view faandct.c @ 1757:3906ddbaffec libavcodec

optimization & bugfix extracted from the 4k line diff between ffmpeg 0.4.7 and http://www.alicestreet.com/ffh263.html the other parts of the diff where 1. spelling fixes (rejected as only a small part of it could be applied automatically) 2. cosmetics (reindention, function reordering, var renaming, ...) with bugs (rejected) 3. rtp related stuff (rejetced as it breaks several codecs) 4. some changes to the intra/inter decission & scene change detection (quality tests needed first)

/*
 * Floating point AAN DCT
 * Copyright (c) 2003 Michael Niedermayer <[e-mail]>
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 *
 * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
 */

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

#include <math.h>

#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 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 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 z1, z2, z3, z4, z5, z11, z13;
    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;
        
        z1= (tmp12 + tmp13)*A1;
        temp[2 + i]= tmp13 + z1;
        temp[6 + i]= tmp13 - z1;
        
        tmp10= tmp4 + tmp5;
        tmp11= tmp5 + tmp6;
        tmp12= tmp6 + tmp7;

        z5= (tmp10 - tmp12) * A5;
        z2= tmp10*A2 + z5;
        z4= tmp12*A4 + z5;
        z3= tmp11*A1;

        z11= tmp7 + z3;
        z13= tmp7 - z3;

        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 z1, z2, z3, z4, z5, z11, z13;
    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));
        
        z1= (tmp12 + tmp13)* A1;
        data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + z1));
        data[8*6 + i]= lrintf(SCALE(8*6 + i) * (tmp13 - z1));
        
        tmp10= tmp4 + tmp5;
        tmp11= tmp5 + tmp6;
        tmp12= tmp6 + tmp7;

        z5= (tmp10 - tmp12) * A5;
        z2= tmp10*A2 + z5;
        z4= tmp12*A4 + z5;
        z3= tmp11*A1;

        z11= tmp7 + z3;
        z13= tmp7 - z3;

        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 z1;
    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));
        
        z1 = (tmp12 + tmp13)* A1;
        data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + z1));
        data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - z1));
        
        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));

	z1 = (tmp12 + tmp13)* A1;
	data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + z1));
	data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - z1));
    }
}