[Ffmpeg-devel] [RFC] IMC decoder

[Benjamin Larsson]

Kostya skrev:
> On Mon, Oct 30, 2006 at 01:26:51AM +0100, Michael Niedermayer wrote:
>   
>> Hi
>>
>> On Sun, Oct 29, 2006 at 08:35:11AM +0200, Kostya wrote:
>>     
>>> Here is Intel Music Codec decoder RE'd by Maxim Poliakovski, FFmpeg'ized
>>> by Benjamin Larsson, cleaned and fixed by me.
>>>       

First thanks for cleaning up my code. And here are some comments.

>>     
>> ------------------------------------------------------------------------
>>
>> Index: Changelog
>> ===================================================================
>> --- Changelog	(revision 6830)
>> +++ Changelog	(working copy)
>> @@ -65,6 +65,7 @@
>>  - MTV demuxer
>>  - TIFF picture decoder
>>  - GIF picture decoder
>> +- Intel Music decoder
>>     

I think that the official name of the codec is "Intel Music Coder".

>>  
>> [...]
>>  
>> +
>> +/**
>> + *  @file imc.c IMC - Intel Music Codec
>>     
s/Codec/Coder/
>> + *  A mdct based codec using a 256 points large transform
>> + *  divied into 32 bands with some mix of scale factors.
>> + *  Only mono is supported.
>> + *
>> + */[...]

[...]

>> +typedef struct {
>> +    float old_floor[BANDS];
>> +    float flcoeffs1[BANDS];
>> +    float flcoeffs2[BANDS];
>> +    float flcoeffs3[BANDS];
>> +    float flcoeffs4[BANDS];
>> +    float flcoeffs5[BANDS];
>> +    float flcoeffs6[BANDS];
>> +    float CWdecoded[COEFFS];
>> +
>> +    /** MDCT tables */
>> +    //@{
>> +    float mdct_sine_window[COEFFS];
>> +    float post_coef1[COEFFS];
>> +    float post_coef2[COEFFS];
>> +    float pre_coef1[COEFFS];
>> +    float pre_coef2[COEFFS];
>> +    float last_fft_im[COEFFS];
>>     

last_fft_im is the old non overlapped non windowed samples that is used 
in the last part of the mdct so the last_fft_im name might not be the best.
And the coef part of the twiddle factors (post/pre_*) could be renamed 
*_sin or *_cos depending on what they are.

>> +
>> +void imc_imdct256(IMCContext *q) {
>> +    int i;
>> +    float re, im;
>> +
>> +    /* prerotation */
>> +    for(i=0; i < COEFFS/2; i++){
>> +        q->samples[i].re = -(q->pre_coef1[i] * q->CWdecoded[COEFFS-1-i*2]) -
>> +                           (q->pre_coef2[i] * q->CWdecoded[i*2]);
>> +        q->samples[i].im = (q->pre_coef2[i] * q->CWdecoded[COEFFS-1-i*2]) -
>> +                           (q->pre_coef1[i] * q->CWdecoded[i*2]);
>> +    }
>> +
>> +    /* FFT */
>> +    ff_fft_permute(&q->fft, q->samples);
>> +    ff_fft_calc (&q->fft, q->samples);
>> +
>> +    /* postrotation, window and reorder */
>> +    for(i = 0; i < COEFFS/2; i++){
>> +        re = (q->samples[i].re * q->post_coef1[i]) + (-q->samples[i].im * q->post_coef2[i]);
>> +        im = (-q->samples[i].im * q->post_coef1[i]) - (q->samples[i].re * q->post_coef2[i]);
>> +        q->out_samples[i*2] = (q->mdct_sine_window[COEFFS-1-i*2] * q->last_fft_im[i]) + (q->mdct_sine_window[i*2] * re);
>> +        q->out_samples[COEFFS-1-i*2] = (q->mdct_sine_window[i*2] * q->last_fft_im[i]) - (q->mdct_sine_window[COEFFS-1-i*2] * re);
>> +        q->last_fft_im[i] = im;
>> +    }
>> +}
>>     

This looks alot like the ffmpeg imdct function with the exception of the 
postrotation, window and reorder part. The thing that can be tried is to 
like the do_imdct_512 function in the soc ac3 decoder. If that doesn't 
work then the transforms aren't identical and the code can't be replaced.
>> +
>> +
>> +static int imc_decode_close(AVCodecContext * avctx)
>> +{
>> +    IMCContext *q = avctx->priv_data;
>> +
>> +    ff_fft_end(&q->fft);
>>     

Shouldn't the VLC table be free'd also ?

>> +    return 0;
>> +}
>> +
>> +
>> +AVCodec imc_decoder = {
>> +    .name = "imc",
>> +    .type = CODEC_TYPE_AUDIO,
>> +    .id = CODEC_ID_IMC,
>> +    .priv_data_size = sizeof(IMCContext),
>> +    .init = imc_decode_init,
>> +    .close = imc_decode_close,
>> +    .decode = imc_decode_frame,
>> +};
>> Index: libavcodec/avcodec.h
>> ===================================================================
>> --- libavcodec/avcodec.h	(revision 6830)
>> +++ libavcodec/avcodec.h	(working copy)
>> @@ -233,6 +233,7 @@
>>      CODEC_ID_QCELP,
>>      CODEC_ID_WAVPACK,
>>      CODEC_ID_DSICINAUDIO,
>> +    CODEC_ID_IMC,
>>  
>>      /* subtitle codecs */
>>      CODEC_ID_DVD_SUBTITLE= 0x17000,
>> @@ -2304,6 +2305,7 @@
>>  extern AVCodec dsicinaudio_decoder;
>>  extern AVCodec tiertexseqvideo_decoder;
>>  extern AVCodec tiff_decoder;
>> +extern AVCodec imc_decoder;
>>  
>>  /* pcm codecs */
>>  #define PCM_CODEC(id, name) \
>> Index: doc/ffmpeg-doc.texi
>> ===================================================================
>> --- doc/ffmpeg-doc.texi	(revision 6830)
>> +++ doc/ffmpeg-doc.texi	(working copy)
>> @@ -934,6 +934,7 @@
>>  @item WavPack Audio          @tab      @tab X
>>  @item Cin Audio              @tab      @tab X
>>  @tab Codec used in Delphine Software games.
>> + at item Intel Music Codec      @tab      @tab X
>>     

s/Codec/Coder/

MvH
Benjamin Larsson