jinhongyii opened a new pull request, #20068: URL: https://github.com/apache/tvm/pull/20068
## Motivation and context The tcgen05 register fragment layouts did not fully describe the physical placement of Layout F sub-slabs, and the default Layout D placement could drift from the public datapath layout factory. ## Changes - Describe both TMEM Layout F sub-slabs and propagate the selected physical half-slab to tcgen05 `.16x*b` load/store rows. - Derive the default Layout D placement from the public datapath layout factory. - Correct canonical tcgen05 atom layout ordering used by register elementwise dispatch. - Document the supported layouts and sub-slab behavior. ## Testing - B200 Layout F mapping and both sub-slab readback regressions - Canonical tcgen05 atom-layout and elementwise dispatch regressions - Changed-files pre-commit checks -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
