aarch64 support went in at 3.11.31. There are kernels for DGEMM and SGEMM, and support...
I looked a little at how 3.11.34 tuning is different on this particular problem....
It appears that CombineStructsMM is identical in 3.11.31 and 3.11.34. I am not sure...
Ah. The Cw is assumed in other places to be cacheline aligned. If I add a cacheline...
I found three separate bugs in CombineStructsMM that lead to the possibility of a...
This does not fail in 3.11.32, only in 3.11.31. It fails without any changes to 3.11.31,...
I will spend time reading through your description in more depth tomorrow. But at...
barriers come in 3 flavors. Write, read, and sys. Sys just does both the read and...