Skip to content

feat: support tile vector/cube#134

Merged
hellozmz merged 4 commits intomainfrom
zmz/fa_base
Jan 9, 2026
Merged

feat: support tile vector/cube#134
hellozmz merged 4 commits intomainfrom
zmz/fa_base

Conversation

@hellozmz
Copy link
Collaborator

@hellozmz hellozmz commented Jan 9, 2026

在ascend中,支持ascendnpu-ir中的优化功能:

  1. limit_auto_multi_buffer_only_for_local_buffer
  2. set_workspace_multibuffer
  3. tile_mix_vector_loop
  4. tile_mix_cube_loop

指令流水中已经展示出来,将vector拆分成了4份(开启了double_buffer,所以整体看是8份):
image

性能比较(seqlen=64k)
image

@CLAassistant
Copy link

CLAassistant commented Jan 9, 2026

CLA assistant check
All committers have signed the CLA.

@hellozmz hellozmz marked this pull request as ready for review January 9, 2026 08:34
@hellozmz hellozmz changed the title support tile loop support tile vector/cube Jan 9, 2026
@hellozmz hellozmz changed the title support tile vector/cube feat: support tile vector/cube Jan 9, 2026
@hellozmz hellozmz added the ascend label Jan 9, 2026
@hellozmz hellozmz merged commit dcb9dc9 into main Jan 9, 2026
5 checks passed
@hellozmz hellozmz deleted the zmz/fa_base branch January 9, 2026 09:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants