perf: use mimalloc for musl target

[xc2]

Summary

closes: #6072

related: #6065

linux musl binding now performs ~3x times faster than before.

sample of 0.5.9

Running case debian (GNU)
Rspack compiled successfully in 1.67 s

Running case alpine (musl)
Rspack compiled successfully in 6.39 s

Running case alpine-mimalloc-v1 (musl, redirecting malloc to mimalloc 1.8.2)
Rspack compiled successfully in 1.78 s

Running case alpine-mimalloc-v2 (musl, redirecting malloc to mimalloc 2.1.2)
Rspack compiled successfully in 1.70 s

sample of this change

Running case debian (GNU, nothing change in this pr)
Rspack compiled successfully in 1.86 s

Running case alpine (musl, with mimalloc rust binding)
Rspack compiled successfully in 1.97 s

Running case alpine-mimalloc-v1 (musl, with mimalloc rust binding, redirecting malloc to mimalloc 1.8.2)
Rspack compiled successfully in 1.89 s

Running case alpine-mimalloc-v2 (musl, with mimalloc rust binding, redirecting malloc to mimalloc 2.1.2)
Rspack compiled successfully in 1.81 s

Checklist

• Tests updated (or not required): no need
• Documentation updated (or not required): no need

[xc2]

!canary

[netlify]

✅ Deploy Preview for rspack canceled.

Built without sensitive environment variables

Name	Link
🔨 Latest commit	c33ff59
🔍 Latest deploy log	https://app.netlify.com/sites/rspack/deploys/6608257364659f0008e12341

[github-actions]

0.5.9-canary-df5312d-20240330095228

[xc2]

!canary

[github-actions]

0.5.9-canary-16e548a-20240330113014

[hardfist]

!bench

[rspack-bot]

📝 Benchmark detail: Open

Name	Base (2024-03-30 48aa3b9)	Current	Change
10000_development-mode + exec	2.07 s ± 41 ms	2.06 s ± 13 ms	-0.61 %
10000_development-mode_hmr + exec	765 ms ± 5.1 ms	764 ms ± 12 ms	-0.16 %
10000_production-mode + exec	2.95 s ± 22 ms	3.05 s ± 25 ms	+3.35 %
arco-pro_development-mode + exec	2.46 s ± 38 ms	2.49 s ± 31 ms	+1.38 %
arco-pro_development-mode_hmr + exec	489 ms ± 3.6 ms	490 ms ± 2.9 ms	+0.05 %
arco-pro_development-mode_hmr_intercept-plugin + exec	506 ms ± 4.8 ms	508 ms ± 7.7 ms	+0.34 %
arco-pro_development-mode_intercept-plugin + exec	3.34 s ± 34 ms	3.38 s ± 72 ms	+1.33 %
arco-pro_production-mode + exec	4.15 s ± 35 ms	4.15 s ± 38 ms	-0.06 %
arco-pro_production-mode_intercept-plugin + exec	5.02 s ± 61 ms	5.01 s ± 25 ms	-0.12 %
threejs_development-mode_10x + exec	1.9 s ± 47 ms	1.87 s ± 28 ms	-1.38 %
threejs_development-mode_10x_hmr + exec	729 ms ± 9 ms	729 ms ± 4.3 ms	-0.01 %
threejs_production-mode_10x + exec	5.73 s ± 34 ms	5.86 s ± 148 ms	+2.29 %

[h-a-n-a]

Seems like we can use mimalloc for everything. https://github.com/napi-rs/node-rs/blob/70d7fa72262c6e547950b30daa2d03583a1b04bd/crates/alloc/Cargo.toml

[xc2]

Seems like we can use mimalloc for everything. https://github.com/napi-rs/node-rs/blob/70d7fa72262c6e547950b30daa2d03583a1b04bd/crates/alloc/Cargo.toml

yea it's good to make the allocator unified but i didn't make the change as i found it didn't help much on performance for glibc linked binary.

and i need more effort to check if these change below works well. let me check if i got time later.

• mimalloc-rust -> mimalloc
• tikv-jemalloc -> mimalloc

[Boshen]

Let's fix the musl performance issue first and try changing to mimalloc in another PR.