nikitaved

Results 69 comments of nikitaved

COO intersection primitives: performance improvement

@cpuhrsch , this one is also not safe to merge due to the very same issues with the MSVC compiler that got some other of my PRs reverted... :(

COO intersection primitives: performance improvement

@pytorchbot merge -g

sparse_mask: extend lhs to sparse COO tensors

There is a way to substantially improve performance, not just for this PR, but for the COO intersection kernel overall (no need for `coalesce` at all, much fewer kernel calls)....

sparse_mask: extend lhs to sparse COO tensors

A gentle ping for reviews :)

sparse_mask: extend lhs to sparse COO tensors

Thank you, @pearu , for your review!

sparse_mask: extend lhs to sparse COO tensors

@pytorchbot merge

COO intersection kernel: respect value intersection order

@cpuhrsch , @amjames , @pearu , could you please stamp this one?

COO intersection kernel: respect value intersection order

@pytorchbot merge -g

COO intersection kernel: respect value intersection order

@jeanschmidt , looks like MSVC is having hard times with cpp17 I guess? And/or structured binding from a tuple... Otherwise the code change is trivial.

COO intersection kernel: respect value intersection order

Closing, it is obsolete.