[FFmpeg-devel] [PATCH 2/2]v7 Opus Pyramid Vector Quantization Search in x86 SIMD asm

[Rostislav Pehlivanov]

On 6 August 2017 at 19:52, Ivan Kalvachev <ikalvachev at gmail.com> wrote:

> This patch requires "Add macros used in opus_pvq_search to x86util.asm"
> as 4 of the macros are moved there.
>
> 1. Cosmetics is completely redone.
>
> 2. I've left the align code as it is.
> I found a really old nasm-2.07 version (from 19 Jan 2010) and made a test
> build.
> I got nasm-2.09.04 (from Jan 11 2011) too, just to be sure.
> They all passed without issues.
>
> The x264 x86inc.asm also uses smartalign without
> checking version number.
>
> Also I had to do a bit more extensive benchmarks,
> because it's hard to tell which version is better
> (with or without align).
> So far it looks like the align might be faster
> with 2-6 cycles at best.
>
> So until somebody finds some concrete issue
> I'd like to keep the code as it is.
>
> (maybe try avx2 without align:)
>
>
> I hope I haven't forgotten to do something.
> And I do hope I haven't messed up something new.
>
> Best Regards.
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
>
Pushed, thanks