https://github.com/skatrak updated
https://github.com/llvm/llvm-project/pull/100155
>From 54e52e8a37fd725976e157cd0f9e0221a355dead Mon Sep 17 00:00:00 2001
From: Sergio Afonso <safon...@amd.com>
Date: Tue, 23 Jul 2024 16:40:18 +0100
Subject: [PATCH 1/2] [Flang][OpenMP] Add frontend support for -fopenmp-targets
This patch adds support for the `-fopenmp-targets` option to the `bbc` and
`flang -fc1` tools. It adds an `OMPTargetTriples` property to the `LangOptions`
structure, which is filled with the triples represented by the compiler option.
This is used to initialize the `omp.target_triples` module attribute for later
use by lowering stages.
---
flang/include/flang/Frontend/LangOptions.h | 6 ++++
flang/include/flang/Tools/CrossToolHelpers.h | 14 ++++++--
flang/lib/Frontend/CompilerInvocation.cpp | 35 ++++++++++++++++++++
flang/test/Lower/OpenMP/offload-targets.f90 | 10 ++++++
flang/tools/bbc/bbc.cpp | 13 +++++++-
5 files changed, 74 insertions(+), 4 deletions(-)
create mode 100644 flang/test/Lower/OpenMP/offload-targets.f90
diff --git a/flang/include/flang/Frontend/LangOptions.h
b/flang/include/flang/Frontend/LangOptions.h
index 7ab2195818863..57d86d46df5ab 100644
--- a/flang/include/flang/Frontend/LangOptions.h
+++ b/flang/include/flang/Frontend/LangOptions.h
@@ -16,6 +16,9 @@
#define FORTRAN_FRONTEND_LANGOPTIONS_H
#include <string>
+#include <vector>
+
+#include "llvm/TargetParser/Triple.h"
namespace Fortran::frontend {
@@ -58,6 +61,9 @@ class LangOptions : public LangOptionsBase {
/// host code generation.
std::string OMPHostIRFile;
+ /// List of triples passed in using -fopenmp-targets.
+ std::vector<llvm::Triple> OMPTargetTriples;
+
LangOptions();
};
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h
b/flang/include/flang/Tools/CrossToolHelpers.h
index 1d890fd8e1f6f..75fd783af237d 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -131,7 +131,9 @@ struct OffloadModuleOpts {
bool OpenMPThreadSubscription, bool OpenMPNoThreadState,
bool OpenMPNoNestedParallelism, bool OpenMPIsTargetDevice,
bool OpenMPIsGPU, bool OpenMPForceUSM, uint32_t OpenMPVersion,
- std::string OMPHostIRFile = {}, bool NoGPULib = false)
+ std::string OMPHostIRFile = {},
+ const std::vector<llvm::Triple> &OMPTargetTriples = {},
+ bool NoGPULib = false)
: OpenMPTargetDebug(OpenMPTargetDebug),
OpenMPTeamSubscription(OpenMPTeamSubscription),
OpenMPThreadSubscription(OpenMPThreadSubscription),
@@ -139,7 +141,9 @@ struct OffloadModuleOpts {
OpenMPNoNestedParallelism(OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(OpenMPIsTargetDevice), OpenMPIsGPU(OpenMPIsGPU),
OpenMPForceUSM(OpenMPForceUSM), OpenMPVersion(OpenMPVersion),
- OMPHostIRFile(OMPHostIRFile), NoGPULib(NoGPULib) {}
+ OMPHostIRFile(OMPHostIRFile),
+ OMPTargetTriples(OMPTargetTriples.begin(), OMPTargetTriples.end()),
+ NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -150,7 +154,7 @@ struct OffloadModuleOpts {
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPForceUSM(Opts.OpenMPForceUSM),
OpenMPVersion(Opts.OpenMPVersion), OMPHostIRFile(Opts.OMPHostIRFile),
- NoGPULib(Opts.NoGPULib) {}
+ OMPTargetTriples(Opts.OMPTargetTriples), NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -162,6 +166,7 @@ struct OffloadModuleOpts {
bool OpenMPForceUSM = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
+ std::vector<llvm::Triple> OMPTargetTriples = {};
bool NoGPULib = false;
};
@@ -185,6 +190,9 @@ struct OffloadModuleOpts {
if (!Opts.OMPHostIRFile.empty())
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
}
+ auto strTriples = llvm::to_vector(llvm::map_range(Opts.OMPTargetTriples,
+ [](llvm::Triple triple) { return triple.normalize(); }));
+ offloadMod.setTargetTriples(strTriples);
}
}
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp
b/flang/lib/Frontend/CompilerInvocation.cpp
index 8c892d9d032e1..19f067a135dd6 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -894,6 +894,7 @@ static bool parseDiagArgs(CompilerInvocation &res,
llvm::opt::ArgList &args,
/// options accordingly. Returns false if new errors are generated.
static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
clang::DiagnosticsEngine &diags) {
+ llvm::Triple t(res.getTargetOpts().triple);
unsigned numErrorsBefore = diags.getNumErrors();
// -fdefault* family
@@ -1012,6 +1013,40 @@ static bool parseDialectArgs(CompilerInvocation &res,
llvm::opt::ArgList &args,
res.getLangOpts().OpenMPIsGPU = 0;
break;
}
+
+ // Get the OpenMP target triples if any.
+ if (auto *arg =
+ args.getLastArg(clang::driver::options::OPT_fopenmp_targets_EQ)) {
+ enum ArchPtrSize { Arch16Bit, Arch32Bit, Arch64Bit };
+ auto getArchPtrSize = [](const llvm::Triple &triple) {
+ if (triple.isArch16Bit())
+ return Arch16Bit;
+ if (triple.isArch32Bit())
+ return Arch32Bit;
+ assert(triple.isArch64Bit() && "Expected 64-bit architecture");
+ return Arch64Bit;
+ };
+
+ for (unsigned i = 0; i < arg->getNumValues(); ++i) {
+ llvm::Triple tt(arg->getValue(i));
+
+ if (tt.getArch() == llvm::Triple::UnknownArch ||
+ !(tt.getArch() == llvm::Triple::aarch64 || tt.isPPC() ||
+ tt.getArch() == llvm::Triple::systemz ||
+ tt.getArch() == llvm::Triple::nvptx ||
+ tt.getArch() == llvm::Triple::nvptx64 ||
+ tt.getArch() == llvm::Triple::amdgcn ||
+ tt.getArch() == llvm::Triple::x86 ||
+ tt.getArch() == llvm::Triple::x86_64))
+ diags.Report(clang::diag::err_drv_invalid_omp_target)
+ << arg->getValue(i);
+ else if (getArchPtrSize(t) != getArchPtrSize(tt))
+ diags.Report(clang::diag::err_drv_incompatible_omp_arch)
+ << arg->getValue(i) << t.str();
+ else
+ res.getLangOpts().OMPTargetTriples.push_back(tt);
+ }
+ }
}
// -pedantic
diff --git a/flang/test/Lower/OpenMP/offload-targets.f90
b/flang/test/Lower/OpenMP/offload-targets.f90
new file mode 100644
index 0000000000000..9d0bb2b2946c5
--- /dev/null
+++ b/flang/test/Lower/OpenMP/offload-targets.f90
@@ -0,0 +1,10 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp
-fopenmp-targets=amdgcn-amd-amdhsa,nvptx64-nvidia-cuda %s -o - | FileCheck %s
+! RUN: bbc -emit-hlfir -fopenmp
-fopenmp-targets=amdgcn-amd-amdhsa,nvptx64-nvidia-cuda %s -o - | FileCheck %s
+
+! This test checks the addition of the omp.target_triples attribute when the
+! -fopenmp-targets option is set
+
+!CHECK: module attributes {
+!CHECK-SAME: omp.target_triples = ["amdgcn-amd-amdhsa", "nvptx64-nvidia-cuda"]
+program targets
+end program targets
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index 07eef065daf6f..736d68219581d 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -149,6 +149,11 @@ static llvm::cl::opt<bool> enableOpenMPForceUSM(
llvm::cl::desc("force openmp unified shared memory mode"),
llvm::cl::init(false));
+static llvm::cl::list<std::string> targetTriplesOpenMP(
+ "fopenmp-targets",
+ llvm::cl::desc("comma-separated list of OpenMP offloading triples"),
+ llvm::cl::CommaSeparated);
+
// A simplified subset of the OpenMP RTL Flags from Flang, only the primary
// positive options are available, no negative options e.g. fopen_assume* vs
// fno_open_assume*
@@ -380,11 +385,17 @@ static llvm::LogicalResult convertFortranSourceToMLIR(
"-fopenmp-is-target-device is also set";
return mlir::failure();
}
+ // Construct offloading target triples vector.
