jdoerfert wrote:

I think we should forward this and then evaluate the 2 kernel reductions.
I left plenty of comments and questions but overall this looks very sensible to 
me.
Since there are a few questions, I would like to see the revised version before 
approving it.

Please explain in the commit message what the "functional change" is.
Something like:
Instead of a fixed size global buffer that is time-shared across blocks, we now 
use a dynamically sized global buffer that does not require time sharing. This 
results in faster execution time but also larger memory footprint, especially 
for array reductions. Users are advised to limit the number of teams if they 
perform array reductions and memory is a concern.

That all said, **thank you** for picking this up.

https://github.com/llvm/llvm-project/pull/195102
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to