Skip to content

[Graph] Translate softmax and reduce to hidet script#242

Merged
yaoyaoding merged 12 commits intohidet-org:mainfrom
Aalanli:hidet_script
Jun 2, 2023
Merged

[Graph] Translate softmax and reduce to hidet script#242
yaoyaoding merged 12 commits intohidet-org:mainfrom
Aalanli:hidet_script

Conversation

@Aalanli
Copy link
Copy Markdown
Contributor

@Aalanli Aalanli commented May 23, 2023

Sorry, I did something to my fork that I could not reverse; I thought that deleting the fork would not close the PR.

In regards to the comments:

  • There seems to be no reason for the prologue to be disabled, but the FP16 reduce version does. I re-enabled the prologue
  • implemented suggestions for xdtype and x_shape
  • yes, warp_reduce(rv, prim.sum) works, can't find operator.sum

Everything works as expected.

Copy link
Copy Markdown
Member

@yaoyaoding yaoyaoding left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @Aalanli!

return True
else:
# use self.cuda_schedule_reduce_by_default
return False
Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just looked at the self.cuda_schedule_reduce_by_default(...) implementation, seems it is okay to fuse prologue.

@yaoyaoding
Copy link
Copy Markdown
Member

Thanks @Aalanli !

@yaoyaoding yaoyaoding merged commit 59e2eae into hidet-org:main Jun 2, 2023
@Aalanli Aalanli deleted the hidet_script branch September 27, 2023 18:09
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants