On Wed, 2011-11-30 at 12:19 +0100, Takashi Iwai wrote:
At Tue, 29 Nov 2011 15:15:01 -0600, Pierre-Louis Bossart wrote: Well, my question is much more simple. In the patch, there are checks of the stream direction in snd_compr_{fragment|frame}_elapsed() functions.
+void snd_compr_fragment_elapsed(struct snd_compr_stream *stream) +{
- pr_debug("fragment elapsed called\n");
- if (stream->direction != SNDRV_DEVICE_TYPE_COMPR_PLAYBACK)
return;- wake_up(&stream->runtime->sleep);
+} +EXPORT_SYMBOL_GPL(snd_compr_fragment_elapsed);
+void snd_compr_frame_elapsed(struct snd_compr_stream *stream) +{
- if (stream->direction != SNDRV_DEVICE_TYPE_COMPR_CAPTURE)
return;- wake_up(&stream->runtime->sleep);
+} +EXPORT_SYMBOL_GPL(snd_compr_frame_elapsed);
And I wonder why this restriction *must* be present in the API-level implementation.
And, furthermore, I see no reason to have this individual functions. Why simply not an inline function like
static inline void snd_compr_stream_elapsed(struct snd_compr_stream *stream) { wake_up(&stream->runtime->sleep); }
?? Then you can reduce two exported symbols and less stack usages.
Thanks, I have already changed this as above based on your suggestion last week :)
But I have kept two different functions doing same thing for sake of confusion. For playback you need to call this on block/fragment/random_bytes but for capture it needs to be frame based.