Skip to content
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.

Sign up for GitHub

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

Jump to bottom

rav1e compilation time regression (4s -> >1hr) on LLVM 17 #116022

Closed
FauxFaux opened this issue Sep 21, 2023 · 3 comments
Closed

rav1e compilation time regression (4s -> >1hr) on LLVM 17 #116022

FauxFaux opened this issue Sep 21, 2023 · 3 comments
Labels
A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. C-bug Category: This is a bug. I-hang Issue: The compiler never terminates, due to infinite loops, deadlock, livelock, etc. regression-from-stable-to-beta Performance or correctness regression from stable to beta.

Comments

@FauxFaux
Copy link

FauxFaux commented Sep 21, 2023 •
edited
Loading

Code

Applications including https://github.com/xiph/rav1e , including the repo's example, never complete compilation in --release. I waited over an hour on my 2019 desktop (i7-8700k).

I tested f8b95d97517a6e223f1c6a90a467188586dd090f.

$ CARGO_INCREMENTAL=0 cargo  beta build --no-default-features --example simple_encoding --release

Version it worked on

It most recently worked on: rustc 1.73.0-nightly (03a119b 2023-08-07)
Stable is fine: rustc 1.72.1 (d5c2e9c 2023-09-13)

rustc 1.73.0-nightly (03a119b0b 2023-08-07)
binary: rustc
commit-hash: 03a119b0b0e310d22d94399b24ed030056050f13
commit-date: 2023-08-07
host: x86_64-unknown-linux-gnu
release: 1.73.0-nightly
LLVM version: 16.0.5

Version with regression

rustc --version --verbose:

rustc 1.73.0-beta.6 (9b9539791 2023-09-17)
binary: rustc
commit-hash: 9b95397915e672e76c43eb013289bca7b9a23370
commit-date: 2023-09-17
host: x86_64-unknown-linux-gnu
release: 1.73.0-beta.6
LLVM version: 17.0.0

rustc 1.73.0-nightly (f88a8b71c 2023-08-08)
binary: rustc
commit-hash: f88a8b71cebb730cbd5058c45ebcae1d4d9be377
commit-date: 2023-08-08
host: x86_64-unknown-linux-gnu
release: 1.73.0-nightly
LLVM version: 17.0.0

I can only assume this is the new LLVM: #114048

@rustbot modify labels: regression-from-stable-to-beta -regression-untriaged

Thoughts

If you add random panics in places, the compilation speeds up. The end of the main loop is a good candidate, this can bring the compile time down to only a few minutes (still a massive regression).

For this, --timings doesn't seem to know what's up: https://b.goeswhere.com/cargo-timing-20230921T083504Z.html

RUSTFLAGS='-Z time-llvm-passes' produces a bunch of output, including this one? I don't know how to read this output and I expect anyone who can is able to reproduce the problem, so I'm not including it in full.

00:01:46 ===-------------------------------------------------------------------------===
00:01:46                           Pass execution timing report
00:01:46 ===-------------------------------------------------------------------------===
00:01:46   Total Execution Time: 104.3365 seconds (101.2781 wall clock)
00:01:46-
00:01:46    ---User Time---   --System Time--   --User System--   ---Wall Time---  --- Name ---
00:01:46   82.7290 ( 86.1%)   7.8338 ( 95.0%)  90.5628 ( 86.8%)  90.5593 ( 89.4%)  GVNPass
00:01:46    5.1964 (  5.4%)   0.2020 (  2.4%)   5.3984 (  5.2%)   3.8898 (  3.8%)  SimpleLoopUnswitchPass
00:01:46    3.2395 (  3.4%)   0.0006 (  0.0%)   3.2401 (  3.1%)   3.2407 (  3.2%)  LoopSimplifyPass
00:01:46    1.3124 (  1.4%)   0.0090 (  0.1%)   1.3214 (  1.3%)   1.3228 (  1.3%)  IndVarSimplifyPass
00:01:46    1.0127 (  1.1%)   0.0333 (  0.4%)   1.0460 (  1.0%)   0.7634 (  0.8%)  InstCombinePass
00:01:46    0.3558 (  0.4%)   0.0043 (  0.1%)   0.3601 (  0.3%)   0.2950 (  0.3%)  SimplifyCFGPass
00:01:46    0.1910 (  0.2%)   0.0047 (  0.1%)   0.1957 (  0.2%)   0.1844 (  0.2%)  SCCPPass
00:01:46    0.1673 (  0.2%)   0.0012 (  0.0%)   0.1685 (  0.2%)   0.1681 (  0.2%)  LoopFullUnrollPass
00:01:46    0.3241 (  0.3%)   0.0189 (  0.2%)   0.3430 (  0.3%)   0.1667 (  0.2%)  LICMPass
00:01:46    0.1501 (  0.2%)   0.0019 (  0.0%)   0.1520 (  0.1%)   0.1510 (  0.1%)  LoopIdiomRecognizePass
00:01:46    0.2368 (  0.2%)   0.0145 (  0.2%)   0.2514 (  0.2%)   0.1318 (  0.1%)  LoopSimplifyCFGPass
00:01:46    0.1849 (  0.2%)   0.0208 (  0.3%)   0.2057 (  0.2%)   0.1049 (  0.1%)  LoopInstSimplifyPass
00:01:46    0.1198 (  0.1%)   0.0219 (  0.3%)   0.1416 (  0.1%)   0.0775 (  0.1%)  InlinerPass
00:01:46    0.0270 (  0.0%)   0.0035 (  0.0%)   0.0304 (  0.0%)   0.0301 (  0.0%)  LoopDeletionPass
00:01:46    0.2481 (  0.3%)   0.0184 (  0.2%)   0.2666 (  0.3%)   0.0288 (  0.0%)  SROAPass
00:01:46    0.0330 (  0.0%)   0.0013 (  0.0%)   0.0344 (  0.0%)   0.0237 (  0.0%)  JumpThreadingPass
00:01:46    0.0896 (  0.1%)   0.0058 (  0.1%)   0.0954 (  0.1%)   0.0173 (  0.0%)  EarlyCSEPass
00:01:46    0.0264 (  0.0%)   0.0020 (  0.0%)   0.0284 (  0.0%)   0.0152 (  0.0%)  CorrelatedValuePropagationPass
00:01:46    0.0159 (  0.0%)   0.0004 (  0.0%)   0.0163 (  0.0%)   0.0135 (  0.0%)  LCSSAPass
00:01:46    0.0257 (  0.0%)   0.0013 (  0.0%)   0.0270 (  0.0%)   0.0130 (  0.0%)  LoopRotatePass
00:01:46    0.1138 (  0.1%)   0.0127 (  0.2%)   0.1265 (  0.1%)   0.0105 (  0.0%)  IPSCCPPass
00:01:46    0.0801 (  0.1%)   0.0131 (  0.2%)   0.0931 (  0.1%)   0.0089 (  0.0%)  GlobalOptPass
00:01:46    0.0267 (  0.0%)   0.0027 (  0.0%)   0.0294 (  0.0%)   0.0070 (  0.0%)  PostOrderFunctionAttrsPass
00:01:46    0.0117 (  0.0%)   0.0011 (  0.0%)   0.0127 (  0.0%)   0.0067 (  0.0%)  DSEPass
00:01:46    0.0121 (  0.0%)   0.0009 (  0.0%)   0.0130 (  0.0%)   0.0063 (  0.0%)  ADCEPass
00:01:46    0.0089 (  0.0%)   0.0003 (  0.0%)   0.0092 (  0.0%)   0.0062 (  0.0%)  VectorCombinePass
00:01:46    0.0087 (  0.0%)   0.0007 (  0.0%)   0.0095 (  0.0%)   0.0055 (  0.0%)  BDCEPass
00:01:46    0.0100 (  0.0%)   0.0025 (  0.0%)   0.0125 (  0.0%)   0.0055 (  0.0%)  MemCpyOptPass
00:01:46    0.0489 (  0.1%)   0.0064 (  0.1%)   0.0553 (  0.1%)   0.0051 (  0.0%)  CalledValuePropagationPass
00:01:46    0.0096 (  0.0%)   0.0012 (  0.0%)   0.0109 (  0.0%)   0.0027 (  0.0%)  ArgumentPromotionPass
00:01:46    0.0061 (  0.0%)   0.0004 (  0.0%)   0.0065 (  0.0%)   0.0024 (  0.0%)  MergedLoadStoreMotionPass
00:01:46    0.0042 (  0.0%)   0.0004 (  0.0%)   0.0046 (  0.0%)   0.0024 (  0.0%)  ReassociatePass
00:01:46    0.0046 (  0.0%)   0.0004 (  0.0%)   0.0051 (  0.0%)   0.0019 (  0.0%)  ConstraintEliminationPass
00:01:46    0.0147 (  0.0%)   0.0005 (  0.0%)   0.0152 (  0.0%)   0.0016 (  0.0%)  LowerExpectIntrinsicPass
00:01:46    0.0115 (  0.0%)   0.0011 (  0.0%)   0.0126 (  0.0%)   0.0015 (  0.0%)  RequireAnalysisPass<llvm::GlobalsAA, llvm::Module, llvm::AnalysisManager<Module>>
00:01:46    0.0027 (  0.0%)   0.0003 (  0.0%)   0.0029 (  0.0%)   0.0013 (  0.0%)  TailCallElimPass
00:01:46    0.0022 (  0.0%)   0.0004 (  0.0%)   0.0025 (  0.0%)   0.0010 (  0.0%)  AggressiveInstCombinePass
