[Flang][OpenMP][MLIR] Add support for -nogpulib option #71045
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
Added an OpenMP attribute which will model nogpulib flag. If nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314
If nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314
DominikAdamski
requested review from
skatrak,
jsjodin,
banach-space,
kiranchandramohan,
agozillon,
dpalermo and
kiranktp
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-mlir Author: Dominik Adamski (DominikAdamski) ChangesIf -nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314 Full diff: https://github.com/llvm/llvm-project/pull/71045.diff 14 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index b1229b2f4562379..cc50b45b7933d5f 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5092,7 +5092,7 @@ def nohipwrapperinc : Flag<["-"], "nohipwrapperinc">, Group<IncludePath_Group>,
HelpText<"Do not include the default HIP wrapper headers and include paths">;
def : Flag<["-"], "nocudainc">, Alias<nogpuinc>;
def nogpulib : Flag<["-"], "nogpulib">, MarshallingInfoFlag<LangOpts<"NoGPULib">>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Do not link device library for CUDA/HIP device compilation">;
def : Flag<["-"], "nocudalib">, Alias<nogpulib>;
def gpulibc : Flag<["-"], "gpulibc">, Visibility<[ClangOption, CC1Option]>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 999039f83ddfb92..17d9ac7632128d6 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -295,6 +295,8 @@ void Flang::addOffloadOptions(Compilation &C, const InputInfoList &Inputs,
CmdArgs.push_back("-fopenmp-assume-no-thread-state");
if (Args.hasArg(options::OPT_fopenmp_assume_no_nested_parallelism))
CmdArgs.push_back("-fopenmp-assume-no-nested-parallelism");
+ if (Args.hasArg(options::OPT_nogpulib))
+ CmdArgs.push_back("-nogpulib");
}
}
diff --git a/flang/include/flang/Frontend/LangOptions.def b/flang/include/flang/Frontend/LangOptions.def
index 3a1d44f7fb472d1..2bf10826120a8bd 100644
--- a/flang/include/flang/Frontend/LangOptions.def
+++ b/flang/include/flang/Frontend/LangOptions.def
@@ -21,6 +21,8 @@ LANGOPT(Name, Bits, Default)
ENUM_LANGOPT(FPContractMode, FPModeKind, 2, FPM_Fast) ///< FP Contract Mode (off/fast)
+/// Indicate a build without the standard GPU libraries.
+LANGOPT(NoGPULib , 1, false)
/// Permit floating point optimization without regard to infinities
LANGOPT(NoHonorInfs, 1, false)
/// Permit floating point optimization without regard to NaN
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index ddec70fa9824c52..3b51a3845625df3 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -53,14 +53,16 @@ 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 OpenMPIsGPU, 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) {}
+ OpenMPVersion(OpenMPVersion), OMPHostIRFile(OMPHostIRFile),
+ NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -70,7 +72,7 @@ struct OffloadModuleOpts {
OpenMPNoNestedParallelism(Opts.OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPVersion(Opts.OpenMPVersion),
- OMPHostIRFile(Opts.OMPHostIRFile) {}
+ OMPHostIRFile(Opts.OMPHostIRFile), NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -81,6 +83,7 @@ struct OffloadModuleOpts {
bool OpenMPIsGPU = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
+ bool NoGPULib = false;
};
// Shares assinging of the OpenMP OffloadModuleInterface and its assorted
@@ -95,7 +98,7 @@ void setOffloadModuleInterfaceAttributes(
if (Opts.OpenMPIsTargetDevice) {
offloadMod.setFlags(Opts.OpenMPTargetDebug, Opts.OpenMPTeamSubscription,
Opts.OpenMPThreadSubscription, Opts.OpenMPNoThreadState,
- Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion);
+ Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion, Opts.NoGPULib);
if (!Opts.OMPHostIRFile.empty())
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2ecab3742587a..ef0db22037283d3 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -881,6 +881,8 @@ static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
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()) {
diff --git a/flang/test/Driver/driver-help-hidden.f90 b/flang/test/Driver/driver-help-hidden.f90
index 6d399f1d179a022..2e65096d0c3fec4 100644
--- a/flang/test/Driver/driver-help-hidden.f90
+++ b/flang/test/Driver/driver-help-hidden.f90
@@ -119,6 +119,7 @@
! CHECK-NEXT: --no-offload-arch=<value>
! CHECK-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! CHECK-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! CHECK-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! CHECK-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! CHECK-NEXT: --offload-device-only Only compile for the offloading device.
