FrozenGene commented on issue #4747: [ThreadPool] Solve ARM BIG.LITTLE 
heterogeneous multicores
URL: https://github.com/apache/incubator-tvm/pull/4747#issuecomment-576376485
 
 
   > Overall I like the idea of binding master thread to the set of cores 
needed rather than a single one, whichi means free migration of master thread, 
would be great if we can document this idea in the code, let others to check.
   
   Newest code has provided one detail example to show which cores the master 
thread could run in the comment. Could check it and to see whether to have 
anything should supply. 
   
   > Given that pthread_atfork is indeed available in most cases, we should put 
that as a separate item(perhaps using macro to detect the existence) if we 
still decides to use it.
   > For example, we can keep the existing behavior of binding master thread on 
linux, but use the new behavior on android
   
   pthread_atfork is not a must for solving this problem in fact. The key is we 
should let master thread could do free migration on available cores. Previous 
pr wants to do but provide cpu cores set wrong as described. Existing behavior 
of linux is still has bug if we deploy it on one arm (has big little cpus) 
linux embed board. So we should resolve linux / android together in my opinion. 
:-)
   

----------------------------------------------------------------
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]


With regards,
Apache Git Services

Reply via email to