[MLIR] [AMDGPU] Use correct flags when building SerializeToHsaco

[krzysz00]

Summary

The SerializeToHsaco pass does not depend on ROCm being available on
the build system - it only requires ROCm to be present at runtime.
However, the CMake file that built it tested for
MLIR_ENABLE_ROCM_RUNNER , which implies that ROCm is currently
available and is used to control building ROCm integration tests.

Referencing MLIR_ENABLE_ROCM_RUNNER instead of
MLIR_ENABLE_ROCM_CONVERSIONS in the SerializeToHsaco build therefore
causes problems for clients who wish to build projects that depend on
this pass on a system without an AMD GPU present.

[whchung]

LGTM.

mlir/lib/Dialect/GPU/CMakeLists.txt

if(MLIR_ENABLE_CUDA_RUNNER)
   target_link_libraries(MLIRGPUTransforms
     PRIVATE
     MLIRNVVMToLLVMIRTranslation
     ${CUDA_DRIVER_LIBRARY}
   )
 
 endif()
 
-if(MLIR_ENABLE_ROCM_RUNNER)
+if(MLIR_ENABLE_ROCM_CONVERSIONS)
   if (NOT ("AMDGPU" IN_LIST LLVM_TARGETS_TO_BUILD))
     message(SEND_ERROR
       "Building mlir with ROCm support requires the AMDGPU backend")
   endif()
 
   set(DEFAULT_ROCM_PATH "/opt/rocm" CACHE PATH "Fallback path to search for ROCm installs")
   target_compile_definitions(obj.MLIRGPUTransforms
     PRIVATE