodeModel::Model::Small)
+      .Case("kernel", llvm::CodeModel::Model::Kernel)
+      .Case("medium", llvm::CodeModel::Model::Medium)
+      .Case("large", llvm::CodeModel::Model::Large)
+      .Default(std::nullopt);
+}
+
 } // end namespace Fortran::frontend

flang/lib/Frontend/CompilerInstance.cpp

@@ -321,11 +321,19 @@ bool CompilerInstance::setUpTargetMachine() {
   assert(OptLevelOrNone && "Invalid optimization level!");
   llvm::CodeGenOptLevel OptLevel = *OptLevelOrNone;
   std::string featuresStr = getTargetFeatures();
+  std::optional<llvm::CodeModel::Model> cm = getCodeModel(CGOpts.CodeModel);
   targetMachine.reset(theTarget->createTargetMachine(
       theTriple, /*CPU=*/targetOpts.cpu,
       /*Features=*/featuresStr, llvm::TargetOptions(),
       /*Reloc::Model=*/CGOpts.getRelocationModel(),
-      /*CodeModel::Model=*/std::nullopt, OptLevel));
+      /*CodeModel::Model=*/cm, OptLevel));
   assert(targetMachine && "Failed to create TargetMachine");
+  if (cm.has_value()) {
+    const llvm::Triple triple(theTriple);
+    if ((cm == llvm::CodeModel::Medium || cm == llvm::CodeModel::Large) &&
+        triple.getArch() == llvm::Triple::x86_64) {
+      targetMachine->setLargeDataThreshold(CGOpts.LargeDataThreshold);
+    }
+  }
   return true;
 }

flang/lib/Frontend/CompilerInvocation.cpp

@@ -32,6 +32,7 @@
 #include "llvm/Option/Arg.h"
 #include "llvm/Option/ArgList.h"
 #include "llvm/Option/OptTable.h"
+#include "llvm/Support/CodeGen.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/FileUtilities.h"
 #include "llvm/Support/Path.h"
@@ -386,6 +387,29 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
       opts.IsPIE = 1;
   }
 
+  // -mcmodel option.
+  if (const llvm::opt::Arg *a =
+          args.getLastArg(clang::driver::options::OPT_mcmodel_EQ)) {
+    llvm::StringRef modelName = a->getValue();
+    std::optional<llvm::CodeModel::Model> codeModel = getCodeModel(modelName);
+
+    if (codeModel.has_value())
+      opts.CodeModel = modelName;
+    else
+      diags.Report(clang::diag::err_drv_invalid_value)
+          << a->getAsString(args) << modelName;
+  }
+
+  if (const llvm::opt::Arg *arg = args.getLastArg(
+          clang::driver::options::OPT_mlarge_data_threshold_EQ)) {
+    uint64_t LDT;
+    if (llvm::StringRef(arg->getValue()).getAsInteger(/*Radix=*/10, LDT)) {
+      diags.Report(clang::diag::err_drv_invalid_value)
+          << arg->getSpelling() << arg->getValue();
+    }
+    opts.LargeDataThreshold = LDT;
+  }
+
   // This option is compatible with -f[no-]underscoring in gfortran.
   if (args.hasFlag(clang::driver::options::OPT_fno_underscoring,
                    clang::driver::options::OPT_funderscoring, false)) {

flang/lib/Frontend/FrontendActions.cpp

@@ -866,6 +866,17 @@ void CodeGenAction::generateLLVMIR() {
       llvmModule->setPIELevel(
           static_cast<llvm::PIELevel::Level>(opts.PICLevel));
   }
+
+  // Set mcmodel level LLVM module flags
+  std::optional<llvm::CodeModel::Model> cm = getCodeModel(opts.CodeModel);
+  if (cm.has_value()) {
+    const llvm::Triple triple(ci.getInvocation().getTargetOpts().triple);
+    llvmModule->setCodeModel(*cm);
+    if ((cm == llvm::CodeModel::Medium || cm == llvm::CodeModel::Large) &&
+        triple.getArch() == llvm::Triple::x86_64) {
+      llvmModule->setLargeDataThreshold(opts.LargeDataThreshold);
+    }
+  }
 }
 
