[Openmp-commits] [openmp] [OpenMP] Enable automatic unified shared memory on MI300A. (PR #77512)

[via Openmp-commits]

================
@@ -183,6 +183,29 @@ Error asyncMemCopy(bool UseMultipleSdmaEngines, void *Dst, hsa_agent_t DstAgent,
 #endif
 }
 
+Expected<std::string> getTargetTripleAndFeatures(hsa_agent_t Agent) {
+  std::string Target;
+  auto Err = utils::iterateAgentISAs(Agent, [&](hsa_isa_t ISA) {
+    uint32_t Length;
+    hsa_status_t Status;
+    Status = hsa_isa_get_info_alt(ISA, HSA_ISA_INFO_NAME_LENGTH, &Length);
+    if (Status != HSA_STATUS_SUCCESS)
+      return Status;
+
+    llvm::SmallVector<char> ISAName(Length);
+    Status = hsa_isa_get_info_alt(ISA, HSA_ISA_INFO_NAME, ISAName.begin());
+    if (Status != HSA_STATUS_SUCCESS)
+      return Status;
+
+    llvm::StringRef TripleTarget(ISAName.begin(), Length);
+    if (TripleTarget.consume_front("amdgcn-amd-amdhsa"))
+      Target = TripleTarget.ltrim('-').rtrim('\0').str();
+    return HSA_STATUS_SUCCESS;
----------------
carlobertolli wrote:

We really loop over a single agent and I checked with rocr team: it can only resolve in a single isa. I am guessing that the HSA API (of which ROCr is only an implementation) leaves room for multiple ISAs associated with each agent, but that doesn't happen in this case.

https://github.com/llvm/llvm-project/pull/77512