[OpenMP] Add flag for specifying the target device architecture for OpenMP device offloading
Summary:
OpenMP has the ability to offload target regions to devices which may have different architectures.
A new -fopenmp-target-arch flag is introduced to specify the device architecture.
In this patch I use the new flag to specify the compute capability of the underlying NVIDIA architecture for the OpenMP offloading CUDA tool chain.
Only a host-offloading test is provided since full device offloading capability will only be available when [[ https://reviews.llvm.org/D29654 | D29654 ]] lands.
Reviewers: hfinkel, Hahnfeld, carlo.bertolli, caomhin, ABataev
Reviewed By: hfinkel
Subscribers: guansong, cfe-commits
Tags: #openmp
Differential Revision: https://reviews.llvm.org/D34784
llvm-svn: 310263
diff --git a/clang/lib/Driver/Compilation.cpp b/clang/lib/Driver/Compilation.cpp
index cf86644..6626662 100644
--- a/clang/lib/Driver/Compilation.cpp
+++ b/clang/lib/Driver/Compilation.cpp
@@ -59,7 +59,12 @@
DerivedArgList *&Entry = TCArgs[{TC, BoundArch, DeviceOffloadKind}];
if (!Entry) {
- Entry = TC->TranslateArgs(*TranslatedArgs, BoundArch, DeviceOffloadKind);
+ // Translate OpenMP toolchain arguments provided via the -Xopenmp-target flags.
+ Entry = TC->TranslateOpenMPTargetArgs(*TranslatedArgs, DeviceOffloadKind);
+ if (!Entry)
+ Entry = TranslatedArgs;
+
+ Entry = TC->TranslateArgs(*Entry, BoundArch, DeviceOffloadKind);
if (!Entry)
Entry = TranslatedArgs;
}
diff --git a/clang/lib/Driver/ToolChain.cpp b/clang/lib/Driver/ToolChain.cpp
index 1e446dc..94d5e4f 100644
--- a/clang/lib/Driver/ToolChain.cpp
+++ b/clang/lib/Driver/ToolChain.cpp
@@ -795,3 +795,65 @@
return VersionTuple();
}
+
+llvm::opt::DerivedArgList *
+ToolChain::TranslateOpenMPTargetArgs(const llvm::opt::DerivedArgList &Args,
+ Action::OffloadKind DeviceOffloadKind) const {
+ if (DeviceOffloadKind == Action::OFK_OpenMP) {
+ DerivedArgList *DAL = new DerivedArgList(Args.getBaseArgs());
+ const OptTable &Opts = getDriver().getOpts();
+
+ // Handle -Xopenmp-target flags
+ for (Arg *A : Args) {
+ // Exclude flags which may only apply to the host toolchain.
+ // Do not exclude flags when the host triple (AuxTriple),
+ // matches the current toolchain triple.
+ if (A->getOption().matches(options::OPT_m_Group)) {
+ if (getAuxTriple() && getAuxTriple()->str() == getTriple().str())
+ DAL->append(A);
+ continue;
+ }
+
+ unsigned Index;
+ unsigned Prev;
+ bool XOpenMPTargetNoTriple = A->getOption().matches(
+ options::OPT_Xopenmp_target);
+
+ if (A->getOption().matches(options::OPT_Xopenmp_target_EQ)) {
+ // Passing device args: -Xopenmp-target=<triple> -opt=val.
+ if (A->getValue(0) == getTripleString())
+ Index = Args.getBaseArgs().MakeIndex(A->getValue(1));
+ else
+ continue;
+ } else if (XOpenMPTargetNoTriple) {
+ // Passing device args: -Xopenmp-target -opt=val.
+ Index = Args.getBaseArgs().MakeIndex(A->getValue(0));
+ } else {
+ DAL->append(A);
+ continue;
+ }
+
+ // Parse the argument to -Xopenmp-target.
+ Prev = Index;
+ std::unique_ptr<Arg> XOpenMPTargetArg(Opts.ParseOneArg(Args, Index));
+ if (!XOpenMPTargetArg || Index > Prev + 1) {
+ getDriver().Diag(diag::err_drv_invalid_Xopenmp_target_with_args)
+ << A->getAsString(Args);
+ continue;
+ }
+ if (XOpenMPTargetNoTriple && XOpenMPTargetArg &&
+ Args.getAllArgValues(
+ options::OPT_fopenmp_targets_EQ).size() != 1) {
+ getDriver().Diag(diag::err_drv_Xopenmp_target_missing_triple);
+ continue;
+ }
+ XOpenMPTargetArg->setBaseArg(A);
+ A = XOpenMPTargetArg.release();
+ DAL->append(A);
+ }
+
+ return DAL;
+ }
+
+ return nullptr;
+}
diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index 935a5a3..86be187 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -212,8 +212,18 @@
static_cast<const toolchains::CudaToolChain &>(getToolChain());
assert(TC.getTriple().isNVPTX() && "Wrong platform");
+ StringRef GPUArchName;
+ // If this is an OpenMP action we need to extract the device architecture
+ // from the -march=arch option. This option may come from -Xopenmp-target
+ // flag or the default value.
+ if (JA.isDeviceOffloading(Action::OFK_OpenMP)) {
+ GPUArchName = Args.getLastArgValue(options::OPT_march_EQ);
+ assert(!GPUArchName.empty() && "Must have an architecture passed in.");
+ } else
+ GPUArchName = JA.getOffloadingArch();
+
// Obtain architecture from the action.
- CudaArch gpu_arch = StringToCudaArch(JA.getOffloadingArch());
+ CudaArch gpu_arch = StringToCudaArch(GPUArchName);
assert(gpu_arch != CudaArch::UNKNOWN &&
"Device action expected to have an architecture.");
@@ -405,7 +415,7 @@
// For OpenMP device offloading, append derived arguments. Make sure
// flags are not duplicated.
- // TODO: Append the compute capability.
+ // Also append the compute capability.
if (DeviceOffloadKind == Action::OFK_OpenMP) {
for (Arg *A : Args){
bool IsDuplicate = false;
@@ -418,6 +428,13 @@
if (!IsDuplicate)
DAL->append(A);
}
+
+ StringRef Arch = DAL->getLastArgValue(options::OPT_march_EQ);
+ if (Arch.empty())
+ // Default compute capability for CUDA toolchain is sm_20.
+ DAL->AddJoinedArg(nullptr,
+ Opts.getOption(options::OPT_march_EQ), "sm_20");
+
return DAL;
}