[Flang][OpenMP] Add -fopenmp-force-usm option to flang #94359
Merged
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
skatrak
requested review from
ergawy,
Meinersbur,
jsjodin,
banach-space,
tblah,
kiranchandramohan,
abidh,
mjklemm,
raghavendhra,
agozillon,
DominikAdamski,
TIFitis,
kparzysz and
bhandarkar-pranav
llvmbot
added
clang
|
@llvm/pr-subscribers-flang-fir-hlfir @llvm/pr-subscribers-clang-driver Author: Sergio Afonso (skatrak) ChangesThis patch enables the This is later combined with any other target device-related REQUIRES clauses that may have been explicitly set in the compilation unit. Full diff: https://github.com/llvm/llvm-project/pull/94359.diff 10 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 57f37c5023110..eefec33e6d333 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -3574,7 +3574,7 @@ def fopenmp_offload_mandatory : Flag<["-"], "fopenmp-offload-mandatory">, Group<
HelpText<"Do not create a host fallback if offloading to the device fails.">,
MarshallingInfoFlag<LangOpts<"OpenMPOffloadMandatory">>;
def fopenmp_force_usm : Flag<["-"], "fopenmp-force-usm">, Group<f_Group>,
- Flags<[NoArgumentUnused]>, Visibility<[ClangOption, CC1Option]>,
+ Flags<[NoArgumentUnused]>, Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Force behvaior as if the user specified pragma omp requires unified_shared_memory.">,
MarshallingInfoFlag<LangOpts<"OpenMPForceUSM">>;
def fopenmp_target_jit : Flag<["-"], "fopenmp-target-jit">, Group<f_Group>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 42ca060186fd8..b7abe7b1c19bc 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -766,6 +766,8 @@ void Flang::ConstructJob(Compilation &C, const JobAction &JA,
Args.AddAllArgs(CmdArgs, options::OPT_fopenmp_version_EQ);
// FIXME: Clang supports a whole bunch more flags here.
+ if (Args.hasArg(options::OPT_fopenmp_force_usm))
+ CmdArgs.push_back("-fopenmp-force-usm");
break;
default:
// By default, if Clang doesn't know how to generate useful OpenMP code
diff --git a/flang/include/flang/Frontend/LangOptions.def b/flang/include/flang/Frontend/LangOptions.def
index 2bf10826120a8..d3e1e972d1519 100644
--- a/flang/include/flang/Frontend/LangOptions.def
+++ b/flang/include/flang/Frontend/LangOptions.def
@@ -42,6 +42,8 @@ LANGOPT(OpenMPVersion, 32, 0)
LANGOPT(OpenMPIsTargetDevice, 1, false)
/// Generate OpenMP target code only for GPUs
LANGOPT(OpenMPIsGPU, 1, false)
+/// Generate OpenMP target code only for GPUs
+LANGOPT(OpenMPForceUSM, 1, false)
/// Enable debugging in the OpenMP offloading device RTL
LANGOPT(OpenMPTargetDebug, 32, 0)
/// Assume work-shared loops do not have more iterations than participating
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 77b68fc6187fa..26fbe51d329c7 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -130,16 +130,16 @@ struct OffloadModuleOpts {
OffloadModuleOpts(uint32_t OpenMPTargetDebug, bool OpenMPTeamSubscription,
bool OpenMPThreadSubscription, bool OpenMPNoThreadState,
