This is the first step to convert vector ops to MMA operations in order to target GPUs tensor core ops. This currently only support simple cases, transpose and element-wise operation will be added later. Differential Revision: https://reviews.llvm.org/D102962