[GitHub] [tvm] masahi commented on pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort

2021-01-12 Thread GitBox 


masahi commented on pull request #7195:
URL: https://github.com/apache/tvm/pull/7195#issuecomment-759241934


   Thanks @mbrookhart @trevor-m 



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:
us...@infra.apache.org

[GitHub] [tvm] masahi commented on pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort

2021-01-11 Thread GitBox 


masahi commented on pull request #7195:
URL: https://github.com/apache/tvm/pull/7195#issuecomment-758343079


   @Laurawly @kazum This is ready to go, please have a look



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:
us...@infra.apache.org

[GitHub] [tvm] masahi commented on pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort

2021-01-04 Thread GitBox 


masahi commented on pull request #7195:
URL: https://github.com/apache/tvm/pull/7195#issuecomment-754199021


   @mbrookhart I have a fast-path for one segment case, so the perf is the same 
between current / new. I'll update the condition to work for dimension other 
then two.
   
https://github.com/apache/tvm/blob/26254f522de531569441eac4fecb45885fcdc30a/src/runtime/contrib/thrust/thrust.cu#L57
   
   @trevor-m Yes I briefly looked at cub's segmented sort. My impression is 
that it launches one thread block per segment. This sounds great when there are 
many segments to sort and each of segment is not so big. I'm not sure if that 
is a good fit for our use case - I think we are more likely to sort a few, but 
large segments, and most likely we only have one segment. I'm actually 
surprised to hear that TRT uses cub's segmented sort.
   



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:
us...@infra.apache.org