 static std::unique_ptr<llvm::raw_pwrite_stream>
@@ -1280,6 +1291,7 @@ void CodeGenAction::executeAction() {
   // and the command-line target option if specified, or the default if not
   // given on the command-line).
   llvm::TargetMachine &targetMachine = ci.getTargetMachine();
+
   const std::string &theTriple = targetMachine.getTargetTriple().str();
 
   if (llvmModule->getTargetTriple() != theTriple) {

flang/test/Driver/large-data-threshold.f90

@@ -0,0 +1,18 @@
+! RUN: %flang -### -c --target=x86_64 -mcmodel=large -mlarge-data-threshold=32768 %s 2>&1 | FileCheck %s
+! RUN: %flang -### -c --target=x86_64 -mcmodel=large -mlarge-data-threshold=59000 %s 2>&1 | FileCheck %s --check-prefix=CHECK-59000
+! RUN: %flang -### -c --target=x86_64 -mcmodel=large -mlarge-data-threshold=1048576 %s 2>&1 | FileCheck %s --check-prefix=CHECK-1M
+! RUN: not %flang -c --target=x86_64 -mcmodel=large -mlarge-data-threshold=nonsense %s 2>&1 | FileCheck %s --check-prefix=INVALID
+! RUN: %flang -### -c --target=x86_64 -mlarge-data-threshold=32768 %s 2>&1 | FileCheck %s --check-prefix=NO-MCMODEL
+! RUN: %flang -### -c --target=x86_64 -mcmodel=small -mlarge-data-threshold=32768 %s 2>&1 | FileCheck %s --check-prefix=NO-MCMODEL
+! RUN: not %flang -### -c --target=aarch64 -mcmodel=small -mlarge-data-threshold=32768 %s 2>&1 | FileCheck %s --check-prefix=NOT-SUPPORTED
+  
+  
+! CHECK: "{{.*}}flang-new" "-fc1"
+! CHECK-SAME: "-mlarge-data-threshold=32768"
+! CHECK-59000: "{{.*}}flang-new" "-fc1"
+! CHECK-59000-SAME: "-mlarge-data-threshold=59000"
+! CHECK-1M: "{{.*}}flang-new" "-fc1"
+! CHECK-1M-SAME: "-mlarge-data-threshold=1048576"
+! NO-MCMODEL: 'mlarge-data-threshold=' only applies to medium and large code models
+! INVALID: error: invalid value 'nonsense' in '-mlarge-data-threshold='
+! NOT-SUPPORTED: error: unsupported option '-mlarge-data-threshold=' for target 'aarch64'

