Skip to content

perf(interp): fuse arithmetic-to-local stores and array loads #167

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

Merged
siyul-park merged 2 commits into from
Jul 30, 2026
Merged

perf(interp): fuse arithmetic-to-local stores and array loads #167

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0b90b1f
perf(interp): fuse arithmetic-to-local stores and array-constant loads
siyul-park Jul 30, 2026
baa6b4e
refactor(interp): simplify array header reuse
siyul-park Jul 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
88 changes: 44 additions & 44 deletions docs/benchmarks.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,11 +21,10 @@ For implementation details, see `docs/jit-internals.md`. For profiling counters,

## Measurement Environment

The cross-runtime comparison table was measured on July 30, 2026. Its
`minivm/default` rows were re-measured the same day after loop-carried scalar
write-back using five interleaved baseline/current pairs. The other
cross-runtime rows use three sequential samples. The public API cost tables
below it were measured on July 16, 2026:
The external-runtime rows were measured on July 30, 2026 with three sequential
samples. Every minivm row was re-measured on July 31, 2026 from five
interleaved baseline/current pairs against commit `10d6adf`. The public API
cost tables below were measured on July 16, 2026:
Comment on lines +24 to +27

@coderabbitai coderabbitai Bot Jul 30, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Correct the future-dated measurement claim.

As of July 30, 2026, the document cannot truthfully state that minivm was re-measured on July 31, 2026. Replace it with the actual completed measurement date, or defer this documentation update until those measurements exist.

🧰 Tools
🪛 LanguageTool

[grammar] ~25-~25: Ensure spelling is correct
Context: ...26 with three sequential samples. Every minivm row was re-measured on July 31, 2026 fr...

(QB_NEW_EN_ORTHOGRAPHY_ERROR_IDS_1)

🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/benchmarks.md` around lines 24 - 27, Correct the measurement-date
statement in the benchmark documentation: replace the future-dated July 31, 2026
minivm re-measurement claim with the actual completed date, or remove/defer that
claim until the measurements have been performed. Preserve the surrounding
sample and commit details.


- Apple M4 Pro, 12 cores
- `darwin/arm64`
Expand All @@ -44,13 +43,10 @@ slower there than in a focused single-kernel run.
cd benchmarks
go test -tags=compare -run='^$' -bench='.' -benchmem -benchtime=300ms -count=3 ./...

# Issue #163 A/B: alternate this command between the baseline and current
# checkout five times, then take each side's median.
# Issue #164 A/B: alternate this command between commit 10d6adf and the
# current checkout five times, then take each side's median.
go test -run='^$' \
-bench='^(BenchmarkControl_IterativeFib|BenchmarkControl_Sieve|BenchmarkCall_IndirectRecursiveFib|BenchmarkCall_ClosureCounter|BenchmarkMemory_TypedArraySum|BenchmarkMemory_AllocationGraph|BenchmarkNumeric_BranchTree)$/^default$' \
-benchmem -benchtime=300ms -count=1 .
go test -run='^$' \
-bench='^BenchmarkCall_RecursiveFib$/^(20|35)$/^default$' \
-bench='^(BenchmarkControl_IterativeFib|BenchmarkControl_Sieve|BenchmarkCall_RecursiveFib|BenchmarkCall_IndirectRecursiveFib|BenchmarkCall_ClosureCounter|BenchmarkMemory_TypedArraySum|BenchmarkMemory_AllocationGraph|BenchmarkNumeric_BranchTree)$' \
-benchmem -benchtime=300ms -count=1 .

# Public interpreter and pool API costs
Expand All @@ -73,16 +69,20 @@ go test -run='^$' -bench='^Benchmark(Array|Struct|TypedMap|Map)_Refs$' \

