On 17.01.2020 12:13, Sven Schreiber wrote:

I varied the environment variables OMP_NUM_THREADS and
OPENBLAS_NUM_THREADS:

OMP...            OPENBLAS...       best of 3 runs
<unset/default>   <unset/default>   6.4 s
4                 4                 6.4 s
4                 1                 6.4 s
1                 4                 2.9 s
1                 1                 2.9 s

Hi Sven again,

my machine is still: 4 Hyper-Threaded Core i7-8550U CPU @ 1.80GHz, but
this time we use new Allin's compiler/linker settings for Debian-like
system to force linking against OpenBLAS (0.3.7 + OpenMP).

My results:

OMP...            OPENBLAS...       best of 3 runs
<unset/default>   <unset/default>   2.067
4                 4                2.83
4                 1                2.83
1                 4                2.81
1                 1                2.80
------------------------------------------
8                 8                2.86
8                 4                2.81
4                 8                2.86
8                 1                2.81
1                 8                2.87
------------------------------------------

Thanks Marcin, that's looking very good! I guess you're using current git for that?

But openblas alone doesn't explain the equality of the 4-thread and the single-thread outcome, at least comparing to the Windows results. Or did you compile openblas yourself? (Perhaps with a different dgemm threshold?)

thanks

sven

I'm attaching shell script to run test for all possibilities.

Cheers,
Marcin

_______________________________________________
Gretl-devel mailing list -- gretl-devel@gretlml.univpm.it
To unsubscribe send an email to gretl-devel-leave@gretlml.univpm.it
Website: https://gretlml.univpm.it/postorius/lists/gretl-devel.gretlml.univpm.it/