Explanation needed for apparent slowdown in v24.1.0 as compared to v24.0.0 in some benchmarks
Apparently, for some reason, when using the bench_compare_branch.sh
script for v24.0.0 vs master (which at the time was basically almost what is now v24.1.0), this is what @freetrader found:
(This is from slack)
freetrader 8:26 AM
CCheckQueue_RealBlock_32MB_WithCacheStore total time increased 381.27%, speed decreased 79.22%:
File Benchmark evals iterations total min max median
bchn_bench_compare_tmp.0zPPPaZKIi CCheckQueue_RealBlock_32MB_WithCacheStore 5 5 2.38939 0.0398996 0.22884 0.0399992
bchn_bench_compare_tmp.QJts2YyzsZ CCheckQueue_RealBlock_32MB_WithCacheStore 5 5 11.4996 0.383178 0.547711 0.457722
freetrader 8:27 AM
^ was running the bench_compare_branch.sh script from !1453 to compare our master against v24.0.0 , the above seemed a big difference...
I re-ran that one with the --filter option a few times, time increases were always positive and significant:
221.84% ,
410.37%,
506.22%,
392.45%
And also:
freetrader 8:30 AM
The next biggest increase was in PrevectorResizeTrivial which had a time increase of 56.74% (more runs: 54.11%, 53.05%, 55.68%)