[GitHub] [tvm] bfgoldstein opened a new pull request, #11602: [ROOFLINE] Bug Fixes

[GitBox <gi...@apache.org>]

bfgoldstein opened a new pull request, #11602:
URL: https://github.com/apache/tvm/pull/11602

   This PR fixes two bugs in the Roofline code:
   
   - In the estimate_peak_bandwidth function, the threads argument (non-array) was not passed via specialize call;
   
   - In the estimate_peak_bandwidth function, the last dimensions of tensor B mismatch the tensor A. Changing from [threads, vec_width, 4] to [threads, 4, vec_width]. The current version adds cache misses and impacts the final kernel performance.
   
   @tkonolige PTAL
   


-- 
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: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

[GitHub] [tvm] bfgoldstein commented on pull request #11602: [ROOFLINE] Bug Fixes

[GitBox <gi...@apache.org>]

bfgoldstein commented on PR #11602:
URL: https://github.com/apache/tvm/pull/11602#issuecomment-1149089177

   I end up forgetting about the dimension mismatch over the TIR function. The last commit fixes that.


-- 
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: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org