## Summary

- `RecursiveFib(35)` places `minivm/default` at **46.95 ms**, within about **5.6%** of wazero's **44.45 ms**, while remaining allocation-free after warmup.
- Adaptive native traces reduce `IterativeFib(30)` from **746.3 ns** threaded to **36.75 ns**, `TypedArraySum(256)` from **6.313 us** to **307.7 ns**, and `BranchTree(96)` from **952.0 ns** to **264.4 ns**.
- `RecursiveFib(35)` places `minivm/default` at **45.90 ms**, within about **3.3%** of wazero's **44.45 ms**, while remaining allocation-free after warmup.
- Generic threaded improvements close the final three cross-runtime gaps: `IterativeFib(30)` is **484.4 ns** versus gopher-lua's **513.9 ns**, `TypedArraySum(256)` is **2.820 us** versus gopher-lua's **3.413 us**, and `AllocationGraph(128)` is **4.816 us** versus gpython's **5.715 us**.
- Interleaved A/B (median of five) cuts those threaded kernels from **711.5 -> 484.4 ns (-31.9%)**, **6.152 -> 2.820 us (-54.2%)**, and **7.260 -> 4.816 us (-33.7%)**. Every other threaded kernel improves by **14.8-33.2%**; every `default` and `jit` median stays within **1.6%** or improves.
- The gains come from a feature-free threaded dispatch loop, bounded scalar arithmetic-to-local fusion, guarded typed-array constant loads, and reset-time reference-array header reuse. These are opcode-family and lifecycle optimizations rather than fixture-specific superinstructions.
- Reset-time reference-array header reuse halves `AllocationGraph(128)` from **256** to **128 allocs/op** and **5,120** to **1,024 B/op** in every minivm mode. Element backing stores remain fresh, and headers detached through `Pop` are not reused.
- Adaptive native traces reduce `IterativeFib(30)` from **484.4 ns** threaded to **19.69 ns**, `TypedArraySum(256)` from **2.820 us** to **289.2 ns**, and `BranchTree(96)` from **603.8 ns** to **230.9 ns**.
- Loop-carried scalar write-back keeps eligible call-free locals authoritative in registers and commits their VM slots only on native exit paths. Interleaved A/B (median of five) cuts `IterativeFib(30)` **91.75 -> 36.75 ns (-59.9%)**, `TypedArraySum(256)` **673.2 -> 307.7 ns (-54.3%)**, and `Sieve(256)` **2,645 -> 1,626 ns (-38.5%)**. Every other canonical `default` median stays within **0.9%**, and allocation counts are unchanged.
- Fused threaded handlers check stack room once for their own net push instead of once per folded source. That removes about 5,400 generated lines and cuts threaded time on every fusion-heavy kernel: `BranchTree(96)` **-4.3%**, `TypedArraySum(256)` **-4.1%**, `IterativeFib(30)` **-3.6%**, and `Sieve(256)` **-2.8%** (interleaved A/B, median of five). Native and adaptive modes are unchanged within noise.
- Primitive array mutation stays on the native loop path in `Sieve(256)`: deferred-ownership elision drops the per-element retain/release pair, so a runtime-allocated array reaches the same cheap native path a typed-array constant already used. All three modes allocate `1,048 B` in `2` allocations.
- Loop-invariant container hoisting (issue #153) removes the per-access heap-cell derivation, itab guard, and slice-header reload from hoisted loop bodies. It shrinks the loop callables but leaves wall time unchanged: the removed loads sat off the out-of-order critical path.
- Branch-leg folding (issue #155) records native loop exits as branches and folds hot legs that rejoin the header back into the native loop as real back-edges. Combined with loop-carried scalar write-back, `Sieve(256)` now runs at **1.626 us** versus **16.2 us** threaded and **687.3 ns** in wazero.
- Branch-leg folding (issue #155) records native loop exits as branches and folds hot legs that rejoin the header back into the native loop as real back-edges. Combined with loop-carried scalar write-back, `Sieve(256)` now runs at **1.508 us** versus **11.77 us** threaded and **687.3 ns** in wazero.
- Threshold-zero `jit` is not a warmed-JIT guarantee. It matches `default` on Sieve and BranchTree, but is slower on IterativeFib, TypedArraySum, and recursive Fibonacci because it can compile before representative traces are learned.
- Allocation-heavy workloads remain interpreter-bound. `AllocationGraph(128)` is fastest in minivm's threaded mode at **7.774 us**; adaptive and eager modes add profiling cost without native coverage.
- Indirect recursion reaches the native self-call path in adaptive mode: `IndirectRecursiveFib(20)` is **54.85 us** in `default`, versus **576 us** threaded and **43.3 us** in wazero. Eager `jit` stays at **594 us**, consistent with the threshold-zero note above.
- Allocation-heavy workloads remain interpreter-bound. `AllocationGraph(128)` is fastest in minivm's threaded mode at **4.816 us**; adaptive and eager modes add profiling cost without native coverage.
- Indirect recursion reaches the native self-call path in adaptive mode: `IndirectRecursiveFib(20)` is **53.56 us** in `default`, versus **478 us** threaded and **43.3 us** in wazero. Eager `jit` stays at **584 us**, consistent with the threshold-zero note above.

