[flang-commits] [flang] [clang] [Flang] Add code-object-version option (PR #72638)
Dominik Adamski via flang-commits
flang-commits at lists.llvm.org
Fri Nov 17 03:54:26 PST 2023
https://github.com/DominikAdamski updated https://github.com/llvm/llvm-project/pull/72638
>From e5d339c24193e4e37013b3b25460009418d6ce6d Mon Sep 17 00:00:00 2001
From: Dominik Adamski <dominik.adamski at amd.com>
Date: Fri, 17 Nov 2023 03:02:49 -0600
Subject: [PATCH] [Flang] Add code-object-version option
Information about code object version can be configured by the user
for AMD GPU target and it needs to be placed in LLVM IR generated
by Flang.
Information about code object version in MLIR generated by the
parser can be reused by other tools. There is no need to specify
extra flags if we want to invoke MLIR tools separately.
---
clang/include/clang/Driver/Options.td | 2 +-
clang/lib/Driver/ToolChains/Flang.cpp | 11 ++++++
clang/lib/Driver/ToolChains/Flang.h | 7 ++++
flang/include/flang/Frontend/CodeGenOptions.h | 13 +++++++
flang/lib/Frontend/CompilerInvocation.cpp | 9 +++++
flang/lib/Frontend/FrontendActions.cpp | 39 +++++++++++++++++--
flang/test/Driver/code-object-version.f90 | 8 ++++
flang/test/Driver/driver-help-hidden.f90 | 2 +
flang/test/Driver/driver-help.f90 | 4 ++
flang/test/Lower/AMD/code_object_version.f90 | 11 ++++++
10 files changed, 101 insertions(+), 5 deletions(-)
create mode 100644 flang/test/Driver/code-object-version.f90
create mode 100644 flang/test/Lower/AMD/code_object_version.f90
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 811550416110b3d..e7eb94d174e75f8 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -4688,7 +4688,7 @@ defm amdgpu_ieee : BoolOption<"m", "amdgpu-ieee",
def mcode_object_version_EQ : Joined<["-"], "mcode-object-version=">, Group<m_Group>,
HelpText<"Specify code object ABI version. Defaults to 4. (AMDGPU only)">,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, FlangOption, CC1Option, FC1Option]>,
Values<"none,4,5">,
NormalizedValuesScope<"TargetOptions">,
NormalizedValues<["COV_None", "COV_4", "COV_5"]>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 8bdd920c3dcbb79..e60c11bfbe8e38b 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -204,6 +204,14 @@ void Flang::AddAArch64TargetArgs(const ArgList &Args,
}
}
+void Flang::AddAMDGPUTargetArgs(const ArgList &Args,
+ ArgStringList &CmdArgs) const {
+ if (Arg *A = Args.getLastArg(options::OPT_mcode_object_version_EQ)) {
+ StringRef Val = A->getValue();
+ CmdArgs.push_back(Args.MakeArgString("-mcode-object-version=" + Val));
+ }
+}
+
void Flang::addTargetOptions(const ArgList &Args,
ArgStringList &CmdArgs) const {
const ToolChain &TC = getToolChain();
@@ -227,6 +235,9 @@ void Flang::addTargetOptions(const ArgList &Args,
case llvm::Triple::r600:
case llvm::Triple::amdgcn:
+ getTargetFeatures(D, Triple, Args, CmdArgs, /*ForAs*/ false);
+ AddAMDGPUTargetArgs(Args, CmdArgs);
+ break;
case llvm::Triple::riscv64:
case llvm::Triple::x86_64:
getTargetFeatures(D, Triple, Args, CmdArgs, /*ForAs*/ false);
diff --git a/clang/lib/Driver/ToolChains/Flang.h b/clang/lib/Driver/ToolChains/Flang.h
index 0141240b5d3ac90..8d35080e1c0c88b 100644
--- a/clang/lib/Driver/ToolChains/Flang.h
+++ b/clang/lib/Driver/ToolChains/Flang.h
@@ -63,6 +63,13 @@ class LLVM_LIBRARY_VISIBILITY Flang : public Tool {
void AddAArch64TargetArgs(const llvm::opt::ArgList &Args,
llvm::opt::ArgStringList &CmdArgs) const;
+ /// Add specific options for AMDGPU target.
