Add convert-vector-to-llvm pass to the pipeline.

`convert-vector-to-llvm` pass uses some `vector` transform passes
that is not added by `convert-to-llvm`, hence this pass is necessary
for proper lowering of vector ops. As not all vector ops are directly
converted to llvm, some ops are gradually lowered to other vector ops
and those ops are in turn covnerted to llvm.

Author: mshahneo

mlir/lib/Dialect/GPU/Pipelines/GPUToXeVMPipeline.cpp

@@ -85,7 +85,6 @@ void buildGPUPassPipeline(OpPassManager &pm,
     pm.addNestedPass<gpu::GPUModuleOp>(
         createConvertGpuOpsToLLVMSPVOps(gpuToLLVMSPVOptions));
   }
-  pm.addNestedPass<gpu::GPUModuleOp>(createConvertToLLVMPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createCSEPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createReconcileUnrealizedCastsPass());
 }
@@ -105,6 +104,7 @@ void buildPostGPUCommonPassPipeline(
     pm.addPass(createGpuToLLVMConversionPass(gpuToLLVMOptions));
   }
   pm.addPass(createLowerAffinePass());
+  pm.addPass(createConvertVectorToLLVMPass());
   pm.addPass(createConvertToLLVMPass());
   pm.addPass(createReconcileUnrealizedCastsPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createCanonicalizerPass());