[FFmpeg-devel] libavcodec/bswapdsp : add AVX2 for bswap_buf

[James Darnley]

On 2017-10-28 16:14, Martin Vignali wrote:
> 2017-10-28 15:47 GMT+02:00 James Darnley <james.darnley at gmail.com>:
> 
>> On 2017-10-22 19:14, Martin Vignali wrote:
>>> @@ -86,16 +101,20 @@ SECTION .text
>>>
>>>  ; void ff_bswap_buf(uint32_t *dst, const uint32_t *src, int w);
>>>  %macro BSWAP32_BUF 0
>>> -%if cpuflag(ssse3)
>>> +%if cpuflag(ssse3)||cpuflag(avx2)
>>>  cglobal bswap32_buf, 3,4,3
>>>      mov      r3, r1
>>> +%if cpuflag(avx2)
>>> +    vbroadcasti128  m2, [pb_bswap32]
>>> +%else
>>>      mova     m2, [pb_bswap32]
>>> +%endif
>>
>> Why don't you increase the size of the constant?  You wouldn't need a
>> preprocessor condition.
>>
>> I do admit I have never considered using a broadact-128 instruction in
>> similar circumstances so I have no idea which might be "better".
>>
>>
> For this, i follow Henrik Gramner's code, in the discussion
> libavcodec/exr : add x86 SIMD for predictor.

That's fine.  I will look at that on my own time.  I don't want to hold
up applying the patch.  Would you like me to push it for you?


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 603 bytes
Desc: OpenPGP digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20171028/95fb1e8b/attachment.sig>