Benchmarking mir.ndslice + lubeck against numpy and Julia

Joseph Rushton Wakeling joseph.wakeling at webdrake.net
Mon Jan 13 13:44:15 UTC 2020


On Monday, 13 January 2020 at 13:34:40 UTC, jmh530 wrote:
> However, mir-glas is intended to be an alternative to blas,
> though I don't think much work has been done recently.

Yes, in principle that's what `mir-glas` is supposed to offer, 
but the current the benchmarks only cover lubeck (which doesn't 
use `mir-glas`), and as you say, `mir-glas` itself is not 
feature-complete when it comes to linear algebra APIs.

If anyone does get round to trying to benchmark functionality 
from `mir-glas`, note that you will have to pay quite a lot of 
attention to optimization flags if you want to get best results.

> On 3, I didn't have time to look into the Julia results, but 
> someone above made the comment that Julia was optimizing away 
> the calculation itself. Dennis also had some interesting points 
> above.

Yes, I would imagine that might also be part of it.  Again, I 
would anticipate that being something where D ought to be able to 
readily implement similar features.


More information about the Digitalmars-d mailing list