tqchen commented on PR #20062:
URL: https://github.com/apache/tvm/pull/20062#issuecomment-5108862026

   Published CI-recovery head `52557a213b`, preserving the exact mechanical 
commit `f03ec06a7a` and amending only the semantic follow-up.
   
   This round closes the remaining typed-buffer failure boundaries: opaque 
`buffer_data` identities are preserved/remapped through software pipelines, 
warp lowering, flattening, FP8 legalization, and merged shared-memory views; 
shared aliases now extend their physical allocation liveness through actual 
accesses; C/CUDA codegen keeps the established physical pointer names without 
redundant aliases; and LLVM debug emission derives the debug type from the 
logical buffer pointer type. The C++ visitor expectation now includes the 
dependent read/write BufferType shapes.
   
   Validation on the published head:
   - LLVM build completed.
   - 400/400 LLVM codegen, BF16 legalization, and host/device split tests 
passed.
   - 34/34 focused flattening, intrinsic lowering, FP8 legalization, and 
shared-memory merge tests passed, including new opaque-access and 
alias-liveness regressions.
   - software-pipeline and PTX async-copy suites: 20 passed, 6 environment 
skips.
   - compile-only CUDA lowering passed for all five MMA numeric variants, both 
meta-schedule MMA GEMMs, multidimensional PTX async copy, and the buffer-print 
suite.
   - exact CUDA source checks passed for cp.async and aligned/unaligned pointer 
naming.
   - Ruff and `git diff --check` passed.
   
   Fresh CI is running on this head; no merge is requested.


-- 
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]

Reply via email to