Performance regression in 67d70b4e192e..c2851ff669c5 (are-we-fast-yet run 00087)

Aperta
#15,192 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
44/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
rust, wasm

Direzione di ricerca

Start by reading the linked commit, “Publish handle-table slot pointers in VMComponentContext,” and the regression guidance in docs/regressions.md. Reproduce the callgrind comparison using the supplied sightglass commands and benchmarks/cm-online-stats/cm-online-stats.wasm, then investigate changes between the two Wasmtime revisions. Done means identifying whether the commit caused the execution slowdown and documenting or fixing its cause.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

awfy performance

are-we-fast-yet found 1 regression and 1 speedup in virtual cycles (callgrind) between Wasmtime 67d70b4e192e and c2851ff669c5.

A change counts when a Mann-Whitney U test finds it significant at the 0.01 level and it clears its phase's minimum relative change:

phase geomean individual
compilation 0.3% 1%
instantiation 2.5% 20%
execution 0.5% 2%
Regressions (1)
benchmark phase Δ virtual cycles virtual cycles noise Δ instructions retired instructions retired noise
cm-online-stats/cm-online-stats.wasm execution +12.41% (+12.40% to +12.42%) 0.015% -> 0.008% +12.43% 0.003% -> 0.001%
Speedups (1)
benchmark phase Δ virtual cycles virtual cycles noise Δ instructions retired instructions retired noise
tinygo/tinygo-regex.wasm execution -2.72% (-2.75% to -2.72%) 0.003% -> 0.029% -0.00% 0.001% -> 0.005%

A positive delta is a slowdown, a negative delta is a speedup.

instructions retired counts work done rather than how well it was predicted. A regression whose instructions didn't move is often Valgrind's branch predictors aliasing on a code layout change; see docs/regressions.md.

Each noise column is how far that event's own counts spread.

Other callgrind counters that changed for these benchmarks
benchmark phase counter Δ noise
cm-online-stats/cm-online-stats.wasm execution ll-dcache-read-misses -39.39% (-54.55% to -36.36%) 36.364% -> 10.000%
cm-online-stats/cm-online-stats.wasm execution l1-dcache-read-misses -23.32% (-33.33% to -4.15%) 30.316% -> 23.562%
cm-online-stats/cm-online-stats.wasm execution ll-icache-misses -16.34% (-24.50% to -14.40%) 11.634% -> 3.636%
cm-online-stats/cm-online-stats.wasm execution data-writes +17.28% (+17.28% to +17.28%) 0.002% -> 0.001%
cm-online-stats/cm-online-stats.wasm execution indirect-branch-misses -14.22% (-27.34% to -5.83%) 22.831% -> 10.526%
cm-online-stats/cm-online-stats.wasm execution instructions-retired +12.43% (+12.43% to +12.43%) 0.003% -> 0.001%
cm-online-stats/cm-online-stats.wasm execution conditional-branches +9.09% (+9.08% to +9.09%) 0.005% -> 0.002%
cm-online-stats/cm-online-stats.wasm execution data-reads +6.30% (+6.30% to +6.30%) 0.002% -> 0.000%
Commits

https://github.com/bytecodealliance/wasmtime/compare/67d70b4e192e88faa597008560811dc7e2b1414b...c2851ff669c588563f2362dfd5d5be7518028132

Reproduce locally

On x86-64 Linux with Valgrind installed:

git clone https://github.com/bytecodealliance/sightglass
cd sightglass
git checkout 2d453335318d05cf42b6f3263aa68f0021985f10
rustc engines/wasmtime/build.rs -o build-engine
mkdir old new
REVISION=67d70b4e192e88faa597008560811dc7e2b1414b ./build-engine old
REVISION=c2851ff669c588563f2362dfd5d5be7518028132 ./build-engine new
cargo run --release -- benchmark --measure callgrind \
  --engine old/libengine.so --engine new/libengine.so \
  --engine-flags-common="-Cparallel-compilation=n --target x86_64-unknown-linux-gnu -Ccranelift-has-sse3 -Ccranelift-has-ssse3 -Ccranelift-has-sse41 -Ccranelift-has-sse42 -Ccranelift-has-popcnt -Ccranelift-has-avx -Ccranelift-has-avx2 -Ccranelift-has-fma -Ccranelift-has-bmi1 -Ccranelift-has-bmi2 -Ccranelift-has-lzcnt -Ccranelift-has-cmpxchg16b" \
  -- benchmarks/cm-online-stats/cm-online-stats.wasm
Lingua principale
Rust
Stelle
18.8k
Fork
1.9k
Merge medio
1g 3h
PR unite (30g)
382

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di bytecodealliance/wasmtime

Tutte le issue di bytecodealliance/wasmtime

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.