[Spice-devel] [spice v10 04/27] server: Add a GStreamer 1.0 MJPEG video encoder and use it by default

[Francois Gouget]

On Wed, 2 Mar 2016, Christophe Fergeau wrote:
[...]
> > +if test "x$enable_gstreamer" != "xno"; then
> > +    SPICE_CHECK_GSTREAMER(GSTREAMER_1_0, 1.0, [gstreamer-1.0 gstreamer-base-1.0 gstreamer-app-1.0 gstreamer-video-1.0],
> > +        [enable_gstreamer="yes"
> > +         SPICE_CHECK_GSTREAMER_ELEMENTS($GST_INSPECT_1_0, [gst-plugins-base 1.0], [appsrc videoconvert appsink])
> > +         SPICE_CHECK_GSTREAMER_ELEMENTS($GST_INSPECT_1_0, [gstreamer-libav 1.0], [avenc_mjpeg])
> > +         SPICE_CHECK_GSTREAMER_ELEMENTS($GST_INSPECT_1_0, [gst-plugins-good 1.0], [vp8enc])
> > +         SPICE_CHECK_GSTREAMER_ELEMENTS($GST_INSPECT_1_0, [gst-plugins-ugly 1.0], [x264enc])
> 
> Maybe the addition of the tests for vp8enc/x264enc should be moved to
> the respective commits making use of these?

Whoops, I missed those. Moved.


> However, I don't think we can go with these compile-time tests and
> disable gstreamer altogether when they are missing.

SPICE_CHECK_GSTREAMER_ELEMENTS() only issues warnings.


[...]
> > +/* A helper for spice_gst_encoder_encode_frame() */
> > +static SpiceFormatForGStreamer *map_format(SpiceBitmapFmt format)
> > +{
> > +    /* See GStreamer's part-mediatype-video-raw.txt and
> > +     * section-types-definitions.html documents.
> > +     */
> > +    static SpiceFormatForGStreamer format_map[] =  {
> 
> Some constness could probably be added here and to the return
> value.

Makes sense. Done.


[...]
> > +/* A helper for push_raw_frame() */
> > +static inline int chunk_copy(SpiceGstEncoder *encoder, const SpiceBitmap *bitmap,
> > +                             uint32_t chunk_offset, uint32_t len, uint8_t *dst)
> > +{
> > +    SpiceChunks *chunks = bitmap->data;
> > +    uint32_t chunk_index = 0;
> > +    /* We can copy the bitmap chunk by chunk */
> > +    while (len && chunk_index < chunks->num_chunks) {
> > +        if (is_chunk_padded(bitmap, chunk_index)) {
> > +            return FALSE;
> > +        }
> > +        if (chunk_offset >= chunks->chunk[chunk_index].len) {
> > +            chunk_offset -= chunks->chunk[chunk_index].len;
> > +            chunk_index++;
> > +            continue;
> > +        }
> 
> I would split the loop here to make it more obvious that we have a first
> loop skipping chunks we need to ignore, and then a loop doing the copy.

Done.



-- 
Francois Gouget <fgouget at codeweavers.com>