perf(pm): adopt mimalloc as global allocator#3149
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request configures mimalloc as the global allocator for the pm crate to optimize performance for allocation-heavy workloads. However, because mimalloc is only declared as a dependency for non-wasm32 targets in Cargo.toml, unconditionally importing and setting it in main.rs will break compilation for wasm32 targets. A review comment correctly identifies this issue and suggests wrapping the global allocator setup in a #[cfg(not(target_arch = "wasm32"))] conditional compilation attribute.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| use mimalloc::MiMalloc; | ||
|
|
||
| #[global_allocator] | ||
| static GLOBAL: MiMalloc = MiMalloc; |
There was a problem hiding this comment.
The mimalloc dependency is only specified for non-wasm32 targets in Cargo.toml. Unconditionally importing and setting it as the global allocator will cause compilation to fail when targeting wasm32. Wrap the global allocator definition in a #[cfg(not(target_arch = "wasm32"))] attribute.
| use mimalloc::MiMalloc; | |
| #[global_allocator] | |
| static GLOBAL: MiMalloc = MiMalloc; | |
| #[cfg(not(target_arch = "wasm32"))] | |
| #[global_allocator] | |
| static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc; |
Re-landed on the post-#3157 next. The workspace [patch.crates-io] already points mimalloc/libmimalloc-sys at the utooland fork (shared with the pack crates' TurboMalloc); pm now opts the binary in via #[global_allocator]. Allocation-heavy install workload (multi-MB JSON parses, graph build, per-file clone bookkeeping) with cross-thread alloc/free where mimalloc beats glibc arena locking. wasm32 target left on the system allocator. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
elrrrrrrr
force-pushed
the
perf/pm-mimalloc
branch
from
5743bc2 to
2168db7
Compare
📊 pm-bench-phases ·
|
| PM | wall (mean ± σ) | min | user | sys | RSS | Δ vs baseline |
|---|---|---|---|---|---|---|
| utoo (PR) | 10.4s ± 2.18s | 7.92s | 9.29s | 12.7s | 1.07G | +4.5% ✅ |
| utoo-next (baseline) | 9.81s ± 1.82s | 7.72s | 9.34s | 12.6s | 947M | — |
| bun | 12.2s ± 1.12s | 11.3s | 11.1s | 10.9s | 690M | +11.4% |
p1 · resolve
| PM | wall (mean ± σ) | min | user | sys | RSS | Δ vs baseline |
|---|---|---|---|---|---|---|
| utoo (PR) | 2.73s ± 0.69s | 2.31s | 4.33s | 1.43s | 754M | -20.6% ✅ |
| utoo-next (baseline) | 3.62s ± 1.32s | 2.45s | 4.42s | 1.69s | 586M | — |
| bun | 4.67s ± 1.89s | 2.60s | 4.44s | 1.23s | 520M | +12.8% ✅ |
p3 · cold install
| PM | wall (mean ± σ) | min | user | sys | RSS | Δ vs baseline |
|---|---|---|---|---|---|---|
| utoo (PR) | 7.00s ± 0.49s | 6.57s | 5.10s | 11.6s | 570M | +6.7% ✅ |
| utoo-next (baseline) | 6.97s ± 0.55s | 6.46s | 5.18s | 11.4s | 517M | — |
| bun | 8.03s ± 0.72s | 7.46s | 6.82s | 10.8s | 566M | +17.0% |
p4 · warm link
| PM | wall (mean ± σ) | min | user | sys | RSS | Δ vs baseline |
|---|---|---|---|---|---|---|
| utoo (PR) | 2.54s ± 0.07s | 2.47s | 0.43s | 3.73s | 208M | +2.0% ✅ |
| utoo-next (baseline) | 2.56s ± 0.23s | 2.33s | 0.48s | 3.77s | 49M | — |
| bun | 3.90s ± 0.10s | 3.80s | 0.19s | 2.59s | 132M | +53.0% |
Resources & footprint
p0 · full cold install
| PM | vCtx | iCtx | netRX | netTX | cache | node_modules | lock |
|---|---|---|---|---|---|---|---|
| utoo | 70564 | 59058 | 1.25G | 7M | 1.83G | 1.82G | 2M |
| utoo-next | 74105 | 59906 | 1.25G | 7M | 1.83G | 1.82G | 2M |
| bun | 15797 | 18091 | 1.27G | 6M | 1.98G | 1.85G | 1M |
p1 · resolve
| PM | vCtx | iCtx | netRX | netTX | cache | node_modules | lock |
|---|---|---|---|---|---|---|---|
| utoo | 33302 | 50046 | 210M | 2M | 7M | 3M | 2M |
| utoo-next | 32493 | 50717 | 210M | 2M | 7M | 3M | 2M |
| bun | 11846 | 3711 | 212M | 3M | 116M | 0B | 1M |
p3 · cold install
| PM | vCtx | iCtx | netRX | netTX | cache | node_modules | lock |
|---|---|---|---|---|---|---|---|
| utoo | 67100 | 49708 | 1.04G | 4M | 1.82G | 1.82G | 2M |
| utoo-next | 72806 | 49150 | 1.04G | 4M | 1.82G | 1.82G | 2M |
| bun | 4466 | 7820 | 1.06G | 4M | 1.87G | 1.87G | 1M |
p4 · warm link
| PM | vCtx | iCtx | netRX | netTX | cache | node_modules | lock |
|---|---|---|---|---|---|---|---|
| utoo | 21685 | 13614 | 1K | 14K | 1.82G | 1.82G | 2M |
| utoo-next | 22714 | 13843 | 1K | 14K | 1.82G | 1.82G | 2M |
| bun | 281 | 22 | 3K | 11K | 1.87G | 1.87G | 1M |
1 participant
Split out of #3146 — single orthogonal change so the benchmark comment shows its isolated effect.
mimalloc as
#[global_allocator]for theutoobinary (via the existing workspace[patch.crates-io]utooland fork — same allocator family the pack crates use through TurboMalloc).Why re-test: the April experiment (
perf/pm-mimalloc-allocator) showed no clear win, but that predates the v1.1.1 demand-resolver rework — parsing now runs on a dedicated rayon pool with results handed across threads, which is exactly the cross-thread alloc/free pattern where mimalloc beats glibc's arena locking.Expected bench signal: p1 (and p0's resolve half) improve a few percent if the hypothesis holds; p4 unchanged (syscall-bound). RSS will read higher (mimalloc retains freed pages).
🤖 Generated with Claude Code