On 09/24/2016 10:26 AM, jmh530 wrote:
On Saturday, 24 September 2016 at 13:49:35 UTC, Andrei Alexandrescu wrote:
I see, thanks. To the extent the Python-specific overheads are
measurable, it might make sense to include the benchmark.
Here are some benchmarks from Eigen and Blaze for comparison
http://eigen.tuxfamily.org/index.php?title=Benchmark
https://bitbucket.org/blaze-lib/blaze/wiki/Benchmarks
They don't include Python, for the reason mentioned above (no one would
use native python implementation of matrix multiplication, it just calls
some other library).
I don't see a reason to include it here.
OK. Yah, native Python wouldn't make sense. It may be worth mentioning
that SciPy uses BLAS so it has the same performance profile.
Also, a great idea for a followup would be a blog post comparing the
source code for a typical linear algebra real-world task. The idea
being, yes the D version has parity with Intel, but there _is_ a reason
to switch to it because of its ease of use.
Andrei
