Replace VM register arrays with contiguous stack (+28% bytecode throughput)#235
Merged
Conversation
Replace TGocciaRegisterArray (dynamic array) with a contiguous stack and typed pointer (PGocciaRegister) for the VM register file. Same for local cells. This eliminates FPC's automatic refcount management (FPC_DYNARRAY_ASSIGN, FPC_DYNARRAY_CLEAR, FPC_DYNARRAY_INCR_REF, FPC_DYNARR_SETLENGTH, FPC_FINALIZE) which profiling showed consumed 16.6% of production execution time. Design: - FRegisterStack: single growing TGocciaRegisterArray (initial 4096 slots) - FRegisterBase: integer index into the stack for the current frame - FRegisters: PGocciaRegister pointer (= @FRegisterStack[FRegisterBase]) - Save/restore saves integer indices, not array references — no refcounting - All 507 FRegisters[X] access sites work unchanged (pointer indexing) - Same pattern for FLocalCellStack / FLocalCells Measured impact (production build, fib(20) bytecode): - Before: 405 ns/call (median of 10 runs) - After: 291 ns/call (median of 10 runs) - Improvement: 28% Profiler confirmation (macOS sample tool): - FPC_DYNARRAY_CLEAR: 103 samples → 0 - FPC_DYNARRAY_ASSIGN: 93 samples → 0 - FPC_DYNARRAY_INCR_REF: 43 samples → 0 - FPC_DYNARR_SETLENGTH: 52 samples → 0 - FPC_FINALIZE: 57 samples → 0 - Total execution: 2100ms → 1605ms (-24%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughThe PR refactors register and local cell management in the VM to use preallocated growable stacks with base/count offset tracking and pointer windows instead of dynamic arrays, with pointer arithmetic support enabled via compiler directives. Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Suggested labels
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Comment |
Contributor
Benchmark Results274 benchmarks Interpreted: 🟢 21 improved · 🔴 134 regressed · 119 unchanged · avg -1.7% arraybuffer.js — Interp: 🟢 8, 🔴 3, 3 unch. · avg +1.0% · Bytecode: 🟢 5, 🔴 9 · avg +0.1%
arrays.js — Interp: 🔴 9, 10 unch. · avg -1.3% · Bytecode: 🟢 9, 🔴 10 · avg +0.3%
async-await.js — Interp: 🟢 2, 4 unch. · avg +1.3% · Bytecode: 🟢 1, 🔴 4, 1 unch. · avg -3.6%
classes.js — Interp: 🔴 20, 11 unch. · avg -1.6% · Bytecode: 🟢 16, 15 unch. · avg +5.5%
closures.js — Interp: 🟢 1, 🔴 4, 6 unch. · avg -1.3% · Bytecode: 🟢 10, 🔴 1 · avg +11.8%
collections.js — Interp: 🔴 11, 1 unch. · avg -3.5% · Bytecode: 🟢 2, 🔴 9, 1 unch. · avg -8.7%
destructuring.js — Interp: 🔴 14, 8 unch. · avg -2.6% · Bytecode: 🟢 9, 🔴 2, 11 unch. · avg +2.2%
fibonacci.js — Interp: 🔴 2, 6 unch. · avg -1.2% · Bytecode: 🟢 6, 2 unch. · avg +10.2%
for-of.js — Interp: 🔴 6, 1 unch. · avg -3.9% · Bytecode: 🔴 5, 2 unch. · avg -2.1%
helpers/bench-module.js — Interp: 0 · Bytecode: 0
iterators.js — Interp: 🔴 14, 6 unch. · avg -3.9% · Bytecode: 🟢 9, 🔴 5, 6 unch. · avg +0.7%
json.js — Interp: 🔴 6, 14 unch. · avg -1.7% · Bytecode: 🔴 17, 3 unch. · avg -6.6%
jsx.jsx — Interp: 🟢 5, 🔴 2, 14 unch. · avg +0.4% · Bytecode: 🟢 15, 6 unch. · avg +3.4%
modules.js — Interp: 🟢 1, 🔴 2, 6 unch. · avg -0.5% · Bytecode: 🟢 7, 🔴 1, 1 unch. · avg +45.9%
numbers.js — Interp: 🟢 1, 🔴 7, 3 unch. · avg -2.4% · Bytecode: 🟢 2, 🔴 9 · avg -2.5%
objects.js — Interp: 🔴 4, 3 unch. · avg -1.6% · Bytecode: 🟢 3, 🔴 2, 2 unch. · avg +1.5%
promises.js — Interp: 🟢 1, 🔴 4, 7 unch. · avg -0.9% · Bytecode: 🟢 1, 🔴 6, 5 unch. · avg -1.4%
regexp.js — Interp: 🔴 6, 5 unch. · avg -2.7% · Bytecode: 🟢 1, 🔴 8, 2 unch. · avg -3.8%
strings.js — Interp: 🔴 3, 8 unch. · avg -1.8% · Bytecode: 🔴 10, 1 unch. · avg -6.6%
typed-arrays.js — Interp: 🟢 2, 🔴 17, 3 unch. · avg -2.5% · Bytecode: 🟢 12, 🔴 6, 4 unch. · avg +1.7%
Measured on ubuntu-latest x64. Benchmark ranges compare cached main-branch min/max ops/sec with the PR run; overlapping ranges are treated as unchanged noise. Percentage deltas are secondary context. |
Contributor
Suite Timing
Measured on ubuntu-latest x64. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
This was referenced Apr 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
TGocciaRegisterArray(dynamic array) allocation with a contiguous register stack andPGocciaRegistertyped pointerProblem
Profiling the production bytecode VM with
macOS samplerevealed that FPC's automatic dynamic array refcounting consumed 16.6% of execution time:FPC_DYNARRAY_CLEARFPC_DYNARRAY_ASSIGNFPC_DYNARRAY_INCR_REFFPC_DYNARR_SETLENGTHFPC_FINALIZEEvery function call saved/restored
FRegisters(a dynamic array), triggering 4+ refcount operations. For fib(20) with 21,891 calls, that's ~87,000 refcount operations.Solution
Replace the per-call dynamic array with a single growing stack:
Save/restore becomes integer index copies (zero refcounting):
All 507
FRegisters[X]access sites in the dispatch loop work unchanged — pointer indexing has the same syntax as array indexing in FPC.Measurements (production build, macOS AArch64)
Profiler confirmation — all dynamic array functions eliminated:
FPC_DYNARRAY_CLEARFPC_DYNARRAY_ASSIGNFPC_DYNARRAY_INCR_REFFPC_DYNARR_SETLENGTHFPC_FINALIZETest plan
🤖 Generated with Claude Code