Re: snd_pcm_hw_rule_noresample

[Raymond Yau <superquad.vortex2@gmail.com>]

2011/9/20 Clemens Ladisch <clemens@ladisch.de>:
> Raymond Yau wrote:
>> ALSA: pcm: add snd_pcm_hw_rule_noresample()
>> ...
>>
>> ymfpci and emu10k1 have SNDRV_PCM_RATE_CONTINUOUS and support 8000Hz to 48000Hz
>>
>> after this patch and if application using
>> snd_pcm_hw_params_set_rate_resample(,pcm , params ,0)
>>
>> why the supported rate is only 48000Hz ?
>
> Because these cards run at 48 kHz and resample anything else.  That
> the resampler is implemented in hardware does not make a conceptual
> difference, and just means that the resampler has low quality.
>
>> but snd_pcm_hw_params_get_rate_min() is still 8000Hz
>
> This should not happen.  I'd guess the rules engine hasn't noticed
> that some parameters affected by the new rule need to change.
>
> Please try the patch below.  But there might be drivers where other
> parameters are affected by SND_PCM_HW_PARAMS_NORESAMPLE, so I guess
> all parameter bits need to be set.
>
>> This patch won't affect the accuracy of the playback position of the ymfpci
>
> Why should it?
>

I think the correct way for ymfpci is to add a period_bytes constraint
(just like emu10k1 and au88x0) in addition of a switch to fixed rate
at 48000Hz to force the application to choose a period time which is
close to the interval of timer interrupt

 snd_pcm_hw_constraint_step(runtime, 0, SNDRV_PCM_HW_PARAM_PERIOD_BYTES, 1024);

are there any problem in snd_pcm_drain(), snd_pcm_rewind() and
snd_pcm_forward() when the driver let the application select any
period time when those hardware mixing sound card 's DSP have to mix
audio streams?

since hwptr is inaccurate, snd_pcm_rewindable() may need to return
zero if it can only rewind to period boundary