size(codegen): cross-module regular and $pshape methods retain equivalent full bodies

Aperta
  • Qualcosa non va?
#13,688 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
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
12/100
Tipo di issue
Refactoring
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
rust, typescript
Ambito
compilers

Direzione di ricerca

The retained bodies come from LLVM IR emitted for the ordinary and $pshape variants of Histogram.score and Histogram.text, so start with the code generator that emits specialized method clones and the feedback-site ID handling. Reproduce with the two-module example and compare the .ll output saved via PERRY_SAVE_LL, then check nm -n for the four symbols. Done means identical-bodied clones are shared without changing GC metadata, unwinding, or feedback observations, and the byte savings are measured.

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

Descrizione

Follow-up to #13663. This is a separate code-generation size opportunity from #13659's feature pruning and #13662's runtime optimization profiles. No size-oriented optimization or performance tradeoff is requested.

Public two-module reproduction

histogram.ts:

// Public reduction for regular/$pshape bodies retained with identical logic.
export class Histogram {
  bins: number[] = new Array<number>(8).fill(0);
  count: number = 0;
  add(index: number): void {
    this.bins[index & 7] = (this.bins[index & 7] ?? 0) + 1;
    this.count++;
  }
  score(): number {
    let result = this.count;
    for (let i = 0; i < 8; i++) {
      const value = this.bins[i] ?? 0;
      result += value * (i + 1);
    }
    return result;
  }
  text(): string {
    let result = "";
    for (let i = 0; i < 8; i++) result += String(this.bins[i] ?? 0) + ":";
    return result;
  }
}

main.ts:

import { Histogram } from "./histogram.ts";
const histogram = new Histogram();
for (let i = 0; i < 1000; i++) histogram.add(i);
console.log(histogram.score());
console.log(histogram.text());


mkdir -p llvm
PERRY_RUNTIME_DIR=/path/to/matching/runtime PERRY_LL_OPT_LEVEL=s PERRY_SPECIALIZED_ABI=1 PERRY_KEEP_SYMBOLS=1 PERRY_DISABLE_BUILD_CACHE=1 PERRY_NO_CACHE=1 PERRY_SAVE_LL="$PWD/llvm" perry compile main.ts -o histogram --trace llvm --strict-eval --strict-dynamic-import --strict-unimplemented
nm -n histogram
./histogram
node main.ts

Pin: 2359699bee1ec7b98478a0db44f195a5a086b30d; matching default release runtime/stdlib static archives and GC-effects sidecars; runtime O3 / stdlib Os / ThinLTO, repository unwind settings. macOS 26.5 ARM64. Both programs print:

5500
125:125:125:125:125:125:125:125:

The two-module boundary matters: the equivalent one-module fixture did not retain both copies in the final executable.

Retained bodies

Method Ordinary body bytes $pshape body bytes
Histogram.score 2,364 2,364
Histogram.text 2,792 2,792

All four symbols have different addresses. score and text have identical ordinary/$pshape LLVM bodies (excluding their declaration lines/attributes). Thus 5,156 bytes are represented by an extra copy in this reduction. Byte extents are address deltas within Mach-O __text, with aliases counted once and padding assigned to the preceding symbol; this is potential sharing, not an implemented saving.

The full private app retains 12 ordinary/$pshape pairs with equal per-pair extents, totaling 110,736 bytes for one side. Examples: evaluation 11,432 bytes each, move generation 22,084 each, command handler 31,980 each. The evaluator's IR bodies differ only in two feedback-site IDs; move generation differs in five feedback-site IDs. Other pairs include regex site identities or additional metadata differences, so the entire 110,736-byte upper bound must not be treated as proved safe to merge.

Requested work and acceptance

  • Suppress or share a representation-specialized body when lowering establishes that it performs the same operations as the ordinary body. Keep distinct public entries/metadata where identity or call conventions require them.
  • Evaluate whether semantically equivalent feedback sites can share the implementation without merging observations incorrectly. Do not simply strip all IDs or apply unsafe machine-code folding.
  • Preserve GC statepoint/root metadata, function identity/name/length, stack traces and exception unwinding, inter-module method entry contracts, and generic fallback behavior. Keep genuinely specialized clones when they reduce hot work.
  • Measure fewer retained code bytes with unchanged results and search/evaluation performance. There is no request to apply Oz, disable useful specialization globally, or accept a throughput regression.

The app's generated engine+UI code is about 0.69 MB of the 9.31 MB stripped binary, so this is an incremental opportunity, not a claim that clone sharing alone achieves the overall size target. Existing profile/retention accounting stays in #13663.

Lingua principale
Rust
Stelle
4.9k
Fork
166
Merge medio
3h 46m
PR unite (30g)
891

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 PerryTS/perry

Tutte le issue di PerryTS/perry

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.