Skip to content

Copybara import of the project:#10121

Merged
copybara-service[bot] merged 4 commits intomasterfrom
test_907335234
Apr 29, 2026
Merged

Copybara import of the project:#10121
copybara-service[bot] merged 4 commits intomasterfrom
test_907335234

Conversation

@copybara-service
Copy link
Copy Markdown
Contributor

Copybara import of the project:

--
d3121ea by velonica0 like@mail.nankai.edu.cn:

[RVV] add rvv f32 kernels for velu, vgelu, vapproxgelu

Add RVV kernels for three elementwise activation ops:

  • f32-velu (ELU) - rr2_p6 polynomial approximation
  • f32-vgelu (GELU) - rational 12/10 approximation (div + nr variants)
  • f32-vapproxgelu - rational 12/10 approximation (div + nr variants)

Tested on SpacemiT K1 CPU and K3 CPU, both VLEN=256.

--
ae23132 by velonica0 like@mail.nankai.edu.cn:

Alphabetize RVV elementwise entries in cmake/bzl lists

Move f32-velu, f32-vgelu, f32-vapproxgelu entries from end of
PROD/NON_PROD lists to alphabetical positions to avoid merge conflicts
with other RVV PRs.

--
0b6f61a by velonica0 like@mail.nankai.edu.cn:

fix cmake bug

FUTURE_COPYBARA_INTEGRATE_REVIEW=#9963 from velonica0:rvv-elementwise 0b6f61a

velonica0 and others added 4 commits April 20, 2026 11:04
Add RVV kernels for three elementwise activation ops:
- f32-velu (ELU)       - rr2_p6 polynomial approximation
- f32-vgelu (GELU)     - rational 12/10 approximation (div + nr variants)
- f32-vapproxgelu      - rational 12/10 approximation (div + nr variants)

Tested on SpacemiT K1 CPU and K3 CPU, both VLEN=256.
Move f32-velu, f32-vgelu, f32-vapproxgelu entries from end of
PROD/NON_PROD lists to alphabetical positions to avoid merge conflicts
with other RVV PRs.
@copybara-service copybara-service Bot merged commit bbc68d9 into master Apr 29, 2026
@copybara-service copybara-service Bot deleted the test_907335234 branch April 29, 2026 17:29
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