+ std::vector<llvm::Triple> targetTriples;
+ targetTriples.reserve(targetTriplesOpenMP.size());
+ for (llvm::StringRef s : targetTriplesOpenMP)
+ targetTriples.emplace_back(s);
+
auto offloadModuleOpts = OffloadModuleOpts(
setOpenMPTargetDebug, setOpenMPTeamSubscription,
setOpenMPThreadSubscription, setOpenMPNoThreadState,
setOpenMPNoNestedParallelism, enableOpenMPDevice, enableOpenMPGPU,
- enableOpenMPForceUSM, setOpenMPVersion, "", setNoGPULib);
+ enableOpenMPForceUSM, setOpenMPVersion, "", targetTriples,
setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
>From e2fe2f891a317dc65280ef192c5b117f9e1caa5d Mon Sep 17 00:00:00 2001
From: Sergio Afonso <safon...@amd.com>
Date: Wed, 24 Jul 2024 15:17:59 +0100
Subject: [PATCH 2/2] Extract OpenMP argument parsing out of parseDialectArgs
---
flang/lib/Frontend/CompilerInvocation.cpp | 241 +++++++++++-----------
1 file changed, 126 insertions(+), 115 deletions(-)
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp
b/flang/lib/Frontend/CompilerInvocation.cpp
index 19f067a135dd6..378406662ed9a 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -894,7 +894,6 @@ static bool parseDiagArgs(CompilerInvocation &res,
llvm::opt::ArgList &args,
/// options accordingly. Returns false if new errors are generated.
static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
clang::DiagnosticsEngine &diags) {
- llvm::Triple t(res.getTargetOpts().triple);
unsigned numErrorsBefore = diags.getNumErrors();
// -fdefault* family
@@ -930,124 +929,11 @@ static bool parseDialectArgs(CompilerInvocation &res,
llvm::opt::ArgList &args,
Fortran::common::LanguageFeature::CUDA);
}
- // -fopenmp and -fopenacc
+ // -fopenacc
if (args.hasArg(clang::driver::options::OPT_fopenacc)) {
res.getFrontendOpts().features.Enable(
Fortran::common::LanguageFeature::OpenACC);
}
- if (args.hasArg(clang::driver::options::OPT_fopenmp)) {
- // By default OpenMP is set to 1.1 version
- res.getLangOpts().OpenMPVersion = 11;
- res.getFrontendOpts().features.Enable(
- Fortran::common::LanguageFeature::OpenMP);
- if (int Version = getLastArgIntValue(
- args, clang::driver::options::OPT_fopenmp_version_EQ,
- res.getLangOpts().OpenMPVersion, diags)) {
- res.getLangOpts().OpenMPVersion = Version;
- }
- if (args.hasArg(clang::driver::options::OPT_fopenmp_force_usm)) {
- res.getLangOpts().OpenMPForceUSM = 1;
- }
