[FFmpeg-devel] Review request - ra288.{c,h} ra144.{c,h}

[Michael Niedermayer]

On Sat, Sep 13, 2008 at 07:07:26PM +0200, Vitor Sessak wrote:
> Michael Niedermayer wrote:
>> On Fri, Sep 05, 2008 at 12:23:58AM +0200, Vitor Sessak wrote:
>>> Vitor Sessak wrote:
[...]
>> [...]
>>> static void colmult(float *tgt, const float *m1, const float *m2, int n)
>>> {
>>>     while (n--)
>>>         *tgt++ = *m1++ * *m2++;
>>> }
>> such function is commonly called apply_window() in other codecs
>>> static void decode(RA288Context *ractx, float gain, int cb_coef)
>>> {
>>>     int i, j;
>>>     double sumsum;
>>>     float sum, buffer[5];
>>>     float *block = ractx->sp_block + 36; // Current block
>>>
>>>     memmove(ractx->sp_block, ractx->sp_block + 5, 
>>> 36*sizeof(*ractx->sp_block));
>>>
>>>     for (i=0; i < 5; i++) {
>>>         block[i] = 0.;
>>>         for (j=0; j < 36; j++)
>>>             block[i] -= block[i-1-j]*ractx->sp_lpc[j];
>>>     }
>>>
>>>     /* block 46 of G.728 spec */
>>>     sum = 32.;
>>>     for (i=0; i < 10; i++)
>>>         sum -= ractx->gain_block[9-i] * ractx->gain_lpc[i];
>>>
>>>     /* block 47 of G.728 spec */
>>>     sum = av_clipf(sum, 0, 60);
>>>
>>>     /* block 48 of G.728 spec */
>>>     sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*gain 
>>> */
>>>
>>>     for (i=0; i < 5; i++)
>>>         buffer[i] = codetable[cb_coef][i] * sumsum;
>>>
>>>     sum = scalar_product_float(buffer, buffer, 5) / 5;
>>>
>>>     sum = FFMAX(sum, 1);
>>>
>>>     /* shift and store */
>>>     memmove(ractx->gain_block, ractx->gain_block + 1,
>>>             9 * sizeof(*ractx->gain_block));
>>>
>>>     ractx->gain_block[9] = 10 * log10(sum) - 32;
>>>
>>>     for (i=1; i < 5; i++)
>>>         for (j=i-1; j >= 0; j--)
>>>             buffer[i] -= ractx->sp_lpc[i-j-1] * buffer[j];
>>>
>>>     /* output */
>>>     for (i=0; i < 5; i++)
>>>         block[i] = av_clipf(block[i] + buffer[i], -4095, 4095);
>> can the buffer values be stored in block and sp_lpc applied over both
>> in one pass instead of this 2 pass and add-clip thing?
>
> I can't apply sp_lpc to buffer+block, so I need two buffers...

What i was thinking about was:

    /* block 46 of G.728 spec */
    sum = 32.;
    for (i=0; i < 10; i++)
        sum -= gain_block[9-i] * ractx->gain_lpc[i];

    /* block 47 of G.728 spec */
    sum = av_clipf(sum, 0, 60);

    /* block 48 of G.728 spec */
    sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*gain */

    for (i=0; i < 5; i++)
        buffer[i] = codetable[cb_coef][i] * sumsum * (1./2048.);

    sum = scalar_product_float(buffer, buffer, 5) / 5;

    sum = FFMAX(sum, 1);

    /* shift and store */
    memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));

    gain_block[9] = 10 * log10(sum) - 32;

    for (i=0; i < 5; i++) {
        block[i] = buffer[i];
        for (j=0; j < 36; j++)
            block[i] -= block[i-1-j]*ractx->sp_lpc[j];
    }

    /* output */
    for (i=0; i < 5; i++)
        block[i] = av_clipf(block[i], -4095, 4095);

am i missing a proble in this?



>
>> [...]
>>> static int ra288_decode_frame(AVCodecContext * avctx, void *data,
>>>                               int *data_size, const uint8_t * buf,
>>>                               int buf_size)
>>> {
>>>     int16_t *out = data;
>>>     int i, j;
>>>     RA288Context *ractx = avctx->priv_data;
>>>     GetBitContext gb;
>>>
>>>     if (buf_size < avctx->block_align) {
>>>         av_log(avctx, AV_LOG_ERROR,
>>>                "Error! Input buffer is too small [%d<%d]\n",
>>>                buf_size, avctx->block_align);
>>>         return 0;
>>>     }
>>>
>>>     if (*data_size < 32*5*2)
>>>         return -1;
>>>
>>>     init_get_bits(&gb, buf, avctx->block_align * 8);
>>>
>>>     for (i=0; i < 32; i++) {
>>>         float gain = amptable[get_bits(&gb, 3)];
>>>         int cb_coef = get_bits(&gb, 6 + (i&1));
>>>
>>>         decode(ractx, gain, cb_coef);
>>>
>>>         for (j=0; j < 5; j++)
>>>             *(out++) = 8 * ractx->sp_block[36 + j];
>> if float output works already, then this could output floats, if not then
>> this could use lrintf()
>
> I've tried the float output (with the attached patch) and it didn't work. 

ok


> Using lrint() changes slightly the output (PSNR about 99), is it expected?

yes, it does round differently (=more correctly)


[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Complexity theory is the science of finding the exact solution to an
approximation. Benchmarking OTOH is finding an approximation of the exact
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080913/3f13d4af/attachment.pgp>