After running some benchmarks with ConjugateGradient (which involves lots of lightweight parallel-for's), I've noticed that the performance of Parallel.For and Parallel.ForReduction are really bad. I plan on doing a heavy optimization session this weekend, but I'm opening this issue just so it's on record.
After running some benchmarks with
ConjugateGradient
(which involves lots of lightweight parallel-for's), I've noticed that the performance ofParallel.For
andParallel.ForReduction
are really bad. I plan on doing a heavy optimization session this weekend, but I'm opening this issue just so it's on record.