Closed frankmcsherry closed 2 months ago
This is substantially faster than master
on large consolidation benchmarks, and seems comparable to master
or a little better on examples/bfs
benchmarks. The improvement is within measurement error, but has been consistently on the side of "improvement".
Consolidation previously used an
unsafe
implementation because it improved performance overslice::split_at_mut
. This no longer appears to be the case, and also unsafe code is bad and probably doesn't work.At the same time, the current consolidation also appears to be almost 2x as slow as a version from @antiguru that copies from one container to another, and perhaps the in-place consolidation results in loop-carried dependencies or something like that. There is certainly some more to look into here, especially as consolidation is such a common event.