[FFmpeg-devel] [PATCH V1 1/3] lavu: Add alpha blending API based on row.

[James Almer]

On 9/25/2018 10:45 PM, mypopy at gmail.com wrote:
> On Wed, Sep 26, 2018 at 3:55 AM Rostislav Pehlivanov <atomnuker at gmail.com>
> wrote:
>>
>> On 25 September 2018 at 16:27, Jun Zhao <mypopydev at gmail.com> wrote:
>>
>>> Add alpha blending API based on row, support global alpha blending/
>>> per-pixel blending, and add SSSE3/AVX2 optimizations of the functions.
>>>
> 
> 
>> We don't use inline asm on x86 and we don't use global contexts. Look at
>> how float_dsp is done.
> 
> I guess you precise mean "prefer NASM assembler over inline asm on x86". :)
> In fact,
> I know some x86 inline asm in FFmpeg, e,g libavcodec/x86/h264_cabac.
> (Use grep "__asm__ volatile" can find more x86 inline asm). And we need to
> update
> the inline asm on x86 rule in
> https://github.com/FFmpeg/FFmpeg/blob/master/doc/optimization.txt?

Yes, we still have some inline asm either because nobody has gotten
around to port it to NASM syntax after the project moved to it, or
because like with CABAC and some single instruction functions in
libavutil it makes sense being inline since the call overhead would kill
performance.

That document could use some polishing, but in any case, as stated in
the "Inline asm vs. external asm" section, we have for several years
required new code that calls external functions to be written in NASM
syntax, as it's the case with this patchset.

> 
> Thanks.
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>