Open lefticus opened 1 month ago
Various versions being tested is here:
https://github.com/lefticus/cpp_weekly/tree/master/parallel_algorithms
Data: https://github.com/lefticus/cpp_weekly/blob/master/parallel_algorithms/performance_stats.csv
So far it looks like the fact that I used signed indexes for some of the code resulted in clang losing its mind, which, if you know anything about the history of the "used signed for performance" stuff, should be rather ironic.
I also need to run callgrind tests against the different versions and see what shakes out.
Related: Alex Dathskovsky's (https://www.youtube.com/@cppnext-alexd) talk "To Int or to Uint?" at CppNorth 2024, to be repeated at CppCon 2024 https://cppcon2024.sched.com/event/1gZg0/to-int-or-to-uint-this-is-the-question
This is a follow-up to the #26 issues that were reported with clang producing code 1/2 the speed of gcc.