[clang] OpenMP: Reimplement getOffloadArch (PR #189561)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-clang

Author: Matt Arsenault (arsenm)

<details>
<summary>Changes</summary>

This function made no sense at all. It was scanning through
the feature map looking for something that parsed as an OffloadArch.
Directly compute the arch from the target device.

I don't know why there isn't just an OffloadArch in TargetOpts,
this shouldn't really require parsing.

---
Full diff: https://github.com/llvm/llvm-project/pull/189561.diff


1 Files Affected:

- (modified) clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp (+3-12) 


``````````diff
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
index 4fc10e4165465..1ca04e18adcb0 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -2261,18 +2261,9 @@ bool CGOpenMPRuntimeGPU::hasAllocateAttributeForGlobalVar(const VarDecl *VD,
   return false;
 }
 
-// Get current OffloadArch and ignore any unknown values
-static OffloadArch getOffloadArch(CodeGenModule &CGM) {
-  if (!CGM.getTarget().hasFeature("ptx"))
-    return OffloadArch::Unknown;
-  for (const auto &Feature : CGM.getTarget().getTargetOpts().FeatureMap) {
-    if (Feature.getValue()) {
-      OffloadArch Arch = StringToOffloadArch(Feature.getKey());
-      if (Arch != OffloadArch::Unknown)
-        return Arch;
-    }
-  }
-  return OffloadArch::Unknown;
+static OffloadArch getOffloadArch(const CodeGenModule &CGM) {
+  // FIXME: This should not require parsing
+  return StringToOffloadArch(CGM.getTarget().getTargetOpts().CPU);
 }
 
 /// Check to see if target architecture supports unified addressing which is

``````````

</details>


https://github.com/llvm/llvm-project/pull/189561