- if (args.hasArg(clang::driver::options::OPT_fopenmp_is_target_device)) {
- res.getLangOpts().OpenMPIsTargetDevice = 1;
-
- // Get OpenMP host file path if any and report if a non existent file is
- // found
- if (auto *arg = args.getLastArg(
- clang::driver::options::OPT_fopenmp_host_ir_file_path)) {
- res.getLangOpts().OMPHostIRFile = arg->getValue();
- if (!llvm::sys::fs::exists(res.getLangOpts().OMPHostIRFile))
- diags.Report(clang::diag::err_drv_omp_host_ir_file_not_found)
- << res.getLangOpts().OMPHostIRFile;
- }
-
- if (args.hasFlag(
-
clang::driver::options::OPT_fopenmp_assume_teams_oversubscription,
- clang::driver::options::
- OPT_fno_openmp_assume_teams_oversubscription,
- /*Default=*/false))
- res.getLangOpts().OpenMPTeamSubscription = true;
-
- if (args.hasArg(
- clang::driver::options::OPT_fopenmp_assume_no_thread_state))
- res.getLangOpts().OpenMPNoThreadState = 1;
-
- if (args.hasArg(
-
clang::driver::options::OPT_fopenmp_assume_no_nested_parallelism))
- res.getLangOpts().OpenMPNoNestedParallelism = 1;
-
- if (args.hasFlag(clang::driver::options::
- OPT_fopenmp_assume_threads_oversubscription,
- clang::driver::options::
- OPT_fno_openmp_assume_threads_oversubscription,
- /*Default=*/false))
- res.getLangOpts().OpenMPThreadSubscription = true;
-
- if ((args.hasArg(clang::driver::options::OPT_fopenmp_target_debug) ||
- args.hasArg(clang::driver::options::OPT_fopenmp_target_debug_EQ))) {
- res.getLangOpts().OpenMPTargetDebug = getLastArgIntValue(
- args, clang::driver::options::OPT_fopenmp_target_debug_EQ,
- res.getLangOpts().OpenMPTargetDebug, diags);
-
- if (!res.getLangOpts().OpenMPTargetDebug &&
- args.hasArg(clang::driver::options::OPT_fopenmp_target_debug))
- res.getLangOpts().OpenMPTargetDebug = 1;
- }
- if (args.hasArg(clang::driver::options::OPT_nogpulib))
- res.getLangOpts().NoGPULib = 1;
- }
-
- switch (llvm::Triple(res.getTargetOpts().triple).getArch()) {
- case llvm::Triple::nvptx:
- case llvm::Triple::nvptx64:
- case llvm::Triple::amdgcn:
- if (!res.getLangOpts().OpenMPIsTargetDevice) {
- const unsigned diagID = diags.getCustomDiagID(
- clang::DiagnosticsEngine::Error,
- "OpenMP AMDGPU/NVPTX is only prepared to deal with device code.");
- diags.Report(diagID);
- }
- res.getLangOpts().OpenMPIsGPU = 1;
- break;
- default:
- res.getLangOpts().OpenMPIsGPU = 0;
- break;
- }
-
- // Get the OpenMP target triples if any.
