kparzysz-quic commented on pull request #6261: URL: https://github.com/apache/incubator-tvm/pull/6261#issuecomment-676308315
The reasons why prefetch and parallel are not supported in this PR are: 1. The prefetch on Hexagon that we want to use is L2 rectangular prefetch (`l2fetch`), which works differently than the TVM's prefetch (at least at the time when we first added it). We made some changes to how TVM's prefetch works to accomodate it, and upstreaming that would require more discussion with the community. 2. There is a bug in the parallelization code that I haven't had time to upstream a fix for. The fix that we have locally is not quite robust, so I'd have to develop a new one for upstreaming. We wanted to get some basic Hexagon codegen working first, hence this PR. We definitely plan to address the issues I mentioned above, it will just happen later. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