+ ///
+ /// \param [in] Args The list of input driver arguments
+ /// \param [out] CmdArgs The list of output command arguments
+ void AddAMDGPUTargetArgs(const llvm::opt::ArgList &Args,
+ llvm::opt::ArgStringList &CmdArgs) const;
+
/// Extract offload options from the driver arguments and add them to
/// the command arguments.
/// \param [in] C The current compilation for the driver invocation
diff --git a/flang/include/flang/Frontend/CodeGenOptions.h b/flang/include/flang/Frontend/CodeGenOptions.h
index b86bb88610a9a4a..8d938c361a0aa23 100644
--- a/flang/include/flang/Frontend/CodeGenOptions.h
+++ b/flang/include/flang/Frontend/CodeGenOptions.h
@@ -85,6 +85,19 @@ class CodeGenOptions : public CodeGenOptionsBase {
RK_WithPattern, // Remark pattern specified via '-Rgroup=regexp'.
};
+ /// \brief Enumeration value for AMDGPU code object version, which is the
+ /// code object version times 100.
+ enum class CodeObjectVersionKind {
+ COV_None,
+ COV_2 = 200, // Unsupported.
+ COV_3 = 300, // Unsupported.
+ COV_4 = 400,
+ COV_5 = 500,
+ };
+
+ /// \brief Code object version for AMDGPU.
+ CodeObjectVersionKind CodeObjectVersion = CodeObjectVersionKind::COV_None;
+
/// Optimization remark with an optional regular expression pattern.
struct OptRemark {
RemarkKind Kind = RemarkKind::RK_Missing;
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index cb4f2d6a6225205..d96d313f85e8185 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -266,6 +266,15 @@ static void parseCodeGenArgs(Fortran::frontend::CodeGenOptions &opts,
opts.PrepareForThinLTO = true;
}
+ if (const llvm::opt::Arg *a = args.getLastArg(
+ clang::driver::options::OPT_mcode_object_version_EQ)) {
+ llvm::StringRef s = a->getValue();
+ if (s == "5")
+ opts.CodeObjectVersion = CodeGenOptions::CodeObjectVersionKind::COV_5;
+ if (s == "4")
+ opts.CodeObjectVersion = CodeGenOptions::CodeObjectVersionKind::COV_4;
+ }
+
// -f[no-]save-optimization-record[=<format>]
if (const llvm::opt::Arg *a =
args.getLastArg(clang::driver::options::OPT_opt_record_file))
diff --git a/flang/lib/Frontend/FrontendActions.cpp b/flang/lib/Frontend/FrontendActions.cpp
index 1e4a2f3035f1f7f..6888ba1cd8b2f3e 100644
--- a/flang/lib/Frontend/FrontendActions.cpp
+++ b/flang/lib/Frontend/FrontendActions.cpp
@@ -244,8 +244,7 @@ static void setMLIRDataLayout(mlir::ModuleOp &mlirModule,
mlirModule->setAttr(mlir::DLTIDialect::kDataLayoutAttrName, dlSpec);
}
-static void addDepdendentLibs(mlir::ModuleOp &mlirModule,
- CompilerInstance &ci) {
+static void addDependentLibs(mlir::ModuleOp &mlirModule, CompilerInstance &ci) {
const std::vector<std::string> &libs =
ci.getInvocation().getCodeGenOpts().DependentLibs;
if (libs.empty()) {
@@ -264,6 +263,37 @@ static void addDepdendentLibs(mlir::ModuleOp &mlirModule,
}
}
+// Add to MLIR code target specific items which are dependent on target
+// configuration specified by the user
+static void addTargetSpecificMLIRItems(mlir::ModuleOp &mlirModule,
+ CompilerInstance &ci) {
