On Tue, 3 Jun 2025, Niklas Haas wrote:
On Mon, 02 Jun 2025 15:41:33 -0300 James Almer <jamr...@gmail.com> wrote:
GCC/Clang is smart enough to emit minss/maxss the same way as these functions.
The only theoretical benefit was in x86_32, where x87 floats are used, but the
penalty of making the clipping opaque to the compiler's scheduler plus moving
values from mmx regs to xmm and back will offset any potential speedup.
x86_32 builds targetting anything made in the last two decades and a half
should use -msse -mfp=sse anyway.
As mention in the another thread, x87 FPU usage causes non-bitexact results in
swscale. Should we at this point consider setting -mfpu=sse by default for
x86_32 builds?
I don't object to doing that - however, if we have float code, that is
expected to be bitexact, the root issue still remains even if we force SSE
fpmath on x86. The same issue could crop up on any
architecture/OS/compiler combo, it's just that x87 math shows the issues
much easier. (Not making a judgement whether that's an issue we want to
deal with, or whether we're ok with having bitexact code that relies on
float math that behaves close enough to some specific reference.)
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