! CHECK-NEXT: --offload-host-device Compile for both the offloading host and device (default).
diff --git a/flang/test/Driver/driver-help.f90 b/flang/test/Driver/driver-help.f90
index 31c9caa32ea8292..a29e8a0b69db46d 100644
--- a/flang/test/Driver/driver-help.f90
+++ b/flang/test/Driver/driver-help.f90
@@ -107,6 +107,7 @@
! HELP-NEXT: --no-offload-arch=<value>
! HELP-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! HELP-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! HELP-NEXT: --offload-device-only Only compile for the offloading device.
! HELP-NEXT: --offload-host-device Compile for both the offloading host and device (default).
@@ -239,6 +240,7 @@
! HELP-FC1-NEXT: -mvscale-max=<value> Specify the vscale maximum. Defaults to the vector length agnostic value of "0". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -mvscale-min=<value> Specify the vscale minimum. Defaults to "1". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-FC1-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-FC1-NEXT: -opt-record-file <value>
! HELP-FC1-NEXT: File name to use for YAML optimization record output
! HELP-FC1-NEXT: -opt-record-format <value>
diff --git a/flang/test/Lower/OpenMP/nogpulib.f90 b/flang/test/Lower/OpenMP/nogpulib.f90
new file mode 100644
index 000000000000000..f2e67136ecd7481
--- /dev/null
+++ b/flang/test/Lower/OpenMP/nogpulib.f90
@@ -0,0 +1,12 @@
+!REQUIRES: amdgpu-registered-target
+
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device %s -o - | FileCheck %s
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -o - %s | FileCheck %s
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device -nogpulib %s -o - | FileCheck %s -check-prefix=FLAG_SET
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -nogpulib -o - %s | FileCheck %s -check-prefix=FLAG_SET
+
+!CHECK-NOT: module attributes {{{.*}}no_gpu_lib
+!FLAG_SET: module attributes {{{.*}}no_gpu_lib = true
+subroutine omp_subroutine()
+end subroutine omp_subroutine
+
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index 378216bd1e51f8f..a66c41eed59651c 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -177,6 +177,12 @@ static llvm::cl::opt<bool> setOpenMPNoNestedParallelism(
"a parallel region."),
llvm::cl::init(false));
+static llvm::cl::opt<bool>
+ setNoGPULib("nogpulib",
+ llvm::cl::desc("Do not link device library for CUDA/HIP device "
+ "compilation"),
+ llvm::cl::init(false));
+
static llvm::cl::opt<bool> enableOpenACC("fopenacc",
llvm::cl::desc("enable openacc"),
llvm::cl::init(false));
@@ -316,7 +322,7 @@ static mlir::LogicalResult convertFortranSourceToMLIR(
OffloadModuleOpts(setOpenMPTargetDebug, setOpenMPTeamSubscription,
setOpenMPThreadSubscription, setOpenMPNoThreadState,
setOpenMPNoNestedParallelism, enableOpenMPDevice,
- enableOpenMPGPU, setOpenMPVersion);
+ enableOpenMPGPU, setOpenMPVersion, "", setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 72121ad4f9e47a5..ae2dd018059e3aa 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -65,6 +65,7 @@ def FlagsAttr : OpenMP_Attr<"Flags", "flags"> {
DefaultValuedParameter<"bool", "false">:$assume_threads_oversubscription,
DefaultValuedParameter<"bool", "false">:$assume_no_thread_state,
DefaultValuedParameter<"bool", "false">:$assume_no_nested_parallelism,
+ DefaultValuedParameter<"bool", "false">:$no_gpu_lib,
DefaultValuedParameter<"uint32_t", "50">:$openmp_device_version
);
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
index 77001fc816cf91f..89d04af64766fc2 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
@@ -198,11 +198,12 @@ def OffloadModuleInterface : OpInterface<"OffloadModuleInterface"> {
"bool":$assumeThreadsOversubscription,
"bool":$assumeNoThreadState,
"bool":$assumeNoNestedParallelism,
- "uint32_t":$openmpDeviceVersion), [{}], [{
+ "uint32_t":$openmpDeviceVersion,
+ "bool":$noGPULib), [{}], [{
$_op->setAttr(("omp." + mlir::omp::FlagsAttr::getMnemonic()).str(),
mlir::omp::FlagsAttr::get($_op->getContext(), debugKind,
assumeTeamsOversubscription, assumeThreadsOversubscription,
- assumeNoThreadState, assumeNoNestedParallelism, openmpDeviceVersion));
+ assumeNoThreadState, assumeNoNestedParallelism, noGPULib, openmpDeviceVersion));
}]>,
InterfaceMethod<
/*description=*/[{
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 1daf60b8659bb66..083b507f405f426 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -2011,6 +2011,12 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