- if (auto *arg =
- args.getLastArg(clang::driver::options::OPT_fopenmp_targets_EQ)) {
- enum ArchPtrSize { Arch16Bit, Arch32Bit, Arch64Bit };
- auto getArchPtrSize = [](const llvm::Triple &triple) {
- if (triple.isArch16Bit())
- return Arch16Bit;
- if (triple.isArch32Bit())
- return Arch32Bit;
- assert(triple.isArch64Bit() && "Expected 64-bit architecture");
- return Arch64Bit;
- };
-
- for (unsigned i = 0; i < arg->getNumValues(); ++i) {
- llvm::Triple tt(arg->getValue(i));
-
- if (tt.getArch() == llvm::Triple::UnknownArch ||
- !(tt.getArch() == llvm::Triple::aarch64 || tt.isPPC() ||
- tt.getArch() == llvm::Triple::systemz ||
- tt.getArch() == llvm::Triple::nvptx ||
- tt.getArch() == llvm::Triple::nvptx64 ||
- tt.getArch() == llvm::Triple::amdgcn ||
- tt.getArch() == llvm::Triple::x86 ||
- tt.getArch() == llvm::Triple::x86_64))
- diags.Report(clang::diag::err_drv_invalid_omp_target)
- << arg->getValue(i);
- else if (getArchPtrSize(t) != getArchPtrSize(tt))
- diags.Report(clang::diag::err_drv_incompatible_omp_arch)
- << arg->getValue(i) << t.str();
- else
- res.getLangOpts().OMPTargetTriples.push_back(tt);
- }
- }
- }
// -pedantic
if (args.hasArg(clang::driver::options::OPT_pedantic)) {
@@ -1077,6 +963,130 @@ static bool parseDialectArgs(CompilerInvocation &res,
llvm::opt::ArgList &args,
return diags.getNumErrors() == numErrorsBefore;
}
+/// Parses all OpenMP related arguments if the -fopenmp option is present,
+/// populating the \c res object accordingly. Returns false if new errors are
+/// generated.
+static bool parseOpenMPArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
+ clang::DiagnosticsEngine &diags) {
+ if (!args.hasArg(clang::driver::options::OPT_fopenmp))
+ return true;
+
+ unsigned numErrorsBefore = diags.getNumErrors();
+ llvm::Triple t(res.getTargetOpts().triple);
+
+ // By default OpenMP is set to 1.1 version
+ res.getLangOpts().OpenMPVersion = 11;
+ res.getFrontendOpts().features.Enable(
+ Fortran::common::LanguageFeature::OpenMP);
+ if (int Version = getLastArgIntValue(
+ args, clang::driver::options::OPT_fopenmp_version_EQ,
+ res.getLangOpts().OpenMPVersion, diags)) {
+ res.getLangOpts().OpenMPVersion = Version;
+ }
+ if (args.hasArg(clang::driver::options::OPT_fopenmp_force_usm)) {
+ res.getLangOpts().OpenMPForceUSM = 1;
+ }
+ if (args.hasArg(clang::driver::options::OPT_fopenmp_is_target_device)) {
+ res.getLangOpts().OpenMPIsTargetDevice = 1;
+
+ // Get OpenMP host file path if any and report if a non existent file is
+ // found
+ if (auto *arg = args.getLastArg(
+ clang::driver::options::OPT_fopenmp_host_ir_file_path)) {
+ res.getLangOpts().OMPHostIRFile = arg->getValue();
+ if (!llvm::sys::fs::exists(res.getLangOpts().OMPHostIRFile))
+ diags.Report(clang::diag::err_drv_omp_host_ir_file_not_found)
+ << res.getLangOpts().OMPHostIRFile;
+ }
+
+ if (args.hasFlag(
+ clang::driver::options::OPT_fopenmp_assume_teams_oversubscription,
+ clang::driver::options::
+ OPT_fno_openmp_assume_teams_oversubscription,
+ /*Default=*/false))
+ res.getLangOpts().OpenMPTeamSubscription = true;
+
+ if
(args.hasArg(clang::driver::options::OPT_fopenmp_assume_no_thread_state))
+ res.getLangOpts().OpenMPNoThreadState = 1;
+
+ if (args.hasArg(