bool OpenMPNoNestedParallelism, bool OpenMPIsTargetDevice,
- bool OpenMPIsGPU, uint32_t OpenMPVersion, std::string OMPHostIRFile = {},
- bool NoGPULib = false)
+ bool OpenMPIsGPU, bool OpenMPForceUSM, uint32_t OpenMPVersion,
+ std::string OMPHostIRFile = {}, bool NoGPULib = false)
: OpenMPTargetDebug(OpenMPTargetDebug),
OpenMPTeamSubscription(OpenMPTeamSubscription),
OpenMPThreadSubscription(OpenMPThreadSubscription),
OpenMPNoThreadState(OpenMPNoThreadState),
OpenMPNoNestedParallelism(OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(OpenMPIsTargetDevice), OpenMPIsGPU(OpenMPIsGPU),
- OpenMPVersion(OpenMPVersion), OMPHostIRFile(OMPHostIRFile),
- NoGPULib(NoGPULib) {}
+ OpenMPForceUSM(OpenMPForceUSM), OpenMPVersion(OpenMPVersion),
+ OMPHostIRFile(OMPHostIRFile), NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -148,8 +148,9 @@ struct OffloadModuleOpts {
OpenMPNoThreadState(Opts.OpenMPNoThreadState),
OpenMPNoNestedParallelism(Opts.OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
- OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPVersion(Opts.OpenMPVersion),
- OMPHostIRFile(Opts.OMPHostIRFile), NoGPULib(Opts.NoGPULib) {}
+ OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPForceUSM(Opts.OpenMPForceUSM),
+ OpenMPVersion(Opts.OpenMPVersion), OMPHostIRFile(Opts.OMPHostIRFile),
+ NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -158,6 +159,7 @@ struct OffloadModuleOpts {
bool OpenMPNoNestedParallelism = false;
bool OpenMPIsTargetDevice = false;
bool OpenMPIsGPU = false;
+ bool OpenMPForceUSM = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
bool NoGPULib = false;
@@ -172,13 +174,17 @@ struct OffloadModuleOpts {
module.getOperation())) {
offloadMod.setIsTargetDevice(Opts.OpenMPIsTargetDevice);
offloadMod.setIsGPU(Opts.OpenMPIsGPU);
+ if (Opts.OpenMPForceUSM) {
+ offloadMod.setRequires(mlir::omp::ClauseRequires::unified_shared_memory);
+ }
if (Opts.OpenMPIsTargetDevice) {
offloadMod.setFlags(Opts.OpenMPTargetDebug, Opts.OpenMPTeamSubscription,
Opts.OpenMPThreadSubscription, Opts.OpenMPNoThreadState,
Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion, Opts.NoGPULib);
- if (!Opts.OMPHostIRFile.empty())
+ if (!Opts.OMPHostIRFile.empty()) {
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
+ }
}
}
}
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index 50c3e8b0113b5..f64a939b785ef 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -906,6 +906,9 @@ static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
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;
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 9598457d123cf..af9e2af24619b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -2608,7 +2608,9 @@ void Fortran::lower::genOpenMPRequires(mlir::Operation *mod,
symbol->details());
}
- MlirRequires mlirFlags = MlirRequires::none;
+ // Use pre-populated omp.requires module attribute if it was set, so that
+ // the "-fopenmp-force-usm" compiler option is honored.
