Re: PLL computation in TLV320AIC3x SoC driver

[Vladimir Barinov <vova.barinov@gmail.com>]

Peter Meerwald wrote:
> Hello Vladimir,
>
>   
>>> I'm trying to use the SoC TLV320AIC3x codec driver with sysclk 16384000 and
>>> ran into some problems with setting PLL; below is a patch against
>>> linux-2.6-asoc
>>>       
>
>   
>> I've made the simple test application to calculate pll_p/r/j/d values using
>> current tlv320aic3x clock calculation scheme and I've got:
>> samplerate=8000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=(sysclk *
>> [pll_j].[pll_d] * pll_r) / (2048 * pll_p) = 48000
>> samplerate=11025: pll_p=1, pll_r=1, pll_j=5, pll_d=5120, hence FSREF= 44096
>> samplerate=16000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=48000
>> samplerate=22050: pll_p=1, pll_r=1, pll_j=5, pll_d=5120, hence FSREF= 44096
>> samplerate=32000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=48000
>> samplerate=44100: pll_p=1, pll_r=1, pll_j=5, pll_d=5120, hence FSREF= 44096
>> samplerate=48000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=48000
>> samplerate=64000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=48000, note
>> that here we used  DUAL_RATE_MODE, hence FSREF=96000
>> samplerate=88200: pll_p=1, pll_r=1, pll_j=5, pll_d=5120, hence FSREF= 44096,
>> note that here we used  DUAL_RATE_MODE hence FSREF=88192
>> samplerate=96000: pll_p=1, pll_r=1, pll_j=6, pll_d=0, hence FSREF=48000, note
>> that here we used  DUAL_RATE_MODE hence FSREF = 96000
>>     
>
>   
>> Hence according to current code the calculations of desired FSREF is correct
>> for sysclk=16384000
>> I've not tried it with your patch since I don't actually understand what do
>> you fix? :)
>>     
>
> 1. loop variables are r and p but for computation of j the variables 
> pll_r and pll_p are used; I think r and p should be used in the 
> computation of j, your code below:
>
>     for (r = 1; r <= 16; r++)
>         for (p = 1; p <= 8; p++) {
>             int clk, tmp = (codec_clk * pll_r * 10) / pll_p;
>             u8 j = tmp / 10000;
>
> further, the codec datasheet mandates certain ranges for 'good 
> performance', 4 <= j <= 55 (for d==0) and 4 <= j <= 11 (for d!=0) which 
> are not checked for
>   
Correct.
> 2. d should actually be 5125 and not 5120
>   
Now I see, I've verified your code handles this. This actually produce 
more accuracy for FSREF.

I just tested your patch works fine with sysclk=16384000 and 12288000 
but fails with 11286900: it returns FSREF=45467 instead of desired 
44100. I think that this is a valuable difference. Please check it.