+ clang::driver::options::OPT_fopenmp_assume_no_nested_parallelism))
+ res.getLangOpts().OpenMPNoNestedParallelism = 1;
+
+ if (args.hasFlag(
+
clang::driver::options::OPT_fopenmp_assume_threads_oversubscription,
+ clang::driver::options::
+ OPT_fno_openmp_assume_threads_oversubscription,
+ /*Default=*/false))
+ res.getLangOpts().OpenMPThreadSubscription = true;
+
+ if ((args.hasArg(clang::driver::options::OPT_fopenmp_target_debug) ||
+ args.hasArg(clang::driver::options::OPT_fopenmp_target_debug_EQ))) {
+ res.getLangOpts().OpenMPTargetDebug = getLastArgIntValue(
+ args, clang::driver::options::OPT_fopenmp_target_debug_EQ,
+ res.getLangOpts().OpenMPTargetDebug, diags);
+
+ if (!res.getLangOpts().OpenMPTargetDebug &&
+ args.hasArg(clang::driver::options::OPT_fopenmp_target_debug))
+ res.getLangOpts().OpenMPTargetDebug = 1;
+ }
+ if (args.hasArg(clang::driver::options::OPT_nogpulib))
+ res.getLangOpts().NoGPULib = 1;
+ }
+
+ switch (llvm::Triple(res.getTargetOpts().triple).getArch()) {
+ case llvm::Triple::nvptx:
+ case llvm::Triple::nvptx64:
+ case llvm::Triple::amdgcn:
+ if (!res.getLangOpts().OpenMPIsTargetDevice) {
+ const unsigned diagID = diags.getCustomDiagID(
+ clang::DiagnosticsEngine::Error,
+ "OpenMP AMDGPU/NVPTX is only prepared to deal with device code.");
+ diags.Report(diagID);
+ }
+ res.getLangOpts().OpenMPIsGPU = 1;
+ break;
+ default:
+ res.getLangOpts().OpenMPIsGPU = 0;
+ break;
+ }
+
+ // Get the OpenMP target triples if any.
+ if (auto *arg =
+ args.getLastArg(clang::driver::options::OPT_fopenmp_targets_EQ)) {
+ enum ArchPtrSize { Arch16Bit, Arch32Bit, Arch64Bit };
+ auto getArchPtrSize = [](const llvm::Triple &triple) {
+ if (triple.isArch16Bit())
+ return Arch16Bit;
+ if (triple.isArch32Bit())
+ return Arch32Bit;
+ assert(triple.isArch64Bit() && "Expected 64-bit architecture");
+ return Arch64Bit;
+ };
+
+ for (unsigned i = 0; i < arg->getNumValues(); ++i) {
+ llvm::Triple tt(arg->getValue(i));
+
+ if (tt.getArch() == llvm::Triple::UnknownArch ||
+ !(tt.getArch() == llvm::Triple::aarch64 || tt.isPPC() ||
+ tt.getArch() == llvm::Triple::systemz ||
+ tt.getArch() == llvm::Triple::nvptx ||
+ tt.getArch() == llvm::Triple::nvptx64 ||
+ tt.getArch() == llvm::Triple::amdgcn ||
+ tt.getArch() == llvm::Triple::x86 ||
+ tt.getArch() == llvm::Triple::x86_64))
+ diags.Report(clang::diag::err_drv_invalid_omp_target)
+ << arg->getValue(i);
+ else if (getArchPtrSize(t) != getArchPtrSize(tt))
+ diags.Report(clang::diag::err_drv_incompatible_omp_arch)
+ << arg->getValue(i) << t.str();
+ else
+ res.getLangOpts().OMPTargetTriples.push_back(tt);
+ }
+ }
+ return diags.getNumErrors() == numErrorsBefore;
+}
+
/// Parses all floating point related arguments and populates the
/// CompilerInvocation accordingly.
/// Returns false if new errors are generated.
@@ -1312,6 +1322,7 @@ bool CompilerInvocation::createFromArgs(
success &= parseVectorLibArg(invoc.getCodeGenOpts(), args, diags);
success &= parseSemaArgs(invoc, args, diags);
success &= parseDialectArgs(invoc, args, diags);
+ success &= parseOpenMPArgs(invoc, args, diags);
success &= parseDiagArgs(invoc, args, diags);
// Collect LLVM (-mllvm) and MLIR (-mmlir) options.
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