+ MlirRequires mlirFlags = offloadMod.getRequires();
if (semaFlags.test(SemaRequires::ReverseOffload))
mlirFlags = mlirFlags | MlirRequires::reverse_offload;
if (semaFlags.test(SemaRequires::UnifiedAddress))
diff --git a/flang/test/Driver/omp-driver-offload.f90 b/flang/test/Driver/omp-driver-offload.f90
index 8f48ca75114ce..6fb4f4eeeeca1 100644
--- a/flang/test/Driver/omp-driver-offload.f90
+++ b/flang/test/Driver/omp-driver-offload.f90
@@ -207,3 +207,23 @@
! RUN: --rocm-path=%S/Inputs/rocm %s 2>&1 \
! RUN: | FileCheck --check-prefix=ROCM-PATH %s
! ROCM-PATH: Found HIP installation: {{.*Inputs.*rocm}}, version 3.6.20214-a2917cd
+
+! Test -fopenmp-force-usm option without offload
+! RUN: %flang -S -### %s -o %t 2>&1 \
+! RUN: -fopenmp -fopenmp-force-usm \
+! RUN: --target=aarch64-unknown-linux-gnu \
+! RUN: | FileCheck %s --check-prefix=FORCE-USM-NO-OFFLOAD
+
+! FORCE-USM-NO-OFFLOAD: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
+! FORCE-USM-NO-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
+
+! Test -fopenmp-force-usm option with offload
+! RUN: %flang -S -### %s -o %t 2>&1 \
+! RUN: -fopenmp -fopenmp-force-usm --offload-arch=gfx90a \
+! RUN: --target=aarch64-unknown-linux-gnu \
+! RUN: | FileCheck %s --check-prefix=FORCE-USM-OFFLOAD
+
+! FORCE-USM-OFFLOAD: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
+! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
+! FORCE-USM-OFFLOAD-NEXT: "{{[^"]*}}flang-new" "-fc1" "-triple" "amdgcn-amd-amdhsa"
+! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
diff --git a/flang/test/Lower/OpenMP/force-usm.f90 b/flang/test/Lower/OpenMP/force-usm.f90
new file mode 100644
index 0000000000000..90bbf3c4d842f
--- /dev/null
+++ b/flang/test/Lower/OpenMP/force-usm.f90
@@ -0,0 +1,12 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-is-target-device -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+
+! This test checks the addition of requires unified_shared_memory when
+! -fopenmp-force-usm is set
+
+!CHECK: module attributes {
+!CHECK-SAME: omp.requires = #omp<clause_requires unified_shared_memory>
+program requires
+end program requires
diff --git a/flang/test/Lower/OpenMP/requires-force-usm.f90 b/flang/test/Lower/OpenMP/requires-force-usm.f90
new file mode 100644
index 0000000000000..5f5cf9e64cd70
--- /dev/null
+++ b/flang/test/Lower/OpenMP/requires-force-usm.f90
@@ -0,0 +1,15 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-is-target-device -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+
+! This test checks the addition of requires unified_shared_memory when
+! -fopenmp-force-usm is set, even when other requires directives are present
+
+!CHECK: module attributes {
+!CHECK-SAME: omp.requires = #omp<clause_requires reverse_offload|unified_shared_memory>
+program requires
+ !$omp requires reverse_offload
+ !$omp target
+ !$omp end target
+end program requires
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index bab21338cef26..3485c1499d3b6 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -144,6 +144,11 @@ static llvm::cl::opt<bool>
llvm::cl::desc("enable openmp GPU target codegen"),
llvm::cl::init(false));
+static llvm::cl::opt<bool> enableOpenMPForceUSM(
+ "fopenmp-force-usm",
+ llvm::cl::desc("force openmp unified shared memory mode"),
+ llvm::cl::init(false));
+
// 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*
@@ -374,11 +379,11 @@ static mlir::LogicalResult convertFortranSourceToMLIR(
"-fopenmp-is-target-device is also set";
return mlir::failure();
}
- auto offloadModuleOpts =
- OffloadModuleOpts(setOpenMPTargetDebug, setOpenMPTeamSubscription,
- setOpenMPThreadSubscription, setOpenMPNoThreadState,
- setOpenMPNoNestedParallelism, enableOpenMPDevice,
