Re: taisei: update to 1.4.1

[tornaria <tornaria@users.noreply.github.com>]

[-- Attachment #1: Type: text/plain, Size: 1744 bytes --]

New comment by tornaria on void-packages repository

https://github.com/void-linux/void-packages/pull/49348#issuecomment-2022901998

Comment:
> `-msse2` is ok, but the proper fix would be to correct this check in CGLM: https://github.com/recp/cglm/blob/1de373a9bd453d1fff6846db3a01ade8270f12bb/include/cglm/simd/intrin.h#L27
> 
> It should only check for SSE2 there. I can try to submit that upstream a bit later, though I'll need to set up a 32-bit environment to test it first.

An easy way is to check out this branch, then do
```
./xbps-src -A i686 binary-bootstrap
./xbps-src -A i686 pkg cglm
./xbps-src -A i686 pkg taisei
```
You can put patches in `srcpkgs/cglm/patches/*.patch` before the `pkg cglm` step.

> For what it's worth, openSUSE doesn't build Taisei for i686, because CGLM fails (or at least used to fail) some unit tests on that platform. It doesn't look like i686 support is well maintained there, and we neither test nor provide i686 Linux builds of Taisei either. I don't believe there's an x86 system around that can handle this game acceptably well while not supporting x86-64.

Isn't that a good reason to build with sse2 by default?

> As for why we enforce `-msse`: what we really want is `-mfpmath=sse`, which prevents the compiler from using x87 FPU instructions for floating point math. It's to reduce the potential for replay compatibility problems between different builds, due to the extended precision used by those instructions.

You can achieve a similar result without sse using `-ffloat-store` (at a cost).

BTW, do you really need level=20 for zstd? That's supposed to use a lot of memory. Maybe 19 is good enough (or you could use something like `20 if sys.maxsize > 2**32 else 19`).