[PATCH] D82031: [scudo][standalone] Release smaller blocks less often

[Christopher Ferris via Phabricator via llvm-commits]

cferris accepted this revision.
cferris added a comment.
This revision is now accepted and ready to land.

I ran the full set of perf/memory benchmarks and nearly everything has the same memory (RSS) values. The performance is nearly the same two.

There are a couple of outliers:

For 32 bit runs of the std::map and std::unordered_map benchmark of certain sizes (8, 16, 32, 64, 128), the performance is a little worse, but the RSS decreases dramatically. I'm not sure why this might happen though, maybe there is a bug in these benchmarks. But it does seem to track, the extra time is because the memory is being actually released.

For 64 bit, these same benchmarks are the same performance/RSS.

Also, for the memory replays on 32 bit, the time it takes is also a bit higher (worse performance) and the RSS is slightly higher in some runs, but the RSS is within normal variations. For the slight performance loss, I'm not sure why this might be happening, since I would expect the opposite, unless something else is triggering more calls to release the memory on 32 bit.

Overall, this seems to be reasonable, but I am a bit worried about the odd 32 bit performance. Is there an explanation for what could be causing the 32 bit calls to increase?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82031/new/

https://reviews.llvm.org/D82031