llvm::OpenMPIRBuilder *ompBuilder = moduleTranslation.getOpenMPBuilder();
+ ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
+ attribute.getOpenmpDeviceVersion());
+
+ if (attribute.getNoGpuLib())
+ return success();
+
ompBuilder->createGlobalFlag(
attribute.getDebugKind() /*LangOpts().OpenMPTargetDebug*/,
"__omp_rtl_debug_kind");
@@ -2032,8 +2038,6 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
.getAssumeNoNestedParallelism() /*LangOpts().OpenMPNoNestedParallelism*/
,
"__omp_rtl_assume_no_nested_parallelism");
- ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
- attribute.getOpenmpDeviceVersion());
return success();
}
diff --git a/mlir/test/Dialect/OpenMP/attr.mlir b/mlir/test/Dialect/OpenMP/attr.mlir
index fc89ebd01b6bf57..2e88dac8ff4d2b9 100644
--- a/mlir/test/Dialect/OpenMP/attr.mlir
+++ b/mlir/test/Dialect/OpenMP/attr.mlir
@@ -54,6 +54,12 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, openmp_device_version = 51>} {
module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, openmp_device_version = 51>} {}
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, no_gpu_lib = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, no_gpu_lib = true, openmp_device_version = 51>} {}
+
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, no_gpu_lib = false, openmp_device_version = 51>} {}
+
// CHECK: module attributes {omp.version = #omp.version<version = 51>} {
module attributes {omp.version = #omp.version<version = 51>} {}
diff --git a/mlir/test/Target/LLVMIR/openmp-llvm.mlir b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
index 116c05f3747c6ae..8da17107466f817 100644
--- a/mlir/test/Target/LLVMIR/openmp-llvm.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
@@ -2530,6 +2530,16 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// -----
+// CHECK-NOT: @__omp_rtl_debug_kind = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_teams_oversubscription = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_threads_oversubscription = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_no_thread_state = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_no_nested_parallelism = weak_odr hidden constant i32 0
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true,
+ no_gpu_lib=true>} {}
+
+// -----
+
module attributes {omp.is_target_device = false} {
// CHECK: define void @filter_nohost
llvm.func @filter_nohost() -> ()
|
|
@llvm/pr-subscribers-flang-driver Author: Dominik Adamski (DominikAdamski) ChangesIf -nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314 Full diff: https://github.com/llvm/llvm-project/pull/71045.diff 14 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index b1229b2f4562379..cc50b45b7933d5f 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5092,7 +5092,7 @@ def nohipwrapperinc : Flag<["-"], "nohipwrapperinc">, Group<IncludePath_Group>,
HelpText<"Do not include the default HIP wrapper headers and include paths">;
def : Flag<["-"], "nocudainc">, Alias<nogpuinc>;
def nogpulib : Flag<["-"], "nogpulib">, MarshallingInfoFlag<LangOpts<"NoGPULib">>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Do not link device library for CUDA/HIP device compilation">;
def : Flag<["-"], "nocudalib">, Alias<nogpulib>;
def gpulibc : Flag<["-"], "gpulibc">, Visibility<[ClangOption, CC1Option]>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 999039f83ddfb92..17d9ac7632128d6 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -295,6 +295,8 @@ void Flang::addOffloadOptions(Compilation &C, const InputInfoList &Inputs,
CmdArgs.push_back("-fopenmp-assume-no-thread-state");
if (Args.hasArg(options::OPT_fopenmp_assume_no_nested_parallelism))
CmdArgs.push_back("-fopenmp-assume-no-nested-parallelism");
+ if (Args.hasArg(options::OPT_nogpulib))
+ CmdArgs.push_back("-nogpulib");
}
}
diff --git a/flang/include/flang/Frontend/LangOptions.def b/flang/include/flang/Frontend/LangOptions.def
index 3a1d44f7fb472d1..2bf10826120a8bd 100644
--- a/flang/include/flang/Frontend/LangOptions.def
+++ b/flang/include/flang/Frontend/LangOptions.def
@@ -21,6 +21,8 @@ LANGOPT(Name, Bits, Default)
ENUM_LANGOPT(FPContractMode, FPModeKind, 2, FPM_Fast) ///< FP Contract Mode (off/fast)
+/// Indicate a build without the standard GPU libraries.
