Conversation

Notices

1. Adrian Cochrane (alcinnz@floss.social)'s status on Tuesday, 28-Mar-2023 03:37:04 JST Adrian Cochrane

   GNU Multiprecision cares a lot about performance (and correctness). As such there's a subsystem which uses C macros to tweak various APIs to capture performance metrics. With the core "cycle-counting" code implemented in Assembly for a variety of CPUs to avoid incurring any overhead & generating misleading numbers.

   There's some noops implemented, presumably to test this infrastructure.

   There's utilities implemented around this to eg. try out some magic numbers to find the most optimal one.
   1/2

   • Adrian Cochrane (alcinnz@floss.social)'s status on Tuesday, 28-Mar-2023 03:54:10 JST Adrian Cochrane

     in reply to

     There's also utility commands to perform statistical analysis on the results & render charts via GNU PlotLib.

     There's some shared utility libraries for these commands, including one to tweak CPU settings. And a benchmark suite, with its own testrunner.

     There's perl-script to generate profiled versions of the MPN library.

     And I'm not even going to dig through the testsuite...

     2/2