These results are workload measurements, not general language rankings. The runtimes use different value models, safety boundaries, host-call conventions, and compilation strategies.

Expand All @@ -104,89 +104,89 @@ Each minivm kernel times `Interpreter.Run` only. Result extraction, reset, fixtu

| Workload | Runtime | ns/op | B/op | allocs/op |
|---|---|---:|---:|---:|
| IterativeFib(30) | minivm/default | 36.75 | 0 | 0 |
| IterativeFib(30) | minivm/threaded | 746.3 | 0 | 0 |
| IterativeFib(30) | minivm/jit | 95.07 | 0 | 0 |
| IterativeFib(30) | minivm/default | 19.69 | 0 | 0 |
| IterativeFib(30) | minivm/threaded | 484.4 | 0 | 0 |
| IterativeFib(30) | minivm/jit | 34.30 | 0 | 0 |
| IterativeFib(30) | native Go | 9.256 | 0 | 0 |
| IterativeFib(30) | wazero | 52.63 | 8 | 1 |
| IterativeFib(30) | Tengo | 9,406 | 90,592 | 61 |
| IterativeFib(30) | gopher-lua | 513.9 | 160 | 0 |
| IterativeFib(30) | Goja | 2,226 | 368 | 20 |
| IterativeFib(30) | gpython | 2,599 | 2,448 | 88 |
| IterativeFib(30) | Yaegi | 2,856 | 2,036 | 101 |
| Sieve(256) | minivm/default | 1,626 | 1,048 | 2 |
| Sieve(256) | minivm/threaded | 16,239 | 1,048 | 2 |
| Sieve(256) | minivm/jit | 2,696 | 1,048 | 2 |
| Sieve(256) | minivm/default | 1,508 | 1,048 | 2 |
| Sieve(256) | minivm/threaded | 11,770 | 1,048 | 2 |
| Sieve(256) | minivm/jit | 1,508 | 1,048 | 2 |
| Sieve(256) | native Go | 237.2 | 0 | 0 |
| Sieve(256) | wazero | 687.3 | 8 | 1 |
| Sieve(256) | Tengo | 53,630 | 122,504 | 1,611 |
| Sieve(256) | gopher-lua | 23,289 | 18,416 | 44 |
| Sieve(256) | Goja | 43,140 | 1,872 | 25 |
| Sieve(256) | gpython | 35,636 | 5,704 | 30 |
| Sieve(256) | Yaegi | 18,762 | 1,800 | 37 |
| RecursiveFib(20) | minivm/default | 38,280 | 0 | 0 |
| RecursiveFib(20) | minivm/threaded | 357,450 | 0 | 0 |
| RecursiveFib(20) | minivm/jit | 368,423 | 0 | 0 |
| RecursiveFib(20) | minivm/default | 38,368 | 0 | 0 |
| RecursiveFib(20) | minivm/threaded | 303,238 | 0 | 0 |
| RecursiveFib(20) | minivm/jit | 376,672 | 0 | 0 |
| RecursiveFib(20) | native Go | 14,455 | 0 | 0 |
| RecursiveFib(20) | wazero | 31,077 | 8 | 1 |
| RecursiveFib(20) | Tengo | 809,410 | 319,346 | 28,655 |
| RecursiveFib(20) | gopher-lua | 1,055,661 | 704 | 2 |
| RecursiveFib(20) | Goja | 1,456,275 | 4,680 | 39 |
| RecursiveFib(20) | gpython | 3,701,017 | 9,807,924 | 109,494 |
| RecursiveFib(20) | Yaegi | 3,791,815 | 8,302,126 | 192,840 |
| RecursiveFib(35) | minivm/default | 46,951,773 | 0 | 0 |
| RecursiveFib(35) | minivm/threaded | 499,079,076 | 0 | 0 |
| RecursiveFib(35) | minivm/jit | 522,672,192 | 0 | 0 |
| RecursiveFib(35) | minivm/default | 45,904,831 | 0 | 0 |
| RecursiveFib(35) | minivm/threaded | 410,581,243 | 0 | 0 |
| RecursiveFib(35) | minivm/jit | 513,509,366 | 0 | 0 |
| RecursiveFib(35) | native Go | 20,107,461 | 0 | 0 |
| RecursiveFib(35) | wazero | 44,453,238 | 9 | 1 |
| RecursiveFib(35) | Tengo | 1,168,612,000 | 312,797,584 | 39,088,176 |
| RecursiveFib(35) | gopher-lua | 1,477,085,041 | 971,008 | 3,793 |
| RecursiveFib(35) | Goja | 2,099,479,958 | 375,360 | 46,373 |
| RecursiveFib(35) | gpython | 5,578,092,292 | 13,378,028,656 | 149,350,236 |
| RecursiveFib(35) | Yaegi | 5,903,047,625 | 11,324,344,728 | 263,043,676 |
| IndirectRecursiveFib(20) | minivm/default | 54,763 | 0 | 0 |
| IndirectRecursiveFib(20) | minivm/threaded | 576,182 | 0 | 0 |
| IndirectRecursiveFib(20) | minivm/jit | 594,117 | 0 | 0 |
| IndirectRecursiveFib(20) | minivm/default | 53,559 | 0 | 0 |
| IndirectRecursiveFib(20) | minivm/threaded | 478,185 | 0 | 0 |
| IndirectRecursiveFib(20) | minivm/jit | 583,934 | 0 | 0 |
| IndirectRecursiveFib(20) | native Go | 15,981 | 0 | 0 |
| IndirectRecursiveFib(20) | wazero | 43,260 | 8 | 1 |
