[clang] [AMDGPU][SPIR-V] Fix treating SPIR-V input as the wrong LLVM-IR (PR #202986)

[Alex Voicu via cfe-commits]

================
@@ -5385,22 +5385,20 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-emit-llvm-uselists");
 
     if (IsUsingLTO) {
+      const Arg *LTOArg = Args.getLastArg(options::OPT_foffload_lto,
+                                          options::OPT_foffload_lto_EQ);
       if (IsDeviceOffloadAction && !JA.isDeviceOffloading(Action::OFK_OpenMP) &&
           !Args.hasFlag(options::OPT_offload_new_driver,
                         options::OPT_no_offload_new_driver,
                         C.getActiveOffloadKinds() != Action::OFK_None) &&
-          !Triple.isAMDGPU()) {
+          !Triple.isAMDGPU() && !Triple.isSPIRV()) {
----------------
AlexVlx wrote:

Does vanilla SPIR-V work @sarnex? If not we'll have to gate on vendor == AMD for SPIRV

https://github.com/llvm/llvm-project/pull/202986