[Ffmpeg-devel] [PATCH] THP PCM decoder (GSoC Qualification)

[Reimar Döffinger]

Hello,
On Mon, Apr 02, 2007 at 05:53:54PM +0200, Marco Gerards wrote:
> +    case CODEC_ID_ADPCM_THP:
> +      {
> +        GetBitContext gb;
> +        float table[16][2];
> +        int samplecnt;
> +        int prev1[2], prev2[2];

unaing "int prev[2][2];" might possibly simplify some code?

[...]
> +        for (ch = 0; ch < 2; ch++)
> +          for (i = 0; i < 16; i++) {
> +              /* Read the fixed point entry and store as floating
> +                 point.  */
> +              int entry = get_sbits(&gb, 16);
> +              table[i][ch] = (float) entry / pow(2, 11);
> +          }

Trying to avoid all that float stuff might be worth it. And at least pow
is a waste of CPU power, (1 << 11) should work just as well.

> +            while (sample > 0) {
> +                uint8_t index = get_bits (&gb, 4) & 7;
> +                float exp = pow(2, get_bits (&gb, 4));

float seems pointless, "int exp = 1 << get_bits(&gb, 4);" should work
the same?

> +                float factor1 = table[index * 2][ch];
> +                float factor2 = table[index * 2 + 1][ch];
> +        
> +                /* Decode 14 samples.  */
> +                for (n = 0; n < 14; n++) {
> +                    int sampledat = get_sbits (&gb, 4);
> +                    *samples = prev1[ch]*factor1 
> +                               + prev2[ch]*factor2 + sampledat * exp;

Actually multiplying with a power of two is a waste of CPU.
Any problem using
int exp = get_bits (&gb, 4);
above and
... + sampledat << exp;
here?

> +                    prev2[ch] = prev1[ch];
> +                    prev1[ch] = *samples++;
> +                    sample--;
> +                }

Hm. sample isn't checked here. Can that cause any problems? If not you
should probably just pull it out of the loop and do "sample -= 14;"

Greetings,
Reimar D?ffinger