[Cin] gcc and avx
phylsmith2017 at gmail.com
Mon Feb 18 17:56:10 CET 2019
A few days ago, GG finally had some time to look into a couple of Andrew's
email about Optimization, CFLAGS, and autovectorization.
The subject "speeding up Cin-GG" through use of some of the aforementioned
techniques would make a great PHD thesis or dissertation (so Andrew is
already half-way there!).
A summary of GG's feedback is:
1) It would be beneficial to improve the speed, and especially those of the
most used plugins.
2) He is concerned that a lot of the speed up potential is too "hardware
architecture specific". (We use mostly AMD here rather than the other
3) For one of his tests, he modified some of the recommended CFLAGS and saw
no improvement for him personally; but keep in mind that we always run in
debug mode so that we are always prepared for a crash while testing.
Running in user mode will more than likely see useful improvements.
It would be very time-consuming to check speed-up improvements for all of
the different Operating Systems - probably best left to individual users on
their specific hardware/software. Maybe some day.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Cin