+ const TargetOptions &targetOpts = ci.getInvocation().getTargetOpts();
+ const llvm::Triple triple(targetOpts.triple);
+ if (triple.isAMDGPU()) {
+ unsigned oclcABIVERsion;
+ const unsigned defaultOclcABIVERsion = 400;
+ mlir::OpBuilder builder(mlirModule.getContext());
+ const CodeGenOptions &codeGenOpts = ci.getInvocation().getCodeGenOpts();
+ if (codeGenOpts.CodeObjectVersion ==
+ CodeGenOptions::CodeObjectVersionKind::COV_None)
+ oclcABIVERsion = defaultOclcABIVERsion;
+ else
+ oclcABIVERsion = static_cast<unsigned>(codeGenOpts.CodeObjectVersion);
+
+ auto int32Type = builder.getI32Type();
+ auto covInfo = builder.create<mlir::LLVM::GlobalOp>(
+ mlirModule.getLoc(), int32Type, true, mlir::LLVM::Linkage::WeakODR,
+ "__oclc_ABI_version",
+ builder.getIntegerAttr(int32Type, oclcABIVERsion));
+ covInfo.setUnnamedAddr(mlir::LLVM::UnnamedAddr::Local);
+ covInfo.setAddrSpace(4);
+ covInfo.setVisibility_(mlir::LLVM::Visibility::Hidden);
+ builder.setInsertionPointToStart(mlirModule.getBody());
+ builder.insert(covInfo);
+ }
+ addDependentLibs(mlirModule, ci);
+}
+
bool CodeGenAction::beginSourceFileAction() {
llvmCtx = std::make_unique<llvm::LLVMContext>();
CompilerInstance &ci = this->getInstance();
@@ -365,8 +395,9 @@ bool CodeGenAction::beginSourceFileAction() {
Fortran::parser::Program &parseTree{*ci.getParsing().parseTree()};
lb.lower(parseTree, ci.getInvocation().getSemanticsContext());
- // Add dependent libraries
- addDepdendentLibs(*mlirModule, ci);
+ // Add target specific items like dependent libraries, target specific
+ // constants etc.
+ addTargetSpecificMLIRItems(*mlirModule, ci);
// run the default passes.
mlir::PassManager pm((*mlirModule)->getName(),
diff --git a/flang/test/Driver/code-object-version.f90 b/flang/test/Driver/code-object-version.f90
new file mode 100644
index 000000000000000..cd88ff7fe29a282
--- /dev/null
+++ b/flang/test/Driver/code-object-version.f90
@@ -0,0 +1,8 @@
+! RUN: not %flang -target amdgcn-amd-amdhsa -target-cpu gfx908 -mcode-object-version=3 -S %s -o \
+! RUN: /dev/null 2>&1 | FileCheck --check-prefix=INVALID_VERSION %s
+
+! RUN: %flang -target x86_64-unknown-linux-gnu -mcode-object-version=3 -S %s -o \
+! RUN: /dev/null 2>&1 | FileCheck --check-prefix=UNUSED_PARAM %s
+
+! INVALID_VERSION: error: invalid integral value '3' in '-mcode-object-version=3'
+! UNUSED_PARAM: warning: argument unused during compilation: '-mcode-object-version=3' [-Wunused-command-line-argument]
diff --git a/flang/test/Driver/driver-help-hidden.f90 b/flang/test/Driver/driver-help-hidden.f90
index b276f1906e1a457..4ea89776416a6b8 100644
--- a/flang/test/Driver/driver-help-hidden.f90
+++ b/flang/test/Driver/driver-help-hidden.f90
@@ -114,6 +114,8 @@
! CHECK-NEXT: -I <dir> Add directory to the end of the list of include search paths
! CHECK-NEXT: -L <dir> Add directory to library search path
! CHECK-NEXT: -march=<value> For a list of available architectures for the target use '-mcpu=help'
