-
Notifications
You must be signed in to change notification settings - Fork 12.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WIP] Always verify incremental hash #84227
Conversation
(rust-highfive has picked a reviewer for you, use r? to override) |
r? @ghost |
@bors try @rust-timer queue |
Awaiting bors try build completion. @rustbot label: +S-waiting-on-perf |
⌛ Trying commit 4c81dce603dbc5cdfc4bae156fca9ecdda12ae88 with merge f4a231fce8df9e7efa79597690860e9a0b28e97f... |
☀️ Try build successful - checks-actions |
Queued f4a231fce8df9e7efa79597690860e9a0b28e97f with parent 2962e7c, future comparison URL. |
Finished benchmarking try commit (f4a231fce8df9e7efa79597690860e9a0b28e97f): comparison url. Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up. @bors rollup=never |
@Aaron1011 what's the difference between this and #83007 ? Did that PR just miss one of the verify calls? |
I would really like to be able to land this, but the perf impact is terrible. I had hoped that switching to HighwayHash would help, but it actually appears to decrease performance: #84229 |
On larger crates the perf impact is only ~5%, so it's not too horrible. I think I'd personally like us to squash known bugs before landing this (e.g., the syn failure), so we're not just accumulating more bugs, but other than that it seems OK. Is there a possibility we can optimize the incremental verification a bit? It looks like it does several things:
At least the first of these seems.. not necessary? Both of the hash table lookups are also with the same key, I think, so we could optimize out the double hashing there with some manual work. I don't know if it makes sense to do a perf run which just computes the hash but doesn't actually compare it - maybe wrapping in black box or something - to check if that's where all of the overhead is coming from. It probably is, but even shaving off a few dozen instructions from each call might make a nice difference here? |
This change would only reveal existing bugs that were currently going undetected. If you think that this PR might cause additional ICEs, then I would argue that that's a reason to land it sooner - those bugs could be causing miscompilations. |
Yeah, that's a good point. FWIW I am personally willing to r+ this PR "as-is" (I'd like to see the description updated with some details on why) -- I do think the performance isn't ideal but it's also not that bad in most cases. |
4c81dce
to
2893057
Compare
@bors try @rust-timer queue |
Awaiting bors try build completion. @rustbot label: +S-waiting-on-perf |
⌛ Trying commit 2893057d60411eab7a939c81f8c081e3a7c61266 with merge d156ee0e47b013725257045a343c9f0863cae288... |
☀️ Try build successful - checks-actions |
Queued d156ee0e47b013725257045a343c9f0863cae288 with parent b0c818c, future comparison URL. |
Finished benchmarking try commit (d156ee0e47b013725257045a343c9f0863cae288): comparison url. Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up. @bors rollup=never |
For clap-rs-check incr-patched this slows down death_check by 55x. |
One alternative to somehow improving the perf is to land this as is but for only one or two nightlies. Then people will use it and report ICEs but it won't be a regression for anyone on stable. |
eed9ea0
to
be102b8
Compare
Yeah, I'm not a big fan of temporary enabling, though we could try it. I did have a thought about switching the hash algorithm for this verification (for example, to fxhash) -- in theory, that should still catch many cases of bugs: we won't error on collisions, but I would expect those to be fairly rare. I suspect this is pretty painful to implement though... I've currently gone for an alternative solution, which is checking only a fraction of the hashes (in expectation, every 256th, though we can alter arbitrarily). This should pretty significantly reduce the cost of verification while still (I think) being deterministic and hopefully catching most bugs over time. We can play with the exact ratio here based on perf results. @bors try @rust-timer queue |
Awaiting bors try build completion. @rustbot label: +S-waiting-on-perf |
⌛ Trying commit be102b8eaeecec51aaa4ce688a1b92e19ef714f1 with merge 25fee14d1c50219333de853ad1d11d901c3f9a8e... |
☀️ Try build successful - checks-actions |
Queued 25fee14d1c50219333de853ad1d11d901c3f9a8e with parent 337e156, future comparison URL. |
Finished benchmarking commit (25fee14d1c50219333de853ad1d11d901c3f9a8e): comparison url. Summary: This change led to moderate relevant mixed results 🤷 in compiler performance.
If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf. Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR led to changes in compiler perf. Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @bors rollup=never |
be102b8
to
59dbb8c
Compare
@bors try @rust-timer queue Dropped the perf opt from this PR so we can more faithfully evaluate this change, and increased the number of affected hashes to 1/32nd. |
Awaiting bors try build completion. @rustbot label: +S-waiting-on-perf |
⌛ Trying commit 59dbb8c with merge 97d5f8ae1b0d7436d56cadb5e405aadafc64a86f... |
☀️ Try build successful - checks-actions |
Queued 97d5f8ae1b0d7436d56cadb5e405aadafc64a86f with parent dd757b9, future comparison URL. |
Finished benchmarking commit (97d5f8ae1b0d7436d56cadb5e405aadafc64a86f): comparison url. Summary: This change led to very large relevant regressions 😿 in compiler performance.
If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf. Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR led to changes in compiler perf. Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @bors rollup=never |
Closing in favor of #90361, which just has the same commits I force pushed here but is a fresh PR for fresh discussion. |
…aelwoerister Enable verification for 1/32th of queries loaded from disk This is a limited enabling of incremental verification for query results loaded from disk, which previously did not run without -Zincremental-verify-ich. If enabled for all queries, we see a probably unacceptable hit of ~50% in the worst case, so this pairs back the verification to a more limited set based on the hash key. Per collected [perf results](rust-lang#84227 (comment)), this is a regression of at most 7% on coercions opt incr-unchanged, and typically less than 0.5% on other benchmarks (largely limited to incr-unchanged). I believe this is acceptable performance to land, and we can either ratchet it up or down fairly easily. We have no real sense of whether this will lead to a large amount of assertions in the wild, but since those assertions may lead to miscompilations today, it seems potentially warranted. We have a good bit of lead time until the next stable release, though the holiday season will also start soon; we may wish to discuss the timing of enabling this and weigh the desire to prevent (possible) miscompilations against assertions. cc `@rust-lang/wg-incr-comp`
No description provided.