Is there a way to describe intrinsics for tensorization with an optional bias? For example, I may want to offload a dense layer or a dense layer with bias to my hardware. Currently, I have one intrinsic for each case, but they are about 95% identical. The only difference is the T.match_buffer and of course the actual bias addition operation.
As this leads to a lot of code duplication I was curious if there is a more elegant way to handle this?