00:01:46    0.0065 (  0.0%)   0.0006 (  0.0%)   0.0071 (  0.0%)   0.0008 (  0.0%)  PromotePass
00:01:46    0.0014 (  0.0%)   0.0002 (  0.0%)   0.0017 (  0.0%)   0.0005 (  0.0%)  RequireAnalysisPass<llvm::ShouldNotRunFunctionPassesAnalysis, llvm::Function, llvm::AnalysisManager<Function>>
00:01:46    0.0027 (  0.0%)   0.0001 (  0.0%)   0.0028 (  0.0%)   0.0005 (  0.0%)  CallSiteSplittingPass
00:01:46    0.0010 (  0.0%)   0.0001 (  0.0%)   0.0011 (  0.0%)   0.0004 (  0.0%)  LibCallsShrinkWrapPass
00:01:46    0.0034 (  0.0%)   0.0003 (  0.0%)   0.0038 (  0.0%)   0.0004 (  0.0%)  InvalidateAnalysisPass<llvm::AAManager>
00:01:46    0.0003 (  0.0%)   0.0000 (  0.0%)   0.0003 (  0.0%)   0.0003 (  0.0%)  DeadArgumentEliminationPass
00:01:46    0.0041 (  0.0%)   0.0000 (  0.0%)   0.0041 (  0.0%)   0.0003 (  0.0%)  InferFunctionAttrsPass
00:01:46    0.0002 (  0.0%)   0.0000 (  0.0%)   0.0002 (  0.0%)   0.0002 (  0.0%)  GlobalDCEPass
00:01:46    0.0054 (  0.0%)   0.0002 (  0.0%)   0.0056 (  0.0%)   0.0002 (  0.0%)  OpenMPOptCGSCCPass
00:01:46    0.0004 (  0.0%)   0.0001 (  0.0%)   0.0005 (  0.0%)   0.0002 (  0.0%)  MoveAutoInitPass
00:01:46    0.0006 (  0.0%)   0.0001 (  0.0%)   0.0006 (  0.0%)   0.0002 (  0.0%)  CoroSplitPass
00:01:46    0.0009 (  0.0%)   0.0001 (  0.0%)   0.0010 (  0.0%)   0.0002 (  0.0%)  CoroElidePass
00:01:46    0.0005 (  0.0%)   0.0001 (  0.0%)   0.0006 (  0.0%)   0.0002 (  0.0%)  SpeculativeExecutionPass
00:01:46    0.0001 (  0.0%)   0.0000 (  0.0%)   0.0001 (  0.0%)   0.0001 (  0.0%)  NameAnonGlobalPass
00:01:46    0.0001 (  0.0%)   0.0000 (  0.0%)   0.0001 (  0.0%)   0.0001 (  0.0%)  InvalidateAnalysisPass<llvm::ShouldNotRunFunctionPassesAnalysis>
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  AnnotationRemarksPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  RequireAnalysisPass<llvm::ProfileSummaryAnalysis, llvm::Module, llvm::AnalysisManager<Module>>
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  OpenMPOptPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  CoroEarlyPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  Annotation2MetadataPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  ForceFunctionAttrsPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  CoroCleanupPass
00:01:46    0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)   0.0000 (  0.0%)  CanonicalizeAliasesPass
00:01:46   96.0900 (100.0%)   8.2465 (100.0%)  104.3365 (100.0%)  101.2781 (100.0%)  Total
@FauxFaux FauxFaux added C-bug Category: This is a bug. regression-untriaged Untriaged performance or correctness regression. labels Sep 21, 2023
@rustbot rustbot added needs-triage This issue may need triage. Remove it if it has been sufficiently triaged. I-prioritize Issue: Indicates that prioritization has been requested for this issue. regression-from-stable-to-beta Performance or correctness regression from stable to beta. and removed regression-untriaged Untriaged performance or correctness regression. labels Sep 21, 2023
@nikic
Copy link
Contributor

nikic commented Sep 21, 2023

Possibly related to #115970. The crate names are similar :)

@nikic nikic added A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. I-hang Issue: The compiler never terminates, due to infinite loops, deadlock, livelock, etc. labels Sep 21, 2023
@FauxFaux
Copy link
Author

Yup, those are the same crate / author / code; that's another application that links the same library.

@FauxFaux
Copy link
Author

Looks like y'all are way ahead of me, let's close this one?

@apiraino apiraino removed I-prioritize Issue: Indicates that prioritization has been requested for this issue. needs-triage This issue may need triage. Remove it if it has been sufficiently triaged. labels Sep 21, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. C-bug Category: This is a bug. I-hang Issue: The compiler never terminates, due to infinite loops, deadlock, livelock, etc. regression-from-stable-to-beta Performance or correctness regression from stable to beta.
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@nikic @FauxFaux @apiraino @rustbot