view faandct.c @ 10874:bcfe2acbf190 libavcodec

AAC: Compress codebook tables and optimise sign bit handling The codebooks each consist of small number of values repeated in groups of 2 or 4. Storing the codebooks as a packed list of 2- or 4-bit indexes into a table reduces their size substantially (from 7.5k to 1.5k), resulting in less cache pressure. For the band types with sign bits in the bitstream, storing the number and position of non-zero codebook values using a few bits avoids multiple get_bits() calls and floating-point comparisons which gcc handles miserably. Some float/int type punning also avoids gcc brain damage. Overall speedup 20-35% on Cortex-A8, 20% on Core i7.

/*
 * Floating point AAN DCT
 * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
 *
 * Copyright (c) 2003 Michael Niedermayer <[e-mail]>
 * Copyright (c) 2003 Roman Shaposhnik
 *
 * Permission to use, copy, modify, and/or distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */

/**
 * @file libavcodec/faandct.c
 * @brief
 *     Floating point AAN DCT
 * @author Michael Niedermayer <[e-mail]>
 */

#include "dsputil.h"
#include "faandct.h"

#define FLOAT float
#ifdef FAAN_POSTSCALE
#    define SCALE(x) postscale[x]
#else
#    define SCALE(x) 1
#endif

//numbers generated by simple c code (not as accurate as they could be)
/*
for(i=0; i<8; i++){
    printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
}
*/
#define B0 1.00000000000000000000
#define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
#define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
#define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
#define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
#define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
#define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
#define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1


#define A1 0.70710678118654752438 // cos(pi*4/16)
#define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
#define A5 0.38268343236508977170 // cos(pi*6/16)
#define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)

static const FLOAT postscale[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,
B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
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,
};

static av_always_inline void row_fdct(FLOAT temp[64], 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;
    int i;

    for (i=0; i<8*8; i+=8) {
        tmp0= data[0 + i] + data[7 + i];
        tmp7= data[0 + i] - data[7 + i];
        tmp1= data[1 + i] + data[6 + i];
        tmp6= data[1 + i] - data[6 + i];
        tmp2= data[2 + i] + data[5 + i];
        tmp5= data[2 + i] - data[5 + i];
        tmp3= data[3 + i] + data[4 + i];
        tmp4= data[3 + i] - data[4 + i];

        tmp10= tmp0 + tmp3;
        tmp13= tmp0 - tmp3;
        tmp11= tmp1 + tmp2;
        tmp12= tmp1 - tmp2;

        temp[0 + i]= tmp10 + tmp11;
        temp[4 + i]= tmp10 - tmp11;

        tmp12 += tmp13;
        tmp12 *= A1;
        temp[2 + i]= tmp13 + tmp12;
        temp[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;

        temp[5 + i]= z13 + z2;
        temp[3 + i]= z13 - z2;
        temp[1 + i]= z11 + z4;
        temp[7 + i]= z11 - z4;
    }
}

void ff_faandct(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;

    emms_c();

    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(SCALE(8*0 + i) * (tmp10 + tmp11));
        data[8*4 + i]= lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));

        tmp12 += tmp13;
        tmp12 *= A1;
        data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
        data[8*6 + i]= lrintf(SCALE(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(SCALE(8*5 + i) * (z13 + z2));
        data[8*3 + i]= lrintf(SCALE(8*3 + i) * (z13 - z2));
        data[8*1 + i]= lrintf(SCALE(8*1 + i) * (z11 + z4));
        data[8*7 + i]= lrintf(SCALE(8*7 + i) * (z11 - z4));
    }
}

void ff_faandct248(DCTELEM * data)
{
    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
    FLOAT tmp10, tmp11, tmp12, tmp13;
    FLOAT temp[64];
    int i;

    emms_c();

    row_fdct(temp, data);

    for (i=0; i<8; i++) {
        tmp0 = temp[8*0 + i] + temp[8*1 + i];
        tmp1 = temp[8*2 + i] + temp[8*3 + i];
        tmp2 = temp[8*4 + i] + temp[8*5 + i];
        tmp3 = temp[8*6 + i] + temp[8*7 + i];
        tmp4 = temp[8*0 + i] - temp[8*1 + i];
        tmp5 = temp[8*2 + i] - temp[8*3 + i];
        tmp6 = temp[8*4 + i] - temp[8*5 + i];
        tmp7 = temp[8*6 + i] - temp[8*7 + i];

        tmp10 = tmp0 + tmp3;
        tmp11 = tmp1 + tmp2;
        tmp12 = tmp1 - tmp2;
        tmp13 = tmp0 - tmp3;

        data[8*0 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
        data[8*4 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));

        tmp12 += tmp13;
        tmp12 *= A1;
        data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
        data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));

        tmp10 = tmp4 + tmp7;
        tmp11 = tmp5 + tmp6;
        tmp12 = tmp5 - tmp6;
        tmp13 = tmp4 - tmp7;

        data[8*1 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
        data[8*5 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));

        tmp12 += tmp13;
        tmp12 *= A1;
        data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
        data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
    }
}