- enableOpenMPGPU, setOpenMPVersion, "", setNoGPULib);
+ auto offloadModuleOpts = OffloadModuleOpts(
+ setOpenMPTargetDebug, setOpenMPTeamSubscription,
+ setOpenMPThreadSubscription, setOpenMPNoThreadState,
+ setOpenMPNoNestedParallelism, enableOpenMPDevice, enableOpenMPGPU,
+ enableOpenMPForceUSM, setOpenMPVersion, "", setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
|
|
@llvm/pr-subscribers-clang Author: Sergio Afonso (skatrak) ChangesThis patch enables the This is later combined with any other target device-related REQUIRES clauses that may have been explicitly set in the compilation unit. Full diff: https://github.com/llvm/llvm-project/pull/94359.diff 10 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 57f37c5023110..eefec33e6d333 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -3574,7 +3574,7 @@ def fopenmp_offload_mandatory : Flag<["-"], "fopenmp-offload-mandatory">, Group<
HelpText<"Do not create a host fallback if offloading to the device fails.">,
MarshallingInfoFlag<LangOpts<"OpenMPOffloadMandatory">>;
def fopenmp_force_usm : Flag<["-"], "fopenmp-force-usm">, Group<f_Group>,
- Flags<[NoArgumentUnused]>, Visibility<[ClangOption, CC1Option]>,
+ Flags<[NoArgumentUnused]>, Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Force behvaior as if the user specified pragma omp requires unified_shared_memory.">,
MarshallingInfoFlag<LangOpts<"OpenMPForceUSM">>;
def fopenmp_target_jit : Flag<["-"], "fopenmp-target-jit">, Group<f_Group>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 42ca060186fd8..b7abe7b1c19bc 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -766,6 +766,8 @@ void Flang::ConstructJob(Compilation &C, const JobAction &JA,
Args.AddAllArgs(CmdArgs, options::OPT_fopenmp_version_EQ);
// FIXME: Clang supports a whole bunch more flags here.
+ if (Args.hasArg(options::OPT_fopenmp_force_usm))
+ CmdArgs.push_back("-fopenmp-force-usm");
break;
default:
// By default, if Clang doesn't know how to generate useful OpenMP code
diff --git a/flang/include/flang/Frontend/LangOptions.def b/flang/include/flang/Frontend/LangOptions.def
index 2bf10826120a8..d3e1e972d1519 100644
--- a/flang/include/flang/Frontend/LangOptions.def
+++ b/flang/include/flang/Frontend/LangOptions.def
@@ -42,6 +42,8 @@ LANGOPT(OpenMPVersion, 32, 0)
LANGOPT(OpenMPIsTargetDevice, 1, false)
/// Generate OpenMP target code only for GPUs
LANGOPT(OpenMPIsGPU, 1, false)
+/// Generate OpenMP target code only for GPUs
+LANGOPT(OpenMPForceUSM, 1, false)
/// Enable debugging in the OpenMP offloading device RTL
LANGOPT(OpenMPTargetDebug, 32, 0)
/// Assume work-shared loops do not have more iterations than participating
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index 77b68fc6187fa..26fbe51d329c7 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -130,16 +130,16 @@ struct OffloadModuleOpts {
OffloadModuleOpts(uint32_t OpenMPTargetDebug, bool OpenMPTeamSubscription,
bool OpenMPThreadSubscription, bool OpenMPNoThreadState,
bool OpenMPNoNestedParallelism, bool OpenMPIsTargetDevice,
- bool OpenMPIsGPU, uint32_t OpenMPVersion, std::string OMPHostIRFile = {},
- bool NoGPULib = false)
+ bool OpenMPIsGPU, bool OpenMPForceUSM, uint32_t OpenMPVersion,
+ std::string OMPHostIRFile = {}, bool NoGPULib = false)
: OpenMPTargetDebug(OpenMPTargetDebug),
OpenMPTeamSubscription(OpenMPTeamSubscription),
OpenMPThreadSubscription(OpenMPThreadSubscription),
OpenMPNoThreadState(OpenMPNoThreadState),
OpenMPNoNestedParallelism(OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(OpenMPIsTargetDevice), OpenMPIsGPU(OpenMPIsGPU),
- OpenMPVersion(OpenMPVersion), OMPHostIRFile(OMPHostIRFile),
- NoGPULib(NoGPULib) {}
+ OpenMPForceUSM(OpenMPForceUSM), OpenMPVersion(OpenMPVersion),