+LANGOPT(NoGPULib , 1, false)
/// Permit floating point optimization without regard to infinities
LANGOPT(NoHonorInfs, 1, false)
/// Permit floating point optimization without regard to NaN
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index ddec70fa9824c52..3b51a3845625df3 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -53,14 +53,16 @@ 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 OpenMPIsGPU, 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) {}
+ OpenMPVersion(OpenMPVersion), OMPHostIRFile(OMPHostIRFile),
+ NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -70,7 +72,7 @@ struct OffloadModuleOpts {
OpenMPNoNestedParallelism(Opts.OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPVersion(Opts.OpenMPVersion),
- OMPHostIRFile(Opts.OMPHostIRFile) {}
+ OMPHostIRFile(Opts.OMPHostIRFile), NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -81,6 +83,7 @@ struct OffloadModuleOpts {
bool OpenMPIsGPU = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
+ bool NoGPULib = false;
};
// Shares assinging of the OpenMP OffloadModuleInterface and its assorted
@@ -95,7 +98,7 @@ void setOffloadModuleInterfaceAttributes(
if (Opts.OpenMPIsTargetDevice) {
offloadMod.setFlags(Opts.OpenMPTargetDebug, Opts.OpenMPTeamSubscription,
Opts.OpenMPThreadSubscription, Opts.OpenMPNoThreadState,
- Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion);
+ Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion, Opts.NoGPULib);
if (!Opts.OMPHostIRFile.empty())
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2ecab3742587a..ef0db22037283d3 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -881,6 +881,8 @@ static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
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()) {
diff --git a/flang/test/Driver/driver-help-hidden.f90 b/flang/test/Driver/driver-help-hidden.f90
index 6d399f1d179a022..2e65096d0c3fec4 100644
--- a/flang/test/Driver/driver-help-hidden.f90
+++ b/flang/test/Driver/driver-help-hidden.f90
@@ -119,6 +119,7 @@
! CHECK-NEXT: --no-offload-arch=<value>
! CHECK-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! CHECK-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! CHECK-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! CHECK-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! CHECK-NEXT: --offload-device-only Only compile for the offloading device.
! CHECK-NEXT: --offload-host-device Compile for both the offloading host and device (default).
diff --git a/flang/test/Driver/driver-help.f90 b/flang/test/Driver/driver-help.f90
index 31c9caa32ea8292..a29e8a0b69db46d 100644
--- a/flang/test/Driver/driver-help.f90
+++ b/flang/test/Driver/driver-help.f90
@@ -107,6 +107,7 @@
! HELP-NEXT: --no-offload-arch=<value>
! HELP-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! HELP-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! HELP-NEXT: --offload-device-only Only compile for the offloading device.
! HELP-NEXT: --offload-host-device Compile for both the offloading host and device (default).
@@ -239,6 +240,7 @@
! HELP-FC1-NEXT: -mvscale-max=<value> Specify the vscale maximum. Defaults to the vector length agnostic value of "0". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -mvscale-min=<value> Specify the vscale minimum. Defaults to "1". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-FC1-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-FC1-NEXT: -opt-record-file <value>
! HELP-FC1-NEXT: File name to use for YAML optimization record output
! HELP-FC1-NEXT: -opt-record-format <value>
diff --git a/flang/test/Lower/OpenMP/nogpulib.f90 b/flang/test/Lower/OpenMP/nogpulib.f90
new file mode 100644
index 000000000000000..f2e67136ecd7481
--- /dev/null
+++ b/flang/test/Lower/OpenMP/nogpulib.f90
@@ -0,0 +1,12 @@
+!REQUIRES: amdgpu-registered-target
+
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device %s -o - | FileCheck %s
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -o - %s | FileCheck %s
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device -nogpulib %s -o - | FileCheck %s -check-prefix=FLAG_SET
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -nogpulib -o - %s | FileCheck %s -check-prefix=FLAG_SET
+
+!CHECK-NOT: module attributes {{{.*}}no_gpu_lib
+!FLAG_SET: module attributes {{{.*}}no_gpu_lib = true
+subroutine omp_subroutine()
+end subroutine omp_subroutine
+
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index 378216bd1e51f8f..a66c41eed59651c 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -177,6 +177,12 @@ static llvm::cl::opt<bool> setOpenMPNoNestedParallelism(
"a parallel region."),
llvm::cl::init(false));
+static llvm::cl::opt<bool>
+ setNoGPULib("nogpulib",
+ llvm::cl::desc("Do not link device library for CUDA/HIP device "
+ "compilation"),
+ llvm::cl::init(false));
+
static llvm::cl::opt<bool> enableOpenACC("fopenacc",
llvm::cl::desc("enable openacc"),
llvm::cl::init(false));
@@ -316,7 +322,7 @@ static mlir::LogicalResult convertFortranSourceToMLIR(
OffloadModuleOpts(setOpenMPTargetDebug, setOpenMPTeamSubscription,
setOpenMPThreadSubscription, setOpenMPNoThreadState,
setOpenMPNoNestedParallelism, enableOpenMPDevice,
- enableOpenMPGPU, setOpenMPVersion);
+ enableOpenMPGPU, setOpenMPVersion, "", setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 72121ad4f9e47a5..ae2dd018059e3aa 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -65,6 +65,7 @@ def FlagsAttr : OpenMP_Attr<"Flags", "flags"> {
DefaultValuedParameter<"bool", "false">:$assume_threads_oversubscription,
DefaultValuedParameter<"bool", "false">:$assume_no_thread_state,
