[FFmpeg-devel] [PATCH] Some ra144.c simplifications

[Vitor Sessak]

Michael Niedermayer wrote:
> On Wed, Jun 04, 2008 at 08:18:10PM +0200, Vitor Sessak wrote:
>> Michael Niedermayer wrote:
>>> On Wed, May 28, 2008 at 09:23:02PM +0200, Vitor Sessak wrote:
>>>> Michael Niedermayer wrote:
>>>>> On Wed, May 28, 2008 at 06:56:45PM +0200, Vitor Sessak wrote:
>>>>>> Michael Niedermayer wrote:
>>>>>>> On Tue, May 27, 2008 at 09:16:09PM +0200, Vitor Sessak wrote:
>>>>>>>> Michael Niedermayer wrote:
>>>>>>>>> On Sun, May 25, 2008 at 07:11:52PM +0200, Vitor Sessak wrote:
>>>>>>>>>> Michael Niedermayer wrote:
>>>>>>>>>>> On Sun, May 25, 2008 at 06:05:15PM +0200, Vitor Sessak wrote:
>>>>>>>>>>> [...]
>>>>>>>>>>>>>> ok
>>>>>>>>>>>>> One more...
>>>>>>>>>>>> ... and some more cleanup:
>>>>>>>>>>>>
>>>>>>>>>>>> ra144_vector_add_wav.diff: Make add_wav() receive a vector 
>>>>>>>>>>>> instead of three integers
>>>>>>>>>>>>
>>>>>>>>>>>> ra144_params_dec2.diff: Do not calculate anything based in l, it 
>>>>>>>>>>>> is unrolled in the loop anyway
>>>>>>>>>>> ok
>>>>>>>>>> Now s/(unsigned) short/(u)int16_t.
>>>>>>>>> ok
>>>>>>>> Next one. dec2() interpolates the block coefficients from the 
>>>>>>>> previous one and fall back to a block-dependent schema if the 
>>>>>>>> interpolation results in an unstable filter...
>>>>>>> [...]
>>>>>>>> +    // Interpolate block coefficients from the this frame forth 
>>>>>>>> block and
>>>>>>>> +    // last frame forth block
>>>>>>>>      for (x=0; x<30; x++)
>>>>>>>> -        decsp[x] = (a * inp[x] + b * inp2[x]) >> 2;
>>>>>>>> +        decsp[x] = (a * ractx->lpc_coef[x] + b * 
>>>>>>>> ractx->lpc_coef_old[x])>> 2;
>>>>>>> ff_acelp_weighted_vector_sum()
>>>>>> Ok, but to do that I need to use int16_t. So I propose to apply my 
>>>>>> original patch and then the attached one.
>>>>> hmm, ok
>>>> Done. Now remove the dec1() function (that was memcpy + 1 line of code). 
>>>> As a side effect, it removes the need of a memcpy (the dec1() call at 
>>>> decode_frame()).
>>> ok
>> Now the first patch (ra144_rescale_energy.diff) split the energy rescaling 
>> out of the rms() function. The next patch remove *lpc_refl from the 
>> context, since the only thing needed from the last frame is the non 
>> rescaled output of rms().
> 
> ok

Now, I'm almost finished with this. Two things remains:

1- When decoding a ra144 encoded file, ffmpeg produces lots of "Multiple 
frames in a packet from stream 0" (see 
http://fate.multimedia.cx/index.php?test_result=1911120 for an example). 
This is because the decoder receives a 1000 byte sample and decode only 
20 bytes. The attached patch fix this (it decode all the 50 blocks).

2- There are lots of unused table entries. Ok to remove them or do you 
thing they can useful for anything (another codec?)?

Finally, if there is anything else you don't like about ra144.{c,h}, now 
is the time to say if you want me to have a look at it...

-Vitor
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ra144_blocks.diff
Type: text/x-patch
Size: 3083 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080621/d19bd999/attachment.bin>