right, but the key point is 'switch between streams'. That means a
more complex/capable implementation that should be advertised as such
to applications. This is not the default behavior assumed initially:
to allow for minimal implementations in memory-constrained devices, we
assumed gapless was supported with a single decoder.
Maybe the right way to do this is extend the snd_compr_caps structure:
/**
* struct snd_compr_caps - caps descriptor
* @codecs: pointer to array of codecs
* @direction: direction supported. Of type snd_compr_direction
* @min_fragment_size: minimum fragment supported by DSP
* @max_fragment_size: maximum fragment supported by DSP
* @min_fragments: min fragments supported by DSP
* @max_fragments: max fragments supported by DSP
* @num_codecs: number of codecs supported
* @reserved: reserved field
*/
struct snd_compr_caps {
__u32 num_codecs;
__u32 direction;
__u32 min_fragment_size;
__u32 max_fragment_size;
__u32 min_fragments;
__u32 max_fragments;
__u32 codecs[MAX_NUM_CODECS];
__u32 reserved[11];
} __attribute__((packed, aligned(4)));
and use a reserved field to provide info on capabilities, and filter
the set_codec_params() addition based this capability - i.e. return
-ENOTSUP in 'traditional' implementations based on a single
'stream'/decoder instance.
I think this is also what Mark was referring to earlier.
Sounds good!
I will give it a go and see how it ends up!
Glad to see this discussion progressing.
We may also want to document the 3 possible ways of supporting gapless
playback while we are at it:
a) with the existing single decoder assumption
b) with your suggested solution with a switch at the DSP level
c) with 2 streams at the userspace level and a switch/x-fade at the DSP
level - which may simplify userspace quite a bit and was the initial
design in a non-Linux OS.
