jinhongyii opened a new pull request #10163: URL: https://github.com/apache/tvm/pull/10163
Previously in evolutionary search, meta schedule will keep sampling unmeasured schedules until it find `populaion_size `valid schedules in total. However, this algorithm is slow on gpu there will likely be many invalid schedules due to shared memory size constraints and thread number constraints. Hence, we only sample unmeasured schedules until it reaches a minimal size. In our experiments, doing so will not harm the performance, but it will accelerate the whole process a lot. cc: @Hzfengsy @junrushao1994 -- 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]
