[MPlayer-dev-eng] SwScaler YV12 to BGR32 having zeroed alpha

[Jason Tackaberry]

I'm using SwScaler to scale and convert between YV12 and BGR32
colorspaces.  I'm noticing that the alpha channel of the resulting BGR32
image is completely zeroed, which means in order to do anything useful
with it (when using it with code that pays attention to per-pixel alpha)
I need to loop through the buffer and adjust every fourth byte, which
is, I think, needless overhead.

If this behaviour is hard-coded, I think it should be considered a bug.
But I'm posting here to find out if there's an argument or other
function I'm missing that can do what I want.

I'm setting up the sws context this way:

    sws = sws_getContext(priv->width, priv->height, priv->outfmt, sw, sh, IMGFMT_BGR32, 
                         get_sws_cpuflags() | SWS_PRINT_INFO | SWS_BILINEAR,
                         NULL, NULL, NULL);

(Where sw and sh are the target scaled width and height.)

And I'm actually scaling the image this way:

    uint8_t *dst[3]= {bgra_imgbuf, NULL, NULL};
    int dst_stride[3]= {scaled_width * 4, 0, 0};
    sws_scale_ordered(sws, mpi->planes, mpi->stride, 0, vf->priv->height, dst, dst_stride);

The resulting bgra_imgbuf ends up with a zeroed alpha channel.  It's
quite simple to fix, of course:

   for (i=3; i < scaled_width * scaled_height * 4; i+=4)
      bgra_imgbuf[i] = 255;

But this is some overhead I don't think should be needed as I feel the
alpha bytes should be initialized to a more sane value (255) within the
sws code.  Hopefully this is possible?  If not, would a patch be
accepted to change the current behaviour?

Also, as an aside, I discovered that SWS_FAST_BILINEAR will cause a
segfault on x86_64 architecture when using CVS from 2005-05-27.  Perhaps
this is already known, but I thought I'd mention it briefly in case
someone wants to have a look at it.

Cheers,
Jason.