DefaultValuedParameter<"bool", "false">:$assume_no_nested_parallelism,
+ DefaultValuedParameter<"bool", "false">:$no_gpu_lib,
DefaultValuedParameter<"uint32_t", "50">:$openmp_device_version
);
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
index 77001fc816cf91f..89d04af64766fc2 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
@@ -198,11 +198,12 @@ def OffloadModuleInterface : OpInterface<"OffloadModuleInterface"> {
"bool":$assumeThreadsOversubscription,
"bool":$assumeNoThreadState,
"bool":$assumeNoNestedParallelism,
- "uint32_t":$openmpDeviceVersion), [{}], [{
+ "uint32_t":$openmpDeviceVersion,
+ "bool":$noGPULib), [{}], [{
$_op->setAttr(("omp." + mlir::omp::FlagsAttr::getMnemonic()).str(),
mlir::omp::FlagsAttr::get($_op->getContext(), debugKind,
assumeTeamsOversubscription, assumeThreadsOversubscription,
- assumeNoThreadState, assumeNoNestedParallelism, openmpDeviceVersion));
+ assumeNoThreadState, assumeNoNestedParallelism, noGPULib, openmpDeviceVersion));
}]>,
InterfaceMethod<
/*description=*/[{
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 1daf60b8659bb66..083b507f405f426 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -2011,6 +2011,12 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
llvm::OpenMPIRBuilder *ompBuilder = moduleTranslation.getOpenMPBuilder();
+ ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
+ attribute.getOpenmpDeviceVersion());
+
+ if (attribute.getNoGpuLib())
+ return success();
+
ompBuilder->createGlobalFlag(
attribute.getDebugKind() /*LangOpts().OpenMPTargetDebug*/,
"__omp_rtl_debug_kind");
@@ -2032,8 +2038,6 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
.getAssumeNoNestedParallelism() /*LangOpts().OpenMPNoNestedParallelism*/
,
"__omp_rtl_assume_no_nested_parallelism");
- ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
- attribute.getOpenmpDeviceVersion());
return success();
}
diff --git a/mlir/test/Dialect/OpenMP/attr.mlir b/mlir/test/Dialect/OpenMP/attr.mlir
index fc89ebd01b6bf57..2e88dac8ff4d2b9 100644
--- a/mlir/test/Dialect/OpenMP/attr.mlir
+++ b/mlir/test/Dialect/OpenMP/attr.mlir
@@ -54,6 +54,12 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, openmp_device_version = 51>} {
module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, openmp_device_version = 51>} {}
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, no_gpu_lib = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, no_gpu_lib = true, openmp_device_version = 51>} {}
+
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, no_gpu_lib = false, openmp_device_version = 51>} {}
+
// CHECK: module attributes {omp.version = #omp.version<version = 51>} {
module attributes {omp.version = #omp.version<version = 51>} {}
diff --git a/mlir/test/Target/LLVMIR/openmp-llvm.mlir b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
index 116c05f3747c6ae..8da17107466f817 100644
--- a/mlir/test/Target/LLVMIR/openmp-llvm.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
@@ -2530,6 +2530,16 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// -----
+// CHECK-NOT: @__omp_rtl_debug_kind = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_teams_oversubscription = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_threads_oversubscription = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_no_thread_state = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_no_nested_parallelism = weak_odr hidden constant i32 0
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true,
+ no_gpu_lib=true>} {}
+
+// -----
+
module attributes {omp.is_target_device = false} {
// CHECK: define void @filter_nohost
llvm.func @filter_nohost() -> ()
|
|
@llvm/pr-subscribers-mlir-llvm Author: Dominik Adamski (DominikAdamski) ChangesIf -nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314 Full diff: https://github.com/llvm/llvm-project/pull/71045.diff 14 Files Affected:
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index b1229b2f4562379..cc50b45b7933d5f 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5092,7 +5092,7 @@ def nohipwrapperinc : Flag<["-"], "nohipwrapperinc">, Group<IncludePath_Group>,
HelpText<"Do not include the default HIP wrapper headers and include paths">;
def : Flag<["-"], "nocudainc">, Alias<nogpuinc>;
def nogpulib : Flag<["-"], "nogpulib">, MarshallingInfoFlag<LangOpts<"NoGPULib">>,
- Visibility<[ClangOption, CC1Option]>,
+ Visibility<[ClangOption, CC1Option, FlangOption, FC1Option]>,
HelpText<"Do not link device library for CUDA/HIP device compilation">;
def : Flag<["-"], "nocudalib">, Alias<nogpulib>;
def gpulibc : Flag<["-"], "gpulibc">, Visibility<[ClangOption, CC1Option]>,
diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
index 999039f83ddfb92..17d9ac7632128d6 100644
--- a/clang/lib/Driver/ToolChains/Flang.cpp
+++ b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -295,6 +295,8 @@ void Flang::addOffloadOptions(Compilation &C, const InputInfoList &Inputs,
CmdArgs.push_back("-fopenmp-assume-no-thread-state");
if (Args.hasArg(options::OPT_fopenmp_assume_no_nested_parallelism))
CmdArgs.push_back("-fopenmp-assume-no-nested-parallelism");
+ if (Args.hasArg(options::OPT_nogpulib))
+ CmdArgs.push_back("-nogpulib");
}
}
diff --git a/flang/include/flang/Frontend/LangOptions.def b/flang/include/flang/Frontend/LangOptions.def
index 3a1d44f7fb472d1..2bf10826120a8bd 100644
--- a/flang/include/flang/Frontend/LangOptions.def
+++ b/flang/include/flang/Frontend/LangOptions.def
@@ -21,6 +21,8 @@ LANGOPT(Name, Bits, Default)
ENUM_LANGOPT(FPContractMode, FPModeKind, 2, FPM_Fast) ///< FP Contract Mode (off/fast)
+/// Indicate a build without the standard GPU libraries.
