[FFmpeg-devel] remove DEC Alpha DSP & support code

[Rémi Denis-Courmont]

Le 10 juin 2024 16:07:14 GMT+03:00, Andreas Rheinhardt <andreas.rheinhardt at outlook.com> a écrit :
>Rémi Denis-Courmont:
>> 
>> 
>> Le 10 juin 2024 14:57:29 GMT+03:00, Michael Niedermayer <michael at niedermayer.cc> a écrit :
>>> Also we have alpha fate clients:
>>> https://fate.ffmpeg.org/?query=subarch:alpha%2F%2F
>> 
>> Are there? I only see Ubuntu 22.04 builds which must be cross-compiled since Ubuntu 22.04 doesn't support DEC Alpha, not even as a port.
>> 
>> So I guess the tests are run on user-mode QEMU? For the sake of the argument and to answer your performance question, SIMD on QEMU is in my experience much *slower* than scalar C, due to limitations of TCG. So if that's the goal post, then removing the optimisations is actually a performance win.
>> 
>
>I doubt that the same can be said about removing the fast 64bit and clz
>flags.

Sure. TCG has express support for CLZ and CTZ. We could default to the GCC built-ins on Clang and GCC regardless of fast CLZ though.

IMO the fast 64-bit detection should use some default heuristic for unknown platforms though. We could at least check if UINTPTR_MAX is larger or equal to UINT64_MAX (won't work for 64-bit ILP32 ABIs, but I don't think Alpha has ever had one).

In any case, this may be relevant for some platforms but probably not for Alpha as of 2024.