Regards,
Vladimir
> 3. appropriate is spell appropriately :)
>
>   
>>> note that the original code uses variables pll_r and pll_p instead of the
>>> loop variable r and p to compute tmp, this seems broken
>>>
>>> further, the original code does not respect the constraints on j (>= 4, <=
>>> 55 for d==0) according to the codec's datasheet, and similarly for d!=0
>>>
>>> I've tested the code with a number of reasonable sysclk values and got sane
>>> PLL values; please apply if acceptable
>>>   thanks, regards, p.
>>>
>>>
>>> diff --git a/sound/soc/codecs/tlv320aic3x.c b/sound/soc/codecs/tlv320aic3x.c
>>> index 3395cf9..e84e473 100644
>>> --- a/sound/soc/codecs/tlv320aic3x.c
>>> +++ b/sound/soc/codecs/tlv320aic3x.c
>>> @@ -766,9 +766,10 @@ static int aic3x_hw_params(struct snd_pcm_substream
>>> *substream,
>>>         struct snd_soc_codec *codec = socdev->card->codec;
>>>         struct aic3x_priv *aic3x = codec->private_data;
>>>         int codec_clk = 0, bypass_pll = 0, fsref, last_clk = 0;
>>> -       u8 data, r, p, pll_q, pll_p = 1, pll_r = 1, pll_j = 1;
>>> -       u16 pll_d = 1;
>>> +       u8 data, j, r, p, pll_q, pll_p = 1, pll_r = 1, pll_j = 1;
>>> +       u16 d, pll_d = 1;
>>>         u8 reg;
>>> +       int clk;
>>>          /* select data word length */
>>>         data =
>>> @@ -835,47 +836,62 @@ static int aic3x_hw_params(struct snd_pcm_substream
>>> *substream,
>>>                 return 0;
>>>          /* Use PLL
>>> -        * find an apropriate setup for j, d, r and p by iterating over
>>> -        * p and r - j and d are calculated for each fraction.
>>> -        * Up to 128 values are probed, the closest one wins the game.
>>> +        * find an appropriate setup for j, d, r and p by iterating over
>>> +        * p, r and j first, then trying to compute the fraction d.
>>> +        * Up to 6528 values are probed, the closest one wins the game.
>>>          * The sysclk is divided by 1000 to prevent integer overflows.
>>>          */
>>>         codec_clk = (2048 * fsref) / (aic3x->sysclk / 1000);
>>>  -       for (r = 1; r <= 16; r++)
>>> -               for (p = 1; p <= 8; p++) {
>>> -                       int clk, tmp = (codec_clk * pll_r * 10) / pll_p;
>>> -                       u8 j = tmp / 10000;
>>> -                       u16 d = tmp % 10000;
>>> -
>>> -                       if (j > 63)
>>> -                               continue;
>>> -
>>> -                       if (d != 0 && aic3x->sysclk < 10000000)
>>> -                               continue;
>>> -
>>> -                       /* This is actually 1000 * ((j + (d/10000)) * r) / p
>>> -                        * The term had to be converted to get rid of the
>>> -                        * division by 10000 */
>>> -                       clk = ((10000 * j * r) + (d * r)) / (10 * p);
>>> -
>>> -                       /* check whether this values get closer than the
>>> best
>>> -                        * ones we had before */
>>> -                       if (abs(codec_clk - clk) < abs(codec_clk -
>>> last_clk)) {
>>> -                               pll_j = j; pll_d = d; pll_r = r; pll_p = p;
>>> -                               last_clk = clk;
>>> -                       }
>>> -
>>> -                       /* Early exit for exact matches */
>>> -                       if (clk == codec_clk)
>>> -                               break;
>>> -               }
>>> +    for (r = 1; r <= 16; r++)
>>> +        for (p = 1; p <= 8; p++) {
>>> +            for (j = 4; j <= 55; j++) {
>>> +                /* This is actually 1000 * ((j + (d/10000)) * r) / p
>>> +                 * The term had to be converted to get rid of the
>>> +                 * division by 10000; d = 0 here */
>>> +                int clk = (1000 * j * r) / p;
>>> +
>>> +                /* check whether this values get closer than the best
>>> +                 * ones we had before */
>>> +                if (abs(codec_clk - clk) < abs(codec_clk - last_clk)) {
>>> +                    pll_j = j; pll_d = 0; pll_r = r; pll_p = p;
>>> +                    last_clk = clk;
>>> +                }
>>> +
>>> +                /* Early exit for exact matches */
>>> +                if (clk == codec_clk)
>>> +                    goto found;
>>> +            }
>>> +        }
>>> +
>>> +    /* try with d != 0 */
>>> +    for (p = 1; p <= 8; p++) {
>>> +        j = codec_clk * p / 1000;
>>> +
>>> +        if (j < 4 || j > 11) continue;
>>> +
>>> +        /* do not use codec_clk here since we'd loose precision */
>>> +        d = ((2048 * fsref * 10) / (aic3x->sysclk / 1000)) % 10000;
>>> +        clk = (10000 * j + d) / (10 * p);
>>> +
>>> +        /* check whether this values get closer than the best
>>> +         * ones we had before */
>>> +        if (abs(codec_clk - clk) < abs(codec_clk - last_clk)) {
>>> +            pll_j = j; pll_d = d; pll_r = 1; pll_p = 1;
>>> +            last_clk = clk;
>>> +        }
>>> +
>>> +        /* Early exit for exact matches */
>>> +        if (clk == codec_clk)
>>> +            goto found;
>>> +    }
>>>          if (last_clk == 0) {
>>>                 printk(KERN_ERR "%s(): unable to setup PLL\n", __func__);
>>>                 return -EINVAL;
>>>         }
>>>  +found:
>>>         data = aic3x_read_reg_cache(codec, AIC3X_PLL_PROGA_REG);
>>>         aic3x_write(codec, AIC3X_PLL_PROGA_REG, data | (pll_p <<
>>> PLLP_SHIFT));
>>>         aic3x_write(codec, AIC3X_OVRF_STATUS_AND_PLLR_REG, pll_r <<
>>> PLLR_SHIFT);
>>>
>>>   
>>>       
>
>