+ OMPHostIRFile(OMPHostIRFile), NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -148,8 +148,9 @@ struct OffloadModuleOpts {
OpenMPNoThreadState(Opts.OpenMPNoThreadState),
OpenMPNoNestedParallelism(Opts.OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
- OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPVersion(Opts.OpenMPVersion),
- OMPHostIRFile(Opts.OMPHostIRFile), NoGPULib(Opts.NoGPULib) {}
+ OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPForceUSM(Opts.OpenMPForceUSM),
+ OpenMPVersion(Opts.OpenMPVersion), OMPHostIRFile(Opts.OMPHostIRFile),
+ NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -158,6 +159,7 @@ struct OffloadModuleOpts {
bool OpenMPNoNestedParallelism = false;
bool OpenMPIsTargetDevice = false;
bool OpenMPIsGPU = false;
+ bool OpenMPForceUSM = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
bool NoGPULib = false;
@@ -172,13 +174,17 @@ struct OffloadModuleOpts {
module.getOperation())) {
offloadMod.setIsTargetDevice(Opts.OpenMPIsTargetDevice);
offloadMod.setIsGPU(Opts.OpenMPIsGPU);
+ if (Opts.OpenMPForceUSM) {
+ offloadMod.setRequires(mlir::omp::ClauseRequires::unified_shared_memory);
+ }
if (Opts.OpenMPIsTargetDevice) {
offloadMod.setFlags(Opts.OpenMPTargetDebug, Opts.OpenMPTeamSubscription,
Opts.OpenMPThreadSubscription, Opts.OpenMPNoThreadState,
Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion, Opts.NoGPULib);
- if (!Opts.OMPHostIRFile.empty())
+ if (!Opts.OMPHostIRFile.empty()) {
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
+ }
}
}
}
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index 50c3e8b0113b5..f64a939b785ef 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -906,6 +906,9 @@ static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
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;
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 9598457d123cf..af9e2af24619b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -2608,7 +2608,9 @@ void Fortran::lower::genOpenMPRequires(mlir::Operation *mod,
symbol->details());
}
- MlirRequires mlirFlags = MlirRequires::none;
+ // Use pre-populated omp.requires module attribute if it was set, so that
+ // the "-fopenmp-force-usm" compiler option is honored.
+ MlirRequires mlirFlags = offloadMod.getRequires();
if (semaFlags.test(SemaRequires::ReverseOffload))
mlirFlags = mlirFlags | MlirRequires::reverse_offload;
if (semaFlags.test(SemaRequires::UnifiedAddress))
diff --git a/flang/test/Driver/omp-driver-offload.f90 b/flang/test/Driver/omp-driver-offload.f90
index 8f48ca75114ce..6fb4f4eeeeca1 100644
--- a/flang/test/Driver/omp-driver-offload.f90
+++ b/flang/test/Driver/omp-driver-offload.f90
@@ -207,3 +207,23 @@
! RUN: --rocm-path=%S/Inputs/rocm %s 2>&1 \
! RUN: | FileCheck --check-prefix=ROCM-PATH %s
! ROCM-PATH: Found HIP installation: {{.*Inputs.*rocm}}, version 3.6.20214-a2917cd
+
+! Test -fopenmp-force-usm option without offload
+! RUN: %flang -S -### %s -o %t 2>&1 \
+! RUN: -fopenmp -fopenmp-force-usm \
+! RUN: --target=aarch64-unknown-linux-gnu \
+! RUN: | FileCheck %s --check-prefix=FORCE-USM-NO-OFFLOAD
+
+! FORCE-USM-NO-OFFLOAD: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
+! FORCE-USM-NO-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
+
+! Test -fopenmp-force-usm option with offload
+! RUN: %flang -S -### %s -o %t 2>&1 \
+! RUN: -fopenmp -fopenmp-force-usm --offload-arch=gfx90a \
+! RUN: --target=aarch64-unknown-linux-gnu \
+! RUN: | FileCheck %s --check-prefix=FORCE-USM-OFFLOAD
+
+! FORCE-USM-OFFLOAD: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
+! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
+! FORCE-USM-OFFLOAD-NEXT: "{{[^"]*}}flang-new" "-fc1" "-triple" "amdgcn-amd-amdhsa"
+! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
diff --git a/flang/test/Lower/OpenMP/force-usm.f90 b/flang/test/Lower/OpenMP/force-usm.f90
new file mode 100644
index 0000000000000..90bbf3c4d842f
--- /dev/null
+++ b/flang/test/Lower/OpenMP/force-usm.f90
@@ -0,0 +1,12 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-is-target-device -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+
+! This test checks the addition of requires unified_shared_memory when
+! -fopenmp-force-usm is set
+
+!CHECK: module attributes {
+!CHECK-SAME: omp.requires = #omp<clause_requires unified_shared_memory>
+program requires
+end program requires
diff --git a/flang/test/Lower/OpenMP/requires-force-usm.f90 b/flang/test/Lower/OpenMP/requires-force-usm.f90
new file mode 100644
index 0000000000000..5f5cf9e64cd70
--- /dev/null
+++ b/flang/test/Lower/OpenMP/requires-force-usm.f90
@@ -0,0 +1,15 @@
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-is-target-device -fopenmp-force-usm %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+! RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-force-usm -emit-hlfir %s -o - | FileCheck %s
+
+! This test checks the addition of requires unified_shared_memory when
+! -fopenmp-force-usm is set, even when other requires directives are present
+
+!CHECK: module attributes {
+!CHECK-SAME: omp.requires = #omp<clause_requires reverse_offload|unified_shared_memory>
+program requires
+ !$omp requires reverse_offload
+ !$omp target
+ !$omp end target
+end program requires
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index bab21338cef26..3485c1499d3b6 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -144,6 +144,11 @@ static llvm::cl::opt<bool>
llvm::cl::desc("enable openmp GPU target codegen"),
llvm::cl::init(false));
+static llvm::cl::opt<bool> enableOpenMPForceUSM(
+ "fopenmp-force-usm",
+ llvm::cl::desc("force openmp unified shared memory mode"),
+ llvm::cl::init(false));
+
// 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*
@@ -374,11 +379,11 @@ static mlir::LogicalResult convertFortranSourceToMLIR(
"-fopenmp-is-target-device is also set";
return mlir::failure();
}
- auto offloadModuleOpts =
- OffloadModuleOpts(setOpenMPTargetDebug, setOpenMPTeamSubscription,
- setOpenMPThreadSubscription, setOpenMPNoThreadState,
- setOpenMPNoNestedParallelism, enableOpenMPDevice,
- enableOpenMPGPU, setOpenMPVersion, "", setNoGPULib);
+ auto offloadModuleOpts = OffloadModuleOpts(
+ setOpenMPTargetDebug, setOpenMPTeamSubscription,
+ setOpenMPThreadSubscription, setOpenMPNoThreadState,
+ setOpenMPNoNestedParallelism, enableOpenMPDevice, enableOpenMPGPU,
+ enableOpenMPForceUSM, setOpenMPVersion, "", setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
|
banach-space
approved these changes
Jun 4, 2024
This patch enables the `-fopenmp-force-usm` option to be passed to the flang driver, which forwards it to the compiler frontend. This flag, when set, results in the introduction of the `unified_shared_memory` bit to the `omp.requires` attribute of the top-level module operation. This is later combined with any other target device-related REQUIRES clauses that may have been explicitly set in the compilation unit.
|
Thank your for the reviews. Merging now since buildbot failure is unrelated. |
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.
This patch enables the
-fopenmp-force-usmoption to be passed to the flang driver, which forwards it to the compiler frontend. This flag, when set, results in the introduction of theunified_shared_memorybit to theomp.requiresattribute of the top-level module operation.This is later combined with any other target device-related REQUIRES clauses that may have been explicitly set in the compilation unit.