Nanmur opened a new pull request, #19996:
URL: https://github.com/apache/tvm/pull/19996

   ## Motivation
   
   Fixes #19976.
   
   `VerifyGPUCode` runs a lowering pipeline while EvolutionarySearch applies 
postprocessors in parallel. On Windows, concurrent executions of this pipeline 
can terminate the process during candidate generation.
   
   ## Changes
   
   - Serialize the `VerifyGPUCode` lowering pipeline on Windows.
   - Keep the existing parallel behavior on other platforms.
   - Add a Windows-only CUDA Conv2D regression test that exercises the 
concurrent EvolutionarySearch path.
   
   ## Validation
   
   - The regression test terminates with `0xC0000374` before the fix and passes 
after the fix.
   - Candidate generation completed 40/40 runs across `num_threads=1`, `2`, 
`4`, and `physical`.
   - Related MetaSchedule tests: `16 passed`.
   - A one-trial CUDA tuning run completed Builder and LocalRunner measurement 
successfully.


-- 
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