================
@@ -1044,82 +1040,78 @@ void
Driver::CreateOffloadingDeviceToolChains(Compilation &C,
<< OpenMPTargets->getAsString(C.getInputArgs());
return;
}
+
+ // Make sure these show up in a deterministic order.
+ std::multiset<StringRef> OpenMPTriples;
for (StringRef T : OpenMPTargets->getValues())
OpenMPTriples.insert(T);
+
+ llvm::StringMap<StringRef> FoundNormalizedTriples;
+ for (StringRef T : OpenMPTriples) {
+ llvm::Triple TT(ToolChain::getOpenMPTriple(T));
+ std::string NormalizedName = TT.normalize();
+
+ // Make sure we don't have a duplicate triple.
+ auto [TripleIt, Inserted] =
+ FoundNormalizedTriples.try_emplace(NormalizedName, T);
+ if (!Inserted) {
+ Diag(clang::diag::warn_drv_omp_offload_target_duplicate)
+ << T << TripleIt->second;
+ continue;
+ }
+
+ // If the specified target is invalid, emit a diagnostic.
+ if (TT.getArch() == llvm::Triple::UnknownArch) {
+ Diag(clang::diag::err_drv_invalid_omp_target) << T;
+ continue;
+ }
+
+ auto &TC = getOffloadToolChain(C.getInputArgs(), Action::OFK_OpenMP,
TT,
+ C.getDefaultToolChain().getTriple());
+ C.addOffloadDeviceToolChain(&TC, Action::OFK_OpenMP);
+ }
} else if (C.getInputArgs().hasArg(options::OPT_offload_arch_EQ) &&
((!IsHIP && !IsCuda) || UseLLVMOffload)) {
- const ToolChain *HostTC =
C.getSingleOffloadToolChain<Action::OFK_Host>();
- auto AMDTriple = getHIPOffloadTargetTriple(*this, C.getInputArgs());
- auto NVPTXTriple = getNVIDIAOffloadTargetTriple(*this, C.getInputArgs(),
- HostTC->getTriple());
+ llvm::Triple AMDTriple("amdgcn-amd-amdhsa");
+ llvm::Triple NVPTXTriple("nvptx64-nvidia-cuda");
// Attempt to deduce the offloading triple from the set of architectures.
// We can only correctly deduce NVPTX / AMDGPU triples currently.
- // We need to temporarily create these toolchains so that we can access
- // tools for inferring architectures.
- llvm::DenseSet<StringRef> Archs;
- for (const std::optional<llvm::Triple> &TT : {NVPTXTriple, AMDTriple}) {
- if (!TT)
- continue;
-
- auto &TC =
- getOffloadToolChain(C.getInputArgs(), Action::OFK_OpenMP, *TT,
- C.getDefaultToolChain().getTriple());
- for (StringRef Arch :
- getOffloadArchs(C, C.getArgs(), Action::OFK_OpenMP, &TC, true))
- Archs.insert(Arch);
- }
+ for (const llvm::Triple &TT : {AMDTriple, NVPTXTriple}) {
+ auto &TC = getOffloadToolChain(C.getInputArgs(), Action::OFK_OpenMP,
TT,
+ C.getDefaultToolChain().getTriple());
+
+ llvm::DenseSet<StringRef> Archs =
+ getOffloadArchs(C, C.getArgs(), Action::OFK_OpenMP, &TC, true);
+ llvm::DenseSet<StringRef> ArchsForTarget;
+ for (StringRef Arch : Archs) {
+ bool IsNVPTX = IsNVIDIAOffloadArch(
+ StringToOffloadArch(getProcessorFromTargetID(NVPTXTriple,
Arch)));
+ bool IsAMDGPU = IsAMDOffloadArch(
+ StringToOffloadArch(getProcessorFromTargetID(AMDTriple, Arch)));
+ if (!IsNVPTX && !IsAMDGPU && !Arch.equals_insensitive("native")) {
+ Diag(clang::diag::err_drv_failed_to_deduce_target_from_arch)
+ << Arch;
+ return;
+ }
- for (StringRef Arch : Archs) {
- if (NVPTXTriple && IsNVIDIAOffloadArch(StringToOffloadArch(
- getProcessorFromTargetID(*NVPTXTriple, Arch))))
{
- DerivedArchs[NVPTXTriple->getTriple()].insert(Arch);
- } else if (AMDTriple &&
- IsAMDOffloadArch(StringToOffloadArch(
- getProcessorFromTargetID(*AMDTriple, Arch)))) {
- DerivedArchs[AMDTriple->getTriple()].insert(Arch);
- } else {
- Diag(clang::diag::err_drv_failed_to_deduce_target_from_arch) << Arch;
- return;
+ if (TT.isNVPTX() && IsNVPTX) {
+ ArchsForTarget.insert(Arch);
+ } else if (TT.isAMDGPU() && IsAMDGPU) {
+ ArchsForTarget.insert(Arch);
+ }
----------------
shiltian wrote:
```suggestion
if (TT.isNVPTX() && IsNVPTX)
ArchsForTarget.insert(Arch);
else if (TT.isAMDGPU() && IsAMDGPU)
ArchsForTarget.insert(Arch);
```
https://github.com/llvm/llvm-project/pull/145549
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
