[Sound-open-firmware] how to find a suitable value for the TFT/RFT parameter on multi-platform

[Pierre-Louis Bossart]

On 05/17/2018 08:30 PM, zhigangw wrote:
>
>
> On 2018年05月17日 21:25, Pierre-Louis Bossart wrote:
>> On 5/17/18 4:31 AM, zhigangw wrote:
>>> Hello Liam:
>>>
>>>          I think it is better to mention this status again in the 
>>> mail list.
>>>
>>> I am working on playback with the TDM8(4slot used)-48KHz-32bit on 
>>> APL-GPMRB platform.
>>>
>>> Currently the status is:
>>>
>>> The music can be heard from the all four channels.
>>>
>>> But the it has the noise on the background. the noise exists all the 
>>> time during playing.
>>>
>>> No xrun happened during the audio playing.
>>>
>>> I have not tried the pause/resume feature.
>>>
>>> I have not tried keyon's patch for the noise issue.
>>
>> Looks to me like you've answered to your own question:
>> if there is no xrun, the problem is not related to TFT settings, so 
>> please try with known workarounds to remove the noise issue.
>>
>
> Sorry! I think I miss one point in this description:
> After I set the TFT = 4 hardcoded, the play can continue. If I follow 
> the old code, the TFT = 16.
> the XRUN will happen 100%, the DMA IRQ interval is less than 1ms,
> the XRUN is caused by this.
> the noise issue will be a next step, i will try keyon's patch later.

Well yes, the FIFO is 16-deep so if you set TFT to 16 you essentially 
have no sample left in the FIFO..
Probably the formula should be TFT = min  (2 * active_tx_slot, 8); or 
something along those lines.
>>>
>>>
>>> Issue:
>>>
>>> During the development, the TFT (in SSP's SSC3 register) parameter 
>>> will affect the playing operation.
>>>
>>> If we keep code: tft = active_tx_slots * 4;  the tft will be 16.
>>>
>>> during the playing, i found the DMA IRQ will happened very 
>>> frequently, less than 1ms,
>>>
>>> which will cause XRUN during playing. But it will be normal if the 
>>> tft value is below 8 (tft < 8).
>>
>> Doesn't this point to a DMA configuration issue? the IRQ should only 
>> come after a set of burst transfers related to the TFT settings. In 
>> other words, if you reduce TFT you should increase the number of 
>> bursts to keep the IRQ rate constant.
>>
>> There is a clear note in the code on this btw, the RFT/TFT settings 
>> cannot be modified in isolation, you need to have a related change on 
>> the DMA side.
>>
> This is not related to DMA configuration, it is about the SSP 
> configuration.
> I did the experiment on my own APL-GPMRB, change the TFT will affect 
> the DMA's IRQ interval.
> I will check the DMA's configuration about the burst you mentioned.
>
> there is communication mechanism between the SSP and DMA,
> when the TFT (FIFO threshold) is reached, the SSP will send request 
> signal to DMA, the DMA will start to coping (into SSP's FIFO).
> this mechanism is black to software level.
> the TFT/RFT configuration determined when SSP sends request to DMA,
>
> thanks
> br
> ~zhigang
>
>>
>>>
>>>
>>> I did some tests on other platform. but I have not found suitable 
>>> way to fit all of them.
>>>
>>> the RFT also have the similar concern.
>>>
>>>
>>> Do you have any suggestions about this?
>>>
>>>
>>> thank you
>>>
>>> br
>>>
>>> ~zhigang
>>>
>>>
>>>
>>> _______________________________________________
>>> Sound-open-firmware mailing list
>>> Sound-open-firmware at alsa-project.org
>>> http://mailman.alsa-project.org/mailman/listinfo/sound-open-firmware
>>
>