[ClangLinkerWrapper][SPIRV] Fix embed bitcode option (#213137)

We need to pass the right linker option and not try to containerize it.

Fixes two OpenMP offload JIT tests, the other has some front end assert
I will look at.

Signed-off-by: Nick Sarnie <nick.sarnie@intel.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

@@ -586,8 +586,13 @@ Expected<StringRef> clang(ArrayRef<StringRef> InputFiles, const ArgList &Args,
   if (SaveTemps && linkerSupportsLTO(Args))
     CmdArgs.push_back("-Wl,--save-temps");
 
-  if (Args.hasArg(OPT_embed_bitcode))
-    CmdArgs.push_back("-Wl,--lto-emit-llvm");
+  if (Args.hasArg(OPT_embed_bitcode)) {
+    // SPIR-V does not use the LTO linker path, it links bitcode via llvm-link.
+    if (Triple.isSPIRV())
+      CmdArgs.push_back("-emit-llvm");
+    else
+      CmdArgs.push_back("-Wl,--lto-emit-llvm");
+  }
 
   // For linking device code with the SYCL offload kind, special handling is
   // required. Passing --sycl-link to clang results in a call to
@@ -640,7 +645,8 @@ Error containerizeRawImage(std::unique_ptr<MemoryBuffer> &Img, OffloadKind Kind,
                            const ArgList &Args) {
   llvm::Triple Triple(Args.getLastArgValue(OPT_triple_EQ));
   if (Kind == OFK_OpenMP && Triple.isSPIRV() &&
-      Triple.getVendor() == llvm::Triple::Intel)
+      Triple.getVendor() == llvm::Triple::Intel &&
+      !Args.hasArg(OPT_embed_bitcode))
     return offloading::intel::containerizeOpenMPSPIRVImage(Img, Triple);
   return Error::success();
 }

offload/test/jit/empty_kernel_lvl1.c

@@ -29,6 +29,5 @@
 // clang-format on
 
 // REQUIRES: gpu
-// XFAIL: intelgpu
 
 #include "empty_kernel.inc"

offload/test/jit/empty_kernel_lvl2.c

@@ -24,6 +24,5 @@
 // clang-format on
 
 // REQUIRES: gpu
-// XFAIL: intelgpu
 
 #include "empty_kernel.inc"