chore: Use mimalloc as the default allocator

[jfecher]

Description

Problem*

Resolves

Summary*

Testing to see if this improves performance

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[github-actions]

Compilation Memory Report

Program	Peak Memory	%
keccak256	107.15K	100%
workspace	107.15K	100%
regression_4709	107.15K	100%
ram_blowup_regression	107.15K	100%
rollup-base-public	107.15K	100%
rollup-base-private	107.15K	100%
private-kernel-tail	107.15K	100%
private-kernel-reset	107.15K	100%
private-kernel-inner	107.15K	100%

[github-actions]

Execution Memory Report

Program	Peak Memory	%
keccak256	107.15K	100%
workspace	107.15K	100%
regression_4709	107.15K	100%
ram_blowup_regression	107.15K	100%
rollup-base-public	107.15K	100%
rollup-base-private	107.15K	100%
private-kernel-tail	107.15K	100%
private-kernel-reset	107.15K	100%
private-kernel-inner	107.15K	100%

[github-actions]

Compilation Report

Program	Compilation Time	%
sha256_regression	1.030s	-19%
regression_4709	0.756s	-5%
ram_blowup_regression	15.200s	-11%
rollup-root	3.206s	-4%
rollup-merge	1.786s	-3%
rollup-block-merge	3.142s	-6%
rollup-base-public	33.120s	-9%
rollup-base-private	10.800s	-14%
private-kernel-tail	0.869s	-11%
private-kernel-reset	5.306s	-16%
private-kernel-inner	1.902s	-5%

[github-actions]

Execution Report

Program	Execution Time	%
sha256_regression	0.051s	-2%
regression_4709	0.003s	200%
ram_blowup_regression	0.610s	0%
rollup-root	0.105s	0%
rollup-merge	0.007s	16%
rollup-block-merge	0.105s	0%
rollup-base-public	1.218s	-1%
rollup-base-private	0.455s	0%
private-kernel-tail	0.019s	0%
private-kernel-reset	0.315s	0%
private-kernel-inner	0.068s	0%

[jfecher]

Some good compilation time improvements here for such a small change but the memory reports are broken

[ludamad]

Interesting, thought rust had quite a good allocator

[aakoshh]

FWIW I tried the profiling I do with Instruments in #7053 and it seems to have bombed (I'm on Mac):

Or I'm not sure what happens exactly, usually it runs for minutes, and the 12GB is also not something I have seen before.

If I un e.g. info I do see output though 🤷
Maybe the profiler could not attach itself and the compilation stayed fast, but the reported memory completely bogus.

But no, even that's not true, I have stack trace:

It just shows way more memory than expected.

[TomAFrench]

Yeah, heaptrack reports that it doesn't work with custom allocators so we may need to replace this tooling for something else if we go forwards with this.

[jfecher]

@ludamad Rust just uses the default system allocator

[jfecher]

Looks like valgrind's massif tool isn't working either

[jfecher]

Profiling with MIMALLOC_SHOW_STATS=1 I also see the 12GB peak usage. Switching this PR to try out TCMalloc instead

[jfecher]

This could be something to continue looking into for the future but I currently can't seem to profile memory usage of jemalloc, and any alternate allocator in general would break our CI so I'm closing this for now despite the possible performance gains.