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

[Matt Arsenault via flang-commits]

================
@@ -1226,11 +1233,19 @@ ToolChain::path_list ToolChain::getArchSpecificLibPaths() const {
 
   AddPath({getTriple().str()});
 
-  // Handle the legacy AMDGPU triple case as well.
-  if (getTriple().getArchName() == "amdgcn") {
+  // For AMDGPU, fall back to the subarch-stripped triple path, trying both the
+  // canonical "amdgpu" name and the legacy "amdgcn" name.
+  //
+  // TODO: Also try major subarch?
----------------
arsenm wrote:

This is just getting the tests to pass, the actual library layout needs work 

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