flang/test/Driver/mcmodel.f90

@@ -0,0 +1,44 @@
+! RUN: not %flang -### -c --target=i686 -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=ERR-MEDIUM %s
+! RUN: %flang --target=x86_64 -### -c -mcmodel=tiny %s 2>&1 | FileCheck --check-prefix=TINY %s
+! RUN: %flang --target=x86_64 -### -c -mcmodel=small %s 2>&1 | FileCheck --check-prefix=SMALL %s
+! RUN: %flang --target=x86_64 -### -S -mcmodel=kernel %s 2>&1 | FileCheck --check-prefix=KERNEL %s
+! RUN: %flang --target=x86_64 -### -c -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=MEDIUM %s
+! RUN: %flang --target=x86_64 -### -S -mcmodel=large %s 2>&1 | FileCheck --check-prefix=LARGE %s
+! RUN: not %flang -### -c --target=powerpc-linux-gnu -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=ERR-MEDIUM %s
+! RUN: %flang --target=powerpc-unknown-aix -### -S -mcmodel=small %s 2>&1 | FileCheck --check-prefix=SMALL %s
+! RUN: %flang --target=powerpc-unknown-aix -### -S -mcmodel=large %s 2>&1 | FileCheck --check-prefix=LARGE %s
+! RUN: %flang --target=powerpc-unknown-aix -### -S -mcmodel=medium %s 2> %t.log
+! RUN: FileCheck --check-prefix=AIX-MCMEDIUM-OVERRIDE %s < %t.log
+! RUN: not %flang -### -c -mcmodel=lager %s 2>&1 | FileCheck --check-prefix=INVALID %s
+! RUN: %flang --target=aarch64 -### -S -mcmodel=large -fno-pic %s 2>&1 | FileCheck --check-prefix=LARGE %s
+! RUN: %flang --target=aarch64-apple-macosx -### -S -mcmodel=large %s 2>&1 | FileCheck --check-prefix=LARGE %s
+! RUN: not %flang --target=aarch64 -### -S -mcmodel=large -fpic %s 2>&1 | FileCheck --check-prefix=AARCH64-PIC-LARGE %s
+! RUN: not %flang -### -c --target=aarch64 -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=ERR-MEDIUM %s
+! RUN: not %flang -### -c --target=aarch64 -mcmodel=kernel %s 2>&1 | FileCheck --check-prefix=ERR-KERNEL %s
+! RUN: not %flang --target=aarch64_32-linux -### -S -mcmodel=small %s 2>&1 | FileCheck --check-prefix=ERR-AARCH64_32 %s
+! RUN: %flang --target=loongarch64 -### -S -mcmodel=normal %s 2>&1 | FileCheck --check-prefix=SMALL %s
+! RUN: %flang --target=loongarch64 -### -S -mcmodel=medium %s 2>&1 | FileCheck --check-prefix=MEDIUM %s
+! RUN: %flang --target=loongarch64 -### -S -mcmodel=extreme %s 2>&1 | FileCheck --check-prefix=LARGE %s
+! RUN: not %flang --target=loongarch64 -### -S -mcmodel=tiny %s 2>&1 | FileCheck --check-prefix=ERR-TINY %s
+! RUN: not %flang --target=loongarch64 -### -S -mcmodel=small %s 2>&1 | FileCheck --check-prefix=ERR-SMALL %s
+! RUN: not %flang --target=loongarch64 -### -S -mcmodel=kernel %s 2>&1 | FileCheck --check-prefix=ERR-KERNEL %s
+! RUN: not %flang --target=loongarch64 -### -S -mcmodel=large %s 2>&1 | FileCheck --check-prefix=ERR-LARGE %s
+
+! TINY: "-mcmodel=tiny"
+! SMALL: "-mcmodel=small"
+! KERNEL: "-mcmodel=kernel"
+! MEDIUM: "-mcmodel=medium"
+! LARGE: "-mcmodel=large"
+! AIX-MCMEDIUM-OVERRIDE: "-mcmodel=large"
+
+! INVALID: error: unsupported argument 'lager' to option '-mcmodel=' for target '{{.*}}'
+
+! ERR-TINY:   error: unsupported argument 'tiny' to option '-mcmodel=' for target '{{.*}}'
+! ERR-SMALL:  error: unsupported argument 'small' to option '-mcmodel=' for target '{{.*}}'
+! ERR-MEDIUM: error: unsupported argument 'medium' to option '-mcmodel=' for target '{{.*}}'
+! ERR-KERNEL: error: unsupported argument 'kernel' to option '-mcmodel=' for target '{{.*}}'
+! ERR-LARGE:  error: unsupported argument 'large' to option '-mcmodel=' for target '{{.*}}'
+
+! AARCH64-PIC-LARGE: error: invalid argument '-mcmodel=large' only allowed with '-fno-pic'
+! ERR-AARCH64_32: error: unsupported argument 'small' to option '-mcmodel=' for target 'aarch64_32-unknown-linux'
+