[Ffmpeg-devel] PIX_FMT_RGB555 and alpha channel

[Steven Johnson]

Before I start, this post has nothing to do with the FLC Encoder.  I am 
happy to have it mask pixels and what not.  It doesn't matter to me 
either way, but I am confused about the pixel formats.

I just looked into libswscale to see what it does, as I had a bunch of 
patches for the swscale inside FFMpeg, but Alex informs me that they 
wont be accepted for a number of reasons, but a big one is that there is 
a move towards using libswscale by preference, so I was seeing what the 
work of adding them to libswscale was. (The other issues I believe I can 
resolve).

I noticed that for rgb15 formats libswscale always sets the MSBit to 0.  
Which is not what FFMpeg does for this format.  With the MSBit = 0, 
FFMpeg would say that the pixel is fully transparent. They would all 
need to be or'd with 0x8000 to have the same format as FFMpeg RGB555.   
Is this a bug?  Should they be or'd with 0x8000 to make the format == to 
FFMpeg's version?

Also, it would appear that there is no Alpha support in libswscale, is 
that correct? 

Steven J



Alex Beregszaszi wrote:
> Hi,
>
>   
>>> Index: libavutil/avutil.h
>>> ===================================================================
>>> --- libavutil/avutil.h	(revision 6738)
>>> +++ libavutil/avutil.h	(working copy)
>>> @@ -78,7 +78,8 @@
>>>      PIX_FMT_YUV410P,   ///< Planar YUV 4:1:0,  9bpp, (1 Cr & Cb
>>>      sample per 4x4 Y samples) PIX_FMT_YUV411P,   ///< Planar YUV
>>>      4:1:1, 12bpp, (1 Cr & Cb sample per 4x1 Y samples)
>>>      PIX_FMT_RGB565,    ///< Packed RGB 5:6:5, 16bpp, (msb)   5R 6G
>>>      5B(lsb), in cpu endianness
>>> -    PIX_FMT_RGB555,    ///< Packed RGB 5:5:5, 16bpp, (msb)1A 5R 5G
>>> 5B(lsb), in cpu endianness most significant bit to 1 +   
>>> PIX_FMT_RGB555,    ///< Packed RGB 5:5:5, 16bpp, (msb)1A 5R 5G
>>> 5B(lsb), in cpu endianness most significant bit to 0 +   
>>> PIX_FMT_RGBA555,    ///< Packed RGB 5:5:5, 16bpp, (msb)1A 5R 5G
>>> 5B(lsb), in cpu endianness most significant bit to 1
>>>      PIX_FMT_GRAY8,     ///<        Y        ,  8bpp
>>>       
>> breaks binary compatibility
>>
>> also it likely breaks the swscaler as you didnt change it
>>     
>
> Does SWScaler sets the MSB to 1 in case of RGB555? If not, this doesnt
> breaks it.
>
>