I was tuning a Matmul op with a specified shape, but Cublas still presented a slightly better performance than TVM tuned. If I prefer superior performance at this Matmul op, is there any suggestions that I could go deeper tuning?
I was tuning a Matmul op with a specified shape, but Cublas still presented a slightly better performance than TVM tuned. If I prefer superior performance at this Matmul op, is there any suggestions that I could go deeper tuning?