[clang] [clang][Tooling] Fix assertion failure when processing CUDA files (PR #173762)

[Joseph Huber via cfe-commits]

================
@@ -85,6 +85,13 @@ TEST(runToolOnCode, CudaSyntaxOnly) {
        "-nocudainc"}));
 }
 
+TEST(runToolOnCode, CudaMultipleArchs) {
+  EXPECT_TRUE(runToolOnCodeWithArgs(
+      std::make_unique<TestAction>(std::make_unique<clang::ASTConsumer>()), "",
+      {"-fsyntax-only", "-x", "cuda", "--offload-arch=sm_70,sm_80",
+       "-nocudalib", "-nocudainc"}));
----------------
jhuber6 wrote:

We should probably test both configurations, one architecture and multiple. They actrually have slightly different semantics because `-o` is forbidden with more than one.

https://github.com/llvm/llvm-project/pull/173762