[flang-commits] [clang] [flang] [llvm] clang: Start using new amdgpu subarch triples (PR #206482)

[Matt Arsenault via flang-commits]

================
@@ -1124,11 +1125,17 @@ ToolChain::getTargetSubDirPath(StringRef BaseDir) const {
   if (auto Path = getPathForTriple(T))
     return *Path;
 
-  // Handle the legacy AMDGPU triple case as well.
-  if (T.getArchName() == "amdgcn") {
-    llvm::Triple Canon(T);
-    Canon.setArchName("amdgpu");
-    if (auto Path = getPathForTriple(Canon))
+  if (T.isAMDGCN()) {
+    // Clear the subarch as a fallback.
+    // TODO: Remove this when libc and compiler-rt builds are migrated.
----------------
arsenm wrote:

The build structure needs reworking, it is still producing the flat single amdgpu name instead of subarch specific builds 

https://github.com/llvm/llvm-project/pull/206482