+LANGOPT(NoGPULib , 1, false)
/// Permit floating point optimization without regard to infinities
LANGOPT(NoHonorInfs, 1, false)
/// Permit floating point optimization without regard to NaN
diff --git a/flang/include/flang/Tools/CrossToolHelpers.h b/flang/include/flang/Tools/CrossToolHelpers.h
index ddec70fa9824c52..3b51a3845625df3 100644
--- a/flang/include/flang/Tools/CrossToolHelpers.h
+++ b/flang/include/flang/Tools/CrossToolHelpers.h
@@ -53,14 +53,16 @@ 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 OpenMPIsGPU, 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) {}
+ OpenMPVersion(OpenMPVersion), OMPHostIRFile(OMPHostIRFile),
+ NoGPULib(NoGPULib) {}
OffloadModuleOpts(Fortran::frontend::LangOptions &Opts)
: OpenMPTargetDebug(Opts.OpenMPTargetDebug),
@@ -70,7 +72,7 @@ struct OffloadModuleOpts {
OpenMPNoNestedParallelism(Opts.OpenMPNoNestedParallelism),
OpenMPIsTargetDevice(Opts.OpenMPIsTargetDevice),
OpenMPIsGPU(Opts.OpenMPIsGPU), OpenMPVersion(Opts.OpenMPVersion),
- OMPHostIRFile(Opts.OMPHostIRFile) {}
+ OMPHostIRFile(Opts.OMPHostIRFile), NoGPULib(Opts.NoGPULib) {}
uint32_t OpenMPTargetDebug = 0;
bool OpenMPTeamSubscription = false;
@@ -81,6 +83,7 @@ struct OffloadModuleOpts {
bool OpenMPIsGPU = false;
uint32_t OpenMPVersion = 11;
std::string OMPHostIRFile = {};
+ bool NoGPULib = false;
};
// Shares assinging of the OpenMP OffloadModuleInterface and its assorted
@@ -95,7 +98,7 @@ void setOffloadModuleInterfaceAttributes(
if (Opts.OpenMPIsTargetDevice) {
offloadMod.setFlags(Opts.OpenMPTargetDebug, Opts.OpenMPTeamSubscription,
Opts.OpenMPThreadSubscription, Opts.OpenMPNoThreadState,
- Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion);
+ Opts.OpenMPNoNestedParallelism, Opts.OpenMPVersion, Opts.NoGPULib);
if (!Opts.OMPHostIRFile.empty())
offloadMod.setHostIRFilePath(Opts.OMPHostIRFile);
diff --git a/flang/lib/Frontend/CompilerInvocation.cpp b/flang/lib/Frontend/CompilerInvocation.cpp
index ba2ecab3742587a..ef0db22037283d3 100644
--- a/flang/lib/Frontend/CompilerInvocation.cpp
+++ b/flang/lib/Frontend/CompilerInvocation.cpp
@@ -881,6 +881,8 @@ static bool parseDialectArgs(CompilerInvocation &res, llvm::opt::ArgList &args,
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()) {
diff --git a/flang/test/Driver/driver-help-hidden.f90 b/flang/test/Driver/driver-help-hidden.f90
index 6d399f1d179a022..2e65096d0c3fec4 100644
--- a/flang/test/Driver/driver-help-hidden.f90
+++ b/flang/test/Driver/driver-help-hidden.f90
@@ -119,6 +119,7 @@
! CHECK-NEXT: --no-offload-arch=<value>
! CHECK-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! CHECK-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! CHECK-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! CHECK-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! CHECK-NEXT: --offload-device-only Only compile for the offloading device.
