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

feature: DTVB with Swizzling (tensorB) #1562

Open
wants to merge 3 commits into
base: develop
Choose a base branch
from

Conversation

solaslin
Copy link
Contributor

@solaslin solaslin commented Jan 16, 2025

Resolved SWDEV-509997

  • Code-gen of Swizzled (SwizzleTensorB): Requires TN/NN and DTVB
  • FP16, BF16 MFMA_16x16x16_x1. Supported useBias and SAV
  • Included supporting for edge tile and tail-loop
  • Implemented for SwizzledB when WaveGroups[0] > 1 (M-Dim)
  • Support arbitrary M & K for SwizzledB
  • pytests for SwizzledB

[gw1] [ 50%] PASSED Tensile/Tests/common/test_config.py::test_config[Tensile/Tests/self_test/swizzleB.yaml]
[gw0] [100%] PASSED Tensile/Tests/common/test_config.py::test_config[Tensile/Tests/self_test/dtl.yaml]
=== 2 passed, 5 warnings in 449.50s (0:07:29) ===
py310: OK (592.75=setup[10.28]+cmd[0.58,132.13,449.76] seconds)
congratulations :) (592.86 seconds)

@solaslin solaslin added the noCI Disable testing on supported CI systems: math libraries CI has this feature enabled.. label Jan 16, 2025
@solaslin solaslin self-assigned this Jan 16, 2025
@solaslin solaslin force-pushed the swizzledB branch 3 times, most recently from 0d524ab to 5bc3ae7 Compare January 21, 2025 06:43
@solaslin solaslin added enhancement New feature or request and removed noCI Disable testing on supported CI systems: math libraries CI has this feature enabled.. labels Jan 23, 2025
@solaslin solaslin marked this pull request as ready for review January 23, 2025 03:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant