Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

in group then group #1735

Merged
merged 2 commits into from
Jan 31, 2024
Merged

in group then group #1735

merged 2 commits into from
Jan 31, 2024

Conversation

DenisYaroshevskiy
Copy link
Collaborator

@DenisYaroshevskiy DenisYaroshevskiy commented Jan 28, 2024

  • in group then group.

The common group of shuffles is: shuffle in groups, then shuffle groups.
Example is avx2 shuffle 128 byte lanes and then intermix them somehow.

There is a question: what to do first: do you shuffle big groups first or in big groups first.
Techincally: big groups first is strictly more powerful.

However, can lead to better 0ing.

It's not an obvious thing at all and I suspect we will come back to this one again.

Now I implemented "in groups then groups" to solve avx2 slide (partially.

  • asimd

for asimd we used to have purely vextq_u16 based solution.
Now this will prefer shift_n when that's avaliable, becasue no constants (horray - the system works).

For neon it's not enabled, because there is a bug in level computation: emulation should always return 0 but sometimes it returns 1. Which messes with neon, double.

@DenisYaroshevskiy DenisYaroshevskiy merged commit 51216c0 into main Jan 31, 2024
36 checks passed
@DenisYaroshevskiy DenisYaroshevskiy deleted the slide_2 branch January 31, 2024 18:38
jtlap pushed a commit that referenced this pull request May 12, 2024
* in group + asimd

* disable accidental sve
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