[FFmpeg-devel] [PATCH v5 4/4] lavf/utils: Fix endianness in ff_get_packet_palette()

[Mats Peterson]

On 03/03/2016 07:15 AM, Mats Peterson wrote:
> On 03/03/2016 06:52 AM, Mats Peterson wrote:
>> On 03/03/2016 06:39 AM, Mats Peterson wrote:
>>> On 03/03/2016 06:33 AM, Mats Peterson wrote:
>>>> On 03/03/2016 03:05 AM, Michael Niedermayer wrote:
>>>>> with this API each muxer must convert the palette from native endian
>>>>> to whatever the format requires
>>>>> feels a bit odd to extact the palette from AVPacket.data where its
>>>>> stored as the container needs (for some containers) and then
>>>
>>> Exactly for the reason that the AVPacket.data is stored in
>>> "container-specific" (exclusively little-endian, i hope) format, it has
>>
>> Well, exclusively little-endian in the case of raw nut, at least. That's
>> the only format that appends the palette to the video data in each frame
>> in the file, as far as I know.
>>
>> Mats
>>
>> _______________________________________________
>
> By the way, would you mind telling me why the palette in AVPacket.data
> is always little endian (regardless of using nut or not)? And where in
> FFmpeg is conversion from native endian to little endian done?
>
> Mats
>
> _______________________________________________

I found it in av_image_copy_to_buffer() in libavutil/imgutils.c. I 
suppose that's the right place:

if (desc->flags & AV_PIX_FMT_FLAG_PAL) {
         uint32_t *d32 = (uint32_t *)dst;

         for (i = 0; i<256; i++)
             AV_WL32(d32 + i, AV_RN32(src_data[1] + 4*i));
     }
}