Eric Shi comments

Results 89 comments of


                                            Eric Shi

[BUG] Warp does not auto-initialize when working with half precision

Moving to backlog for now as this has the potential to add a lot of overhead.

[QUESTION] Why `Warp`'s tile-based matmul is much slower than torch's one?

Hi @chaoming0625, there are various improvements on the way to close the performance gap between cuBLAS and cuBLASDx. Could you please share complete details about your benchmark so that we...

[QUESTION] Why `Warp`'s tile-based matmul is much slower than torch's one?

Thanks! Will come back to this thread when we have an update on performance.

[QUESTION] Why `Warp`'s tile-based matmul is much slower than torch's one?

- There was a performance regression introduced at the time you were running the benchmark that was fixed before 1.6.0 was published (42812b58fa592b2a73e6ea238bdbc4853b9a782b). - I also made a minor update...

[BUG] Inconsistent sign + inaccuracy in SDF computation

Thank you @ehsanhaghighat for the detailed bug report. @AnkaChan will look into this issue. It might help to provide links to the assets you used for the figures.

[QUESTION] Gradients wrt. to vertex position in FEM simulation

Are you asking about `warp.fem` specifically?

feat: disable hashing for modules and kernels to allow loading from ptx and so

Hi @YuyangLee, please sign off the commit as described by the GitHub action: https://github.com/NVIDIA/warp/pull/661/checks?check_run_id=40703124341 This is required before we can accept this.

feat: disable hashing for modules and kernels to allow loading from ptx and so

@YuyangLee: I think if you can get this pull request to pass the DCO checker, we can handle making additional changes on top of your work to get it into...

[BUG] Gradient compilation error for nested functions with custom gradient

Thanks for reporting this and providing a simple repro!

[BUG] Gradient compilation error for nested functions with custom gradient

Hi @rbregier, apologies for the late reply. A fix for this issue in 6695e68716751cfe82384b0ea3ff4371e56529df was just merged into the `main` branch and will be a part of the v1.10.1 release.