Use mimalloc allocator for static build

[danleh]

Which fixes performance problems, especially on heavily multi-threaded workloads (many functions, high core count machines), see #5561.

[tlively]

LGTM! @kripken?

[kripken]

In general I worry about this kind of thing because the pinned version can get stale, but given the huge benefit here, it seems reasonable!

[dschuff]

So I tried this out on my macbook and got some odd errors (they are different depending on whether mimalloc is linked as a static archive or an object library). Haven't figured out what's going on yet but we might want to gate this behind a CMake flag for now (otherwise it may break when we try to do an emscripten-releases release build).

[danleh]

So I tried this out on my macbook and got some odd errors (they are different depending on whether mimalloc is linked as a static archive or an object library). Haven't figured out what's going on yet but we might want to gate this behind a CMake flag for now (otherwise it may break when we try to do an emscripten-releases release build).

My bad. Added a MIMALLOC_STATIC option, configured the mimalloc build to only produce the static library, and enabled MIMALLOC_STATIC only in the Alpine Linux build (in both ci.yml and create_release.yml). Could you rerun CI?

[arsnyder16]

looking forward to this being pull through into emsdk

[kripken]

lgtm but the CI error looks like it might be real?

[danleh]

https://github.com/WebAssembly/binaryen/actions/runs/13925295048/job/38975354336?pr=7378#step:10:2629

mimalloc: assertion failed: at "/src/third_party/mimalloc/src/segment.c":527, _mi_segments_collect
assertion: "tld->pages_purge.first == NULL"

Mh, it seems we are hitting microsoft/mimalloc#1031.

(@kripken Indeed. Seems we had a race with our comments.)

I'll try if a different commit/branch of mimalloc fixes the issue. Can reproduce locally (not sure why I didn't hit this yesterday locally. Maybe I only tested with LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libmimalloc.so.2 instead of static linking, not sure.)

[kripken]

Current error on CI looks like a warning from mimalloc. We don't want such warnings shown to our users - is there a way to disable them?

I must say, the warnings + that error do make me a little worried here. Is there perhaps a "stable" branch of mimalloc we should be using?