| IndirectRecursiveFib(20) | Tengo | 953,537 | 319,346 | 28,655 |
| IndirectRecursiveFib(20) | gopher-lua | 954,423 | 704 | 2 |
| IndirectRecursiveFib(20) | Goja | 1,377,150 | 4,680 | 39 |
| IndirectRecursiveFib(20) | gpython | 4,018,147 | 10,158,201 | 109,494 |
| IndirectRecursiveFib(20) | Yaegi | 11,430,601 | 13,059,854 | 394,041 |
| ClosureCounter(128) | minivm/default | 3,387 | 64 | 2 |
| ClosureCounter(128) | minivm/threaded | 2,978 | 64 | 2 |
| ClosureCounter(128) | minivm/jit | 3,372 | 64 | 2 |
| ClosureCounter(128) | minivm/default | 3,286 | 64 | 2 |
| ClosureCounter(128) | minivm/threaded | 2,438 | 64 | 2 |
| ClosureCounter(128) | minivm/jit | 3,291 | 64 | 2 |
| ClosureCounter(128) | native Go | 38.22 | 0 | 0 |
| ClosureCounter(128) | wazero | N/A | N/A | N/A |
| ClosureCounter(128) | Tengo | 13,503 | 92,272 | 261 |
| ClosureCounter(128) | gopher-lua | 5,910 | 151 | 3 |
| ClosureCounter(128) | Goja | 10,173 | 1,264 | 13 |
| ClosureCounter(128) | gpython | 28,235 | 58,312 | 659 |
| ClosureCounter(128) | Yaegi | 34,704 | 34,784 | 786 |
| TypedArraySum(256) | minivm/default | 307.7 | 0 | 0 |
| TypedArraySum(256) | minivm/threaded | 6,313 | 0 | 0 |
| TypedArraySum(256) | minivm/jit | 621.5 | 0 | 0 |
| TypedArraySum(256) | minivm/default | 289.2 | 0 | 0 |
| TypedArraySum(256) | minivm/threaded | 2,820 | 0 | 0 |
| TypedArraySum(256) | minivm/jit | 489.3 | 0 | 0 |
| TypedArraySum(256) | native Go | 73.2 | 0 | 0 |
| TypedArraySum(256) | wazero | 157 | 8 | 1 |
| TypedArraySum(256) | Tengo | 15,507 | 94,208 | 513 |
| TypedArraySum(256) | gopher-lua | 3,413 | 4,000 | 15 |
| TypedArraySum(256) | Goja | 13,399 | 2,080 | 238 |
| TypedArraySum(256) | gpython | 7,652 | 2,496 | 246 |
| TypedArraySum(256) | Yaegi | 4,246 | 296 | 8 |
| AllocationGraph(128) | minivm/default | 9,301 | 5,120 | 256 |
| AllocationGraph(128) | minivm/threaded | 7,774 | 5,120 | 256 |
| AllocationGraph(128) | minivm/jit | 9,362 | 5,120 | 256 |
| AllocationGraph(128) | minivm/default | 6,722 | 1,024 | 128 |
| AllocationGraph(128) | minivm/threaded | 4,816 | 1,024 | 128 |
| AllocationGraph(128) | minivm/jit | 6,737 | 1,024 | 128 |
| AllocationGraph(128) | native Go | 943.8 | 1,024 | 128 |
| AllocationGraph(128) | wazero | N/A | N/A | N/A |
| AllocationGraph(128) | Tengo | 14,516 | 96,288 | 388 |
| AllocationGraph(128) | gopher-lua | 6,543 | 14,376 | 256 |
| AllocationGraph(128) | Goja | 26,551 | 78,016 | 770 |
| AllocationGraph(128) | gpython | 5,715 | 5,712 | 266 |
| AllocationGraph(128) | Yaegi | 12,190 | 1,492 | 142 |
| BranchTree(96) | minivm/default | 264.4 | 0 | 0 |
| BranchTree(96) | minivm/threaded | 952 | 0 | 0 |
| BranchTree(96) | minivm/jit | 264.7 | 0 | 0 |
| BranchTree(96) | minivm/default | 230.9 | 0 | 0 |
| BranchTree(96) | minivm/threaded | 603.8 | 0 | 0 |
| BranchTree(96) | minivm/jit | 230.5 | 0 | 0 |
| BranchTree(96) | native Go | 79.98 | 0 | 0 |
| BranchTree(96) | wazero | 172.1 | 16 | 1 |
| BranchTree(96) | Tengo | 18,088 | 95,384 | 660 |
Expand Down Expand Up @@ -302,7 +302,7 @@ Do not infer native execution solely from the `jit` sub-benchmark name. A benchm
- External parsing, compilation, module creation, and function lookup remain outside the timer where the runtime API permits.
- Wazero uses its default compiler runtime; module compilation and instantiation are excluded from timing.
- Cross-runtime comparisons live in the separate `benchmarks/` module and require the `compare` build tag.
- Output was grouped by exact benchmark name, and each documented value is the median of exactly three samples.
- Output was grouped by exact benchmark name. Minivm rows use the median of five interleaved samples; external rows use the median of three sequential samples.