! CHECK-NEXT: --offload-host-device Compile for both the offloading host and device (default).
diff --git a/flang/test/Driver/driver-help.f90 b/flang/test/Driver/driver-help.f90
index 31c9caa32ea8292..a29e8a0b69db46d 100644
--- a/flang/test/Driver/driver-help.f90
+++ b/flang/test/Driver/driver-help.f90
@@ -107,6 +107,7 @@
! HELP-NEXT: --no-offload-arch=<value>
! HELP-NEXT: Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. 'all' resets the list to its default value.
! HELP-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-NEXT: --offload-arch=<value> Specify an offloading device architecture for CUDA, HIP, or OpenMP. (e.g. sm_35). If 'native' is used the compiler will detect locally installed architectures. For HIP offloading, the device architecture can be followed by target ID features delimited by a colon (e.g. gfx908:xnack+:sramecc-). May be specified more than once.
! HELP-NEXT: --offload-device-only Only compile for the offloading device.
! HELP-NEXT: --offload-host-device Compile for both the offloading host and device (default).
@@ -239,6 +240,7 @@
! HELP-FC1-NEXT: -mvscale-max=<value> Specify the vscale maximum. Defaults to the vector length agnostic value of "0". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -mvscale-min=<value> Specify the vscale minimum. Defaults to "1". (AArch64/RISC-V only)
! HELP-FC1-NEXT: -nocpp Disable predefined and command line preprocessor macros
+! HELP-FC1-NEXT: -nogpulib Do not link device library for CUDA/HIP device compilation
! HELP-FC1-NEXT: -opt-record-file <value>
! HELP-FC1-NEXT: File name to use for YAML optimization record output
! HELP-FC1-NEXT: -opt-record-format <value>
diff --git a/flang/test/Lower/OpenMP/nogpulib.f90 b/flang/test/Lower/OpenMP/nogpulib.f90
new file mode 100644
index 000000000000000..f2e67136ecd7481
--- /dev/null
+++ b/flang/test/Lower/OpenMP/nogpulib.f90
@@ -0,0 +1,12 @@
+!REQUIRES: amdgpu-registered-target
+
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device %s -o - | FileCheck %s
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -o - %s | FileCheck %s
+!RUN: %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device -nogpulib %s -o - | FileCheck %s -check-prefix=FLAG_SET
+!RUN: bbc -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -emit-hlfir -nogpulib -o - %s | FileCheck %s -check-prefix=FLAG_SET
+
+!CHECK-NOT: module attributes {{{.*}}no_gpu_lib
+!FLAG_SET: module attributes {{{.*}}no_gpu_lib = true
+subroutine omp_subroutine()
+end subroutine omp_subroutine
+
diff --git a/flang/tools/bbc/bbc.cpp b/flang/tools/bbc/bbc.cpp
index 378216bd1e51f8f..a66c41eed59651c 100644
--- a/flang/tools/bbc/bbc.cpp
+++ b/flang/tools/bbc/bbc.cpp
@@ -177,6 +177,12 @@ static llvm::cl::opt<bool> setOpenMPNoNestedParallelism(
"a parallel region."),
llvm::cl::init(false));
+static llvm::cl::opt<bool>
+ setNoGPULib("nogpulib",
+ llvm::cl::desc("Do not link device library for CUDA/HIP device "
+ "compilation"),
+ llvm::cl::init(false));
+
static llvm::cl::opt<bool> enableOpenACC("fopenacc",
llvm::cl::desc("enable openacc"),
llvm::cl::init(false));
@@ -316,7 +322,7 @@ static mlir::LogicalResult convertFortranSourceToMLIR(
OffloadModuleOpts(setOpenMPTargetDebug, setOpenMPTeamSubscription,
setOpenMPThreadSubscription, setOpenMPNoThreadState,
setOpenMPNoNestedParallelism, enableOpenMPDevice,
- enableOpenMPGPU, setOpenMPVersion);
+ enableOpenMPGPU, setOpenMPVersion, "", setNoGPULib);
setOffloadModuleInterfaceAttributes(mlirModule, offloadModuleOpts);
setOpenMPVersionAttribute(mlirModule, setOpenMPVersion);
}
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 72121ad4f9e47a5..ae2dd018059e3aa 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -65,6 +65,7 @@ def FlagsAttr : OpenMP_Attr<"Flags", "flags"> {
DefaultValuedParameter<"bool", "false">:$assume_threads_oversubscription,
DefaultValuedParameter<"bool", "false">:$assume_no_thread_state,
DefaultValuedParameter<"bool", "false">:$assume_no_nested_parallelism,
+ DefaultValuedParameter<"bool", "false">:$no_gpu_lib,
DefaultValuedParameter<"uint32_t", "50">:$openmp_device_version
);
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
index 77001fc816cf91f..89d04af64766fc2 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOpsInterfaces.td
@@ -198,11 +198,12 @@ def OffloadModuleInterface : OpInterface<"OffloadModuleInterface"> {
"bool":$assumeThreadsOversubscription,
"bool":$assumeNoThreadState,
"bool":$assumeNoNestedParallelism,
- "uint32_t":$openmpDeviceVersion), [{}], [{