+! CHECK-NEXT: -mcode-object-version=<value>
+! CHECK-NEXT: Specify code object ABI version. Defaults to 4. (AMDGPU only)
! CHECK-NEXT: -mcpu=<value> For a list of available CPUs for the target use '-mcpu=help'
! CHECK-NEXT: -mllvm=<arg> Alias for -mllvm
! CHECK-NEXT: -mllvm <value> Additional arguments to forward to LLVM's option processing
diff --git a/flang/test/Driver/driver-help.f90 b/flang/test/Driver/driver-help.f90
index 452c62541e72e61..6fb306d3196fbae 100644
--- a/flang/test/Driver/driver-help.f90
+++ b/flang/test/Driver/driver-help.f90
@@ -100,6 +100,8 @@
! HELP-NEXT: -I <dir> Add directory to the end of the list of include search paths
! HELP-NEXT: -L <dir> Add directory to library search path
! HELP-NEXT: -march=<value> For a list of available architectures for the target use '-mcpu=help'
+! HELP-NEXT: -mcode-object-version=<value>
+! HELP-NEXT: Specify code object ABI version. Defaults to 4. (AMDGPU only)
! HELP-NEXT: -mcpu=<value> For a list of available CPUs for the target use '-mcpu=help'
! HELP-NEXT: -mllvm=<arg> Alias for -mllvm
! HELP-NEXT: -mllvm <value> Additional arguments to forward to LLVM's option processing
@@ -232,6 +234,8 @@
! HELP-FC1-NEXT: -init-only Only execute frontend initialization
! HELP-FC1-NEXT: -I <dir> Add directory to the end of the list of include search paths
! HELP-FC1-NEXT: -load <dsopath> Load the named plugin (dynamic shared object)
+! HELP-FC1-NEXT: -mcode-object-version=<value>
+! HELP-FC1-NEXT: Specify code object ABI version. Defaults to 4. (AMDGPU only)
! HELP-FC1-NEXT: -menable-no-infs Allow optimization to assume there are no infinities.
! HELP-FC1-NEXT: -menable-no-nans Allow optimization to assume there are no NaNs.
! HELP-FC1-NEXT: -mllvm <value> Additional arguments to forward to LLVM's option processing
diff --git a/flang/test/Lower/AMD/code_object_version.f90 b/flang/test/Lower/AMD/code_object_version.f90
new file mode 100644
index 000000000000000..160d2c9fb784e4c
--- /dev/null
+++ b/flang/test/Lower/AMD/code_object_version.f90
@@ -0,0 +1,11 @@
+!REQUIRES: amdgpu-registered-target
+!RUN: %flang_fc1 -emit-hlfir -triple amdgcn-amd-amdhsa -target-cpu gfx908 %s -o - | FileCheck --check-prefix=COV_NONE %s
+!RUN: %flang_fc1 -emit-hlfir -triple amdgcn-amd-amdhsa -target-cpu gfx908 -mcode-object-version=4 %s -o - | FileCheck --check-prefix=COV_4 %s
+!RUN: %flang_fc1 -emit-hlfir -triple amdgcn-amd-amdhsa -target-cpu gfx908 -mcode-object-version=5 %s -o - | FileCheck --check-prefix=COV_5 %s
+
+!COV_NONE: llvm.mlir.global weak_odr hidden local_unnamed_addr constant @__oclc_ABI_version(400 : i32) {addr_space = 4 : i32} : i32
+!COV_4: llvm.mlir.global weak_odr hidden local_unnamed_addr constant @__oclc_ABI_version(400 : i32) {addr_space = 4 : i32} : i32
+!COV_5: llvm.mlir.global weak_odr hidden local_unnamed_addr constant @__oclc_ABI_version(500 : i32) {addr_space = 4 : i32} : i32
+subroutine target_simple
+end subroutine target_simple
+
More information about the flang-commits
mailing list