[FFmpeg-devel] FW: [PATCH] 10-bit DNxHD decoding and encoding

[Michael Niedermayer]

On Wed, Mar 30, 2011 at 04:34:03PM +0000, Joseph Artsimovich wrote:
[...]
> diff --git a/libavcodec/faandct.c b/libavcodec/faandct.c
> index a986f65..dcd0e63 100644
> --- a/libavcodec/faandct.c
> +++ b/libavcodec/faandct.c
> @@ -30,7 +30,7 @@
>  
>  #define FLOAT float
>  #ifdef FAAN_POSTSCALE
> -#    define SCALE(x) postscale[x]
> +#    define SCALE(x) postscale8[x]
>  #else
>  #    define SCALE(x) 1
>  #endif
> @@ -56,7 +56,8 @@ for(i=0; i<8; i++){
>  #define A5 0.38268343236508977170 // cos(pi*6/16)
>  #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
>  
> -static const FLOAT postscale[64]={
> +// To get DCT coefficients multipled by 8.
> +static const FLOAT postscale8[64]={
>  B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
>  B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
>  B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
> @@ -67,6 +68,18 @@ B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
>  B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
>  };
>  
> +// To get DCT coefficients multiplied by 4.
> +static const FLOAT postscale4[64]={
> +0.5*B0*B0, 0.5*B0*B1, 0.5*B0*B2, 0.5*B0*B3, 0.5*B0*B4, 0.5*B0*B5, 0.5*B0*B6, 0.5*B0*B7,
> +0.5*B1*B0, 0.5*B1*B1, 0.5*B1*B2, 0.5*B1*B3, 0.5*B1*B4, 0.5*B1*B5, 0.5*B1*B6, 0.5*B1*B7,
> +0.5*B2*B0, 0.5*B2*B1, 0.5*B2*B2, 0.5*B2*B3, 0.5*B2*B4, 0.5*B2*B5, 0.5*B2*B6, 0.5*B2*B7,
> +0.5*B3*B0, 0.5*B3*B1, 0.5*B3*B2, 0.5*B3*B3, 0.5*B3*B4, 0.5*B3*B5, 0.5*B3*B6, 0.5*B3*B7,
> +0.5*B4*B0, 0.5*B4*B1, 0.5*B4*B2, 0.5*B4*B3, 0.5*B4*B4, 0.5*B4*B5, 0.5*B4*B6, 0.5*B4*B7,
> +0.5*B5*B0, 0.5*B5*B1, 0.5*B5*B2, 0.5*B5*B3, 0.5*B5*B4, 0.5*B5*B5, 0.5*B5*B6, 0.5*B5*B7,
> +0.5*B6*B0, 0.5*B6*B1, 0.5*B6*B2, 0.5*B6*B3, 0.5*B6*B4, 0.5*B6*B5, 0.5*B6*B6, 0.5*B6*B7,
> +0.5*B7*B0, 0.5*B7*B1, 0.5*B7*B2, 0.5*B7*B3, 0.5*B7*B4, 0.5*B7*B5, 0.5*B7*B6, 0.5*B7*B7,
> +};
> +
>  static av_always_inline void row_fdct(FLOAT temp[64], DCTELEM * data)
>  {
>      FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
> @@ -230,3 +243,61 @@ void ff_faandct248(DCTELEM * data)
>          data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
>      }
>  }
> +
> +void ff_faandct_10bit_safe(DCTELEM * data)
> +{
> +    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
> +    FLOAT tmp10, tmp11, tmp12, tmp13;
> +    FLOAT z2, z4, z11, z13;
> +    FLOAT av_unused z5;
> +    FLOAT temp[64];
> +    int i;
> +
> +    row_fdct(temp, data);
> +
> +    for (i=0; i<8; i++) {
> +        tmp0= temp[8*0 + i] + temp[8*7 + i];
> +        tmp7= temp[8*0 + i] - temp[8*7 + i];
> +        tmp1= temp[8*1 + i] + temp[8*6 + i];
> +        tmp6= temp[8*1 + i] - temp[8*6 + i];
> +        tmp2= temp[8*2 + i] + temp[8*5 + i];
> +        tmp5= temp[8*2 + i] - temp[8*5 + i];
> +        tmp3= temp[8*3 + i] + temp[8*4 + i];
> +        tmp4= temp[8*3 + i] - temp[8*4 + i];
> +
> +        tmp10= tmp0 + tmp3;
> +        tmp13= tmp0 - tmp3;
> +        tmp11= tmp1 + tmp2;
> +        tmp12= tmp1 - tmp2;
> +
> +        data[8*0 + i]= lrintf(postscale4[8*0 + i] * (tmp10 + tmp11));
> +        data[8*4 + i]= lrintf(postscale4[8*4 + i] * (tmp10 - tmp11));
> +
> +        tmp12 += tmp13;
> +        tmp12 *= A1;
> +        data[8*2 + i]= lrintf(postscale4[8*2 + i] * (tmp13 + tmp12));
> +        data[8*6 + i]= lrintf(postscale4[8*6 + i] * (tmp13 - tmp12));
> +
> +        tmp4 += tmp5;
> +        tmp5 += tmp6;
> +        tmp6 += tmp7;
> +
> +#if 0
> +        z5= (tmp4 - tmp6) * A5;
> +        z2= tmp4*A2 + z5;
> +        z4= tmp6*A4 + z5;
> +#else
> +        z2= tmp4*(A2+A5) - tmp6*A5;
> +        z4= tmp6*(A4-A5) + tmp4*A5;
> +#endif
> +        tmp5*=A1;
> +
> +        z11= tmp7 + tmp5;
> +        z13= tmp7 - tmp5;
> +
> +        data[8*5 + i]= lrintf(postscale4[8*5 + i] * (z13 + z2));
> +        data[8*3 + i]= lrintf(postscale4[8*3 + i] * (z13 - z2));
> +        data[8*1 + i]= lrintf(postscale4[8*1 + i] * (z11 + z4));
> +        data[8*7 + i]= lrintf(postscale4[8*7 + i] * (z11 - z4));
> +    }
> +}

this can be factored with the existing 8bit code, the only difference
is which table is used

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

I hate to see young programmers poisoned by the kind of thinking
Ulrich Drepper puts forward since it is simply too narrow -- Roman Shaposhnik
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20110330/d26ac8dd/attachment.asc>