Sasha Rush

I think we need a couple long term fixes. 1) figure out why large scaling ops break the degenerate check in affine. 2) make `image` backend specific. Diagrams also has...

We don't need to merge. I'll fix Cairo for real here. Mostly just sent this to remind myself since I needed it in my code.

https://github.com/srush/annotated-mamba/issues/1

Let me see if I can find a workaround, otherwise, sure, that would be fun.

Here's an in-memory reverse hack that seems to work for me. Unfortunately with `tl.dot` I get a segfault. (any recs for debugging those?) ```python L = 32 @triton.jit def reverse(x_ptr,...

Just to be clear the `tl.dot` version is a 2D mat mul. (code below that segfaults) But yes I agree that a `reverse=True` is the best way. Unfortunately I am...

Sorry, shouldn't have called it a hack. Here's an explanation of what it is doing.  However as @Jokeren notes my method requires creating an B x L x L...

What's your version? I was using nightly.

Nice. Yeah I think at this point it is okay to put out answers. Was thinking of making a youtube walkthrough video as well. Any issues? Happy for feedback or...

It looks like numba has prelim support for ROC and SYCL : https://intelpython.github.io/numba-dpex/latest/user_guides/kernel_programming_guide/writing_kernels.html If you want to send a PR that supports these backends too I would be appreciative. In...