Thanks for the information about DGEMM Rahul.
Maybe somebody will comment on the slicing issue at length.
On Fri, 28 Aug 2020, Rahul Ghangas wrote:
Optimizing LinearAlgebra `dot`(specifically matrix-matrix
multiplication) was part of my GSOC project this year.
I noticed at the time but for some reason could not find the details with
my simple Github search.
Engin had previously done some work in this direction for the Intel PRK
test suite for Chapel
Sadly, that does not compile for me against 1.22. It seems to have been
hijacked by the 0-based index brigade. I hope LinearAlgebra is not going
that way.
and we based our implementation on it. The new `dot` has been merged
with master and you can have a look at the merged PR here.
Can you remind me how to grab a copy of master?
Also, I do not quite understand those graph in the Github Issues. What are
they normalized against and to what is the slowness relative?
There's also a performance test for the native implementation of `dot`
which shows similar performance against CrayBLAS gemm. Although, we
haven't done a comparison for much larger matrices.
That graph in the link seems to be showing that your implementation is
getting faster each day but it is not obvious as to the problem or its
details. I do not quite understand how to drive that interface.
Thanks - Damian
Pacific Engineering Systems International, 277-279 Broadway, Glebe NSW 2037
Ph:+61-2-8571-0847 .. Fx:+61-2-9692-9623 | unsolicited email not wanted here
Views & opinions here are mine and not those of any past or present employer
_______________________________________________
Chapel-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-developers