+ "uint32_t":$openmpDeviceVersion,
+ "bool":$noGPULib), [{}], [{
$_op->setAttr(("omp." + mlir::omp::FlagsAttr::getMnemonic()).str(),
mlir::omp::FlagsAttr::get($_op->getContext(), debugKind,
assumeTeamsOversubscription, assumeThreadsOversubscription,
- assumeNoThreadState, assumeNoNestedParallelism, openmpDeviceVersion));
+ assumeNoThreadState, assumeNoNestedParallelism, noGPULib, openmpDeviceVersion));
}]>,
InterfaceMethod<
/*description=*/[{
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 1daf60b8659bb66..083b507f405f426 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -2011,6 +2011,12 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
llvm::OpenMPIRBuilder *ompBuilder = moduleTranslation.getOpenMPBuilder();
+ ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
+ attribute.getOpenmpDeviceVersion());
+
+ if (attribute.getNoGpuLib())
+ return success();
+
ompBuilder->createGlobalFlag(
attribute.getDebugKind() /*LangOpts().OpenMPTargetDebug*/,
"__omp_rtl_debug_kind");
@@ -2032,8 +2038,6 @@ LogicalResult convertFlagsAttr(Operation *op, mlir::omp::FlagsAttr attribute,
.getAssumeNoNestedParallelism() /*LangOpts().OpenMPNoNestedParallelism*/
,
"__omp_rtl_assume_no_nested_parallelism");
- ompBuilder->M.addModuleFlag(llvm::Module::Max, "openmp-device",
- attribute.getOpenmpDeviceVersion());
return success();
}
diff --git a/mlir/test/Dialect/OpenMP/attr.mlir b/mlir/test/Dialect/OpenMP/attr.mlir
index fc89ebd01b6bf57..2e88dac8ff4d2b9 100644
--- a/mlir/test/Dialect/OpenMP/attr.mlir
+++ b/mlir/test/Dialect/OpenMP/attr.mlir
@@ -54,6 +54,12 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, openmp_device_version = 51>} {
module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, openmp_device_version = 51>} {}
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true, no_gpu_lib = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_no_thread_state = true, assume_teams_oversubscription = true, no_gpu_lib = true, openmp_device_version = 51>} {}
+
+// CHECK: module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, openmp_device_version = 51>} {
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, no_gpu_lib = false, openmp_device_version = 51>} {}
+
// CHECK: module attributes {omp.version = #omp.version<version = 51>} {
module attributes {omp.version = #omp.version<version = 51>} {}
diff --git a/mlir/test/Target/LLVMIR/openmp-llvm.mlir b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
index 116c05f3747c6ae..8da17107466f817 100644
--- a/mlir/test/Target/LLVMIR/openmp-llvm.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-llvm.mlir
@@ -2530,6 +2530,16 @@ module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true,
// -----
+// CHECK-NOT: @__omp_rtl_debug_kind = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_teams_oversubscription = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_threads_oversubscription = weak_odr hidden constant i32 0
+// CHECK-NOT: @__omp_rtl_assume_no_thread_state = weak_odr hidden constant i32 1
+// CHECK-NOT: @__omp_rtl_assume_no_nested_parallelism = weak_odr hidden constant i32 0
+module attributes {omp.flags = #omp.flags<assume_teams_oversubscription = true, assume_no_thread_state = true,
+ no_gpu_lib=true>} {}
+
+// -----
+
module attributes {omp.is_target_device = false} {
// CHECK: define void @filter_nohost
llvm.func @filter_nohost() -> ()
|
|
It would be very nice to have this option added to Flang in the near future, as this option would allow the OpenMP libomptarget runtime tests for offload to be ran without failure due to this missing option. I believe we are currently in the process of setting up a buildbot that includes these tests, it's not a blocker for the buildbot from what I understand, but it'd be very nice to be able to test for success of these tests on commits in the near future, especially as the suite expands. |
|
We have the buildbot now up in staging: https://lab.llvm.org/staging/#/builders/188 |
justinfargnoli
pushed a commit
to justinfargnoli/llvm-project
that referenced
this pull request
Jan 28, 2024
If -nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime. Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314
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.
If -nogpulib option is passed by the user, then the OpenMP device runtime is not used and we should not emit globals to configure debugging at compile-time for the device runtime.
Link to -nogpulib flag implementation for Clang: https://reviews.llvm.org/D125314