Improve the Sparse matrix multiplication computational speed #16187 #16905
Conversation
There was a problem hiding this comment.
@ezyang is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
…#16905) Summary: Instead of converting coo to csr format of the sparse matrix in the original implementation, in my revision I directly use coo format for sparse dense matrix mutliplication. On my linux machine it is 5 times faster than the original code: ``` (original code) SIZE: 15000 DENSITY: 0.01 DEVICE: cpu torch: 0.39403 seconds np: 0.00496674 seconds torch/np: 79.3338 ---------------------------------------- (my update) SIZE: 15000 DENSITY: 0.01 DEVICE: cpu torch: 0.0812583 seconds np: 0.00501871 seconds torch/np: 16.1911 ``` Further code feedback and running time tests are highly welcomed. I will keep revise my code if needed. Pull Request resolved: pytorch/pytorch#16905 Differential Revision: D14020095 Pulled By: ezyang fbshipit-source-id: 4ab94075344a55b375f22421e97a690e682baed5
|
Hi @ezyang, @yf225 @fmassa, my PR is automatically closed. |
|
yes the current commit has landed to master. for more work / changes, open a new PR. thanks! |
Hi @soumith, thank you for your reply. Is it possible that I can have more feedback on my code? both of the error reports of internal tests involve cuda, but I only change the cpu part implementation. I didn't get very useful feedback from the internal tests. |
|
Hi @musikisomorphie; our CI is a bit flaky, so one of the things I do before merging is check the failure and see if they are valid or not. In your case, the CUDA errors were flaky and not related to your PR, so I merged. |
Thanks @ezyang, I am glad my code is merged. I misunderstood what @soumith said, I will keep contributing to pytorch. it is quite fun. |
4 participants
Instead of converting coo to csr format of the sparse matrix in the original implementation, in my revision I directly use coo format for sparse dense matrix mutliplication.
On my linux machine it is 5 times faster than the original code:
Further code feedback and running time tests are highly welcomed. I will keep revise my code if needed.