[FFmpeg-devel] Hardware CSC conversion segfault

Reimar Döffinger Reimar.Doeffinger at gmx.de
Sun Jul 14 09:44:48 CEST 2013


On 13.07.2013, at 20:29, Derek Buitenhuis <derek.buitenhuis at gmail.com> wrote:
> On 7/13/2013 2:28 PM, Derek Buitenhuis wrote:
>> On 7/13/2013 3:37 AM, Reimar Döffinger wrote:
>>> None of these say that get_buffer is not allowed to return continuous memory (i.e. malloc each plane separately - which would be an idiotic requirement) so I have no idea what you are talking about.
>> 
>> The original email refers "contiguous memory buffer holding the different
>> planes of YUV data or can they discontinuous", and I'm pretty certain he meant
>> "with no padding", i.e. width == linesize, which is impossible on some resolutions,
>> given get_buffer's alignment requirements.
> 
> I should note the key word here is contiguous, which a padded/aligned buffer
> is not.

Even with stride, the buffer is still one block, even though the data itself may have gaps.
Also, for color space conversion stride isn't an issue (at least not with the constraints we have), it's possible to just tune the width.
Either way, had I concluded the question wasn't about stride since that wouldn't have caused a crash. But yes, it is another thing that might make (especially a memcpy-less) implementation harder.


More information about the ffmpeg-devel mailing list