Cross-runtime library versions:

Expand Down
4 changes: 2 additions & 2 deletions docs/fusion.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,13 +28,13 @@ Every valid opcode has exactly one `lowerers` entry and one semantic emitter. Th

The generator validates the concrete patterns returned by `catalog` in `internal/cmd/geninterp/pattern.go`.

Patterns cover ref consumption, constant calls and closure creation, numeric operations and comparisons, conditional branches, and constant aggregate indexes. Trapping numeric operations materialize completed sources before evaluating the trap so stack ownership and instruction offsets match exact execution.
Patterns cover ref consumption, constant calls and closure creation, numeric operations and comparisons, conditional branches, constant aggregate indexes, direct non-trapping arithmetic stores to typed locals, and typed-array constants indexed by scalar producers. Trapping numeric operations materialize completed sources before evaluating the trap so stack ownership and instruction offsets match exact execution.

## Threaded Compilation

Threading checks the opcode-indexed fusion table before standalone opcode dispatch. Fusion preflight uses a local cursor and mutates nothing on a miss. A match installs one direct handler and advances compile-time IP by only the first opcode width. Absorbed offsets are still threaded separately, so branches into them execute standalone handlers. Exact threading disables fusion.

Compile-time specialization resolves operands, declared slot kinds, constants, heap objects, and cached coroutine metadata. Final handlers do not dispatch source functions, decode operands, inspect concrete heap types, or rescan bytecode for yields.
Compile-time specialization resolves operands, declared slot kinds, constants, heap objects, and cached coroutine metadata. Final handlers do not dispatch source functions, decode operands, or rescan bytecode for yields. A typed-array constant load still validates the current heap value's concrete slice type and bounds on every execution; specialization removes temporary stack materialization and balanced container retain/release work, not those runtime guards.

## JIT Separation

Expand Down
5 changes: 4 additions & 1 deletion docs/instruction-set.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -318,6 +318,8 @@ Examples:
- primitive constants feeding primitive binary operations
- typed locals feeding primitive binary operations
- typed locals plus primitive constants feeding binary operations
- non-trapping primitive binary results stored directly into typed locals
- typed-array constants plus scalar index producers feeding `array.get`
- constant indexes feeding `array.get` or `struct.get`
- constant ref cell plus `ref.get`
- structured-error creation followed by a raise
Expand All @@ -326,7 +328,8 @@ A fused source stays in a temporary instead of being pushed, so a fused
handler checks stack room once for its own net push rather than once per
folded source. Bounds, segmentation, and underflow checks stay per source.
Trapping arithmetic (`div`/`rem`/`mod`) still materializes its operands on the
stack, so it keeps a check per push.
stack, so it keeps a check per push. Typed-array constant loads validate the
current heap value's concrete array type and bounds on every execution.

