[mlir][SerializeToHsaco] Minimize dependencies of AMDGPU compilation

[krzysz00]

Summary

The SerializeToHsaco uses functions from ExecutionEngineUtils to set
up LLVM pass pipelines, but does not otherwise depend on the execution
engine (except indirectly via a dependency on IPO). This commit
removes the dependency on the execution engine to prevent
unnecessarily compilations.

mlir/lib/Dialect/GPU/CMakeLists.txt

 if (MLIR_ENABLE_CUDA_CONVERSIONS)
   set(NVPTX_LIBS
     NVPTXCodeGen
     NVPTXDesc
     NVPTXInfo
   )
 endif()
 
 if (MLIR_ENABLE_ROCM_CONVERSIONS)
   set(AMDGPU_LIBS
     IRReader
+    IPO
     linker
     MCParser
     AMDGPUAsmParser
     AMDGPUCodeGen
     AMDGPUDesc
     AMDGPUInfo
     target
   )
if(MLIR_ENABLE_ROCM_CONVERSIONS)
   target_compile_definitions(obj.MLIRGPUTransforms
     PRIVATE
     __DEFAULT_ROCM_PATH__="${DEFAULT_ROCM_PATH}"
     MLIR_GPU_TO_HSACO_PASS_ENABLE=1
   )
 
   target_link_libraries(MLIRGPUTransforms
     PRIVATE
-    MLIRExecutionEngine
     MLIRROCDLToLLVMIRTranslation
   )
 endif()