[llvm-branch-commits] [clang] clang/AMDGPU: Use TranslateArgs from the base toolchain instead of the host (PR #198627)

[Matt Arsenault via llvm-branch-commits]

================
@@ -3485,12 +3485,19 @@ Generic_GCC::TranslateArgs(const llvm::opt::DerivedArgList &Args,
 
   // Add the bound architecture to the arguments list if present.
   if (!BoundArch.empty()) {
-    options::ID Opt =
-        getTriple().isARM() || getTriple().isPPC() || getTriple().isAArch64()
-            ? options::OPT_mcpu_EQ
-            : options::OPT_march_EQ;
-    DAL->eraseArg(Opt);
-    DAL->AddJoinedArg(nullptr, Opts.getOption(Opt), BoundArch);
+    if (getTriple().isAMDGPU()) {
----------------
arsenm wrote:

As far as naming -march kind of makes more sense for a GPU 

https://github.com/llvm/llvm-project/pull/198627