## Maintenance Notes

Expand Down
13 changes: 13 additions & 0 deletions docs/memory-model.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,6 +124,19 @@ do not block future allocations.

Public host APIs that allocate, such as `Alloc`, `Push`, and `Marshal`, return `ErrHeapExhausted` as ordinary errors.

### Reset-time reference-array header reuse

`Reset` invalidates every live dynamic object. Before clearing those slots, the
interpreter keeps released `*types.Array` headers in an interpreter-local pool
capped by that run's dynamic heap size. VM array construction, slicing, and map
key extraction may reuse those headers on the next run.

Only the Go object header is reused. `Typ` and `Elems` are cleared at reset, and
every new array receives a fresh element backing store, so zeroing and retained
memory behavior stay unchanged. Arrays detached through `Pop` or reclaimed
before reset are not pooled; their returned Go values remain intact. `Close`
Comment on lines +129 to +137

@coderabbitai coderabbitai Bot Jul 30, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Clarify Reset’s ownership scope.

“Every live dynamic object” contradicts the following Pop exception. Limit this to dynamic objects still owned by the interpreter.

Proposed fix 
-`Reset` invalidates every live dynamic object. Before clearing those slots, the
+`Reset` invalidates every dynamic object still owned by the interpreter. Before clearing those slots, the
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
`Reset` invalidates every live dynamic object. Before clearing those slots, the
interpreter keeps released `*types.Array` headers in an interpreter-local pool
capped by that run's dynamic heap size. VM array construction, slicing, and map
key extraction may reuse those headers on the next run.
Only the Go object header is reused. `Typ` and `Elems` are cleared at reset, and
every new array receives a fresh element backing store, so zeroing and retained
memory behavior stay unchanged. Arrays detached through `Pop` or reclaimed
before reset are not pooled; their returned Go values remain intact. `Close`
`Reset` invalidates every dynamic object still owned by the interpreter. Before clearing those slots, the
interpreter keeps released `*types.Array` headers in an interpreter-local pool
capped by that run's dynamic heap size. VM array construction, slicing, and map
key extraction may reuse those headers on the next run.
Only the Go object header is reused. `Typ` and `Elems` are cleared at reset, and
every new array receives a fresh element backing store, so zeroing and retained
memory behavior stay unchanged. Arrays detached through `Pop` or reclaimed
before reset are not pooled; their returned Go values remain intact. `Close`
🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/memory-model.md` around lines 129 - 137, Update the Reset description to
say it invalidates every live dynamic object still owned by the interpreter,
rather than every live dynamic object. Keep the existing Pop exception and
pooling behavior unchanged.

drops the pool.

## GC

GC uses trial deletion to derive roots from exact reference counts instead of
Expand Down
23 changes: 23 additions & 0 deletions internal/cmd/geninterp/generate_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,29 @@ func TestGenerate(t *testing.T) {
require.NotContains(t, source, "types.KindI32")
})

t.Run("stores local arithmetic without stack materialization", func(t *testing.T) {
pattern := seq(op(instr.LOCAL_GET), op(instr.I32_CONST), op(instr.I32_ADD), op(instr.LOCAL_SET))
body, err := compose(pattern, pattern.width(), "")
require.NoError(t, err)
file := jen.NewFile("review")
file.Func().Id("render").Params().Block(body...)
source := file.GoString()
require.Contains(t, source, "stack[addr] = v2")
require.NotContains(t, source, "stack[i.sp] = v2")
})

t.Run("loads typed arrays without materializing the container", func(t *testing.T) {
pattern := seq(constant[types.TypedArray[int32]](), op(instr.LOCAL_GET), op(instr.ARRAY_GET))
body, err := compose(pattern, pattern.width(), "")
require.NoError(t, err)
file := jen.NewFile("review")
file.Func().Id("render").Params().Block(body...)
source := file.GoString()
require.Contains(t, source, "array, ok := i.heap[c0].(types.TypedArray[int32])")
require.NotContains(t, source, "i.retain(")
require.NotContains(t, source, "i.release(")
})

t.Run("accepts the catalog", func(t *testing.T) {
require.NoError(t, validate(catalog()))
})
Expand Down
Loading
Loading