yading@10: /*
yading@10:  * Floating point AAN DCT
yading@10:  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
yading@10:  *
yading@10:  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
yading@10:  * Copyright (c) 2003 Roman Shaposhnik
yading@10:  *
yading@10:  * Permission to use, copy, modify, and/or distribute this software for any
yading@10:  * purpose with or without fee is hereby granted, provided that the above
yading@10:  * copyright notice and this permission notice appear in all copies.
yading@10:  *
yading@10:  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
yading@10:  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
yading@10:  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
yading@10:  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
yading@10:  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
yading@10:  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
yading@10:  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * @brief
yading@10:  *     Floating point AAN DCT
yading@10:  * @author Michael Niedermayer <michaelni@gmx.at>
yading@10:  */
yading@10: 
yading@10: #include "faandct.h"
yading@10: #include "libavutil/internal.h"
yading@10: #include "libavutil/libm.h"
yading@10: 
yading@10: #define FLOAT float
yading@10: 
yading@10: //numbers generated by simple c code (not as accurate as they could be)
yading@10: /*
yading@10: for(i=0; i<8; i++){
yading@10:     printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
yading@10: }
yading@10: */
yading@10: #define B0 1.00000000000000000000
yading@10: #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
yading@10: #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
yading@10: #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
yading@10: #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
yading@10: #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
yading@10: #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
yading@10: #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
yading@10: 
yading@10: 
yading@10: #define A1 0.70710678118654752438 // cos(pi*4/16)
yading@10: #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
yading@10: #define A5 0.38268343236508977170 // cos(pi*6/16)
yading@10: #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
yading@10: 
yading@10: static const FLOAT postscale[64]={
yading@10: B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
yading@10: B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
yading@10: B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
yading@10: B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
yading@10: B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
yading@10: B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
yading@10: B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
yading@10: B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
yading@10: };
yading@10: 
yading@10: static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
yading@10: {
yading@10:     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
yading@10:     FLOAT tmp10, tmp11, tmp12, tmp13;
yading@10:     FLOAT z2, z4, z11, z13;
yading@10:     FLOAT av_unused z5;
yading@10:     int i;
yading@10: 
yading@10:     for (i=0; i<8*8; i+=8) {
yading@10:         tmp0= data[0 + i] + data[7 + i];
yading@10:         tmp7= data[0 + i] - data[7 + i];
yading@10:         tmp1= data[1 + i] + data[6 + i];
yading@10:         tmp6= data[1 + i] - data[6 + i];
yading@10:         tmp2= data[2 + i] + data[5 + i];
yading@10:         tmp5= data[2 + i] - data[5 + i];
yading@10:         tmp3= data[3 + i] + data[4 + i];
yading@10:         tmp4= data[3 + i] - data[4 + i];
yading@10: 
yading@10:         tmp10= tmp0 + tmp3;
yading@10:         tmp13= tmp0 - tmp3;
yading@10:         tmp11= tmp1 + tmp2;
yading@10:         tmp12= tmp1 - tmp2;
yading@10: 
yading@10:         temp[0 + i]= tmp10 + tmp11;
yading@10:         temp[4 + i]= tmp10 - tmp11;
yading@10: 
yading@10:         tmp12 += tmp13;
yading@10:         tmp12 *= A1;
yading@10:         temp[2 + i]= tmp13 + tmp12;
yading@10:         temp[6 + i]= tmp13 - tmp12;
yading@10: 
yading@10:         tmp4 += tmp5;
yading@10:         tmp5 += tmp6;
yading@10:         tmp6 += tmp7;
yading@10: 
yading@10: #if 0
yading@10:         z5= (tmp4 - tmp6) * A5;
yading@10:         z2= tmp4*A2 + z5;
yading@10:         z4= tmp6*A4 + z5;
yading@10: #else
yading@10:         z2= tmp4*(A2+A5) - tmp6*A5;
yading@10:         z4= tmp6*(A4-A5) + tmp4*A5;
yading@10: #endif
yading@10:         tmp5*=A1;
yading@10: 
yading@10:         z11= tmp7 + tmp5;
yading@10:         z13= tmp7 - tmp5;
yading@10: 
yading@10:         temp[5 + i]= z13 + z2;
yading@10:         temp[3 + i]= z13 - z2;
yading@10:         temp[1 + i]= z11 + z4;
yading@10:         temp[7 + i]= z11 - z4;
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_faandct(int16_t *data)
yading@10: {
yading@10:     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
yading@10:     FLOAT tmp10, tmp11, tmp12, tmp13;
yading@10:     FLOAT z2, z4, z11, z13;
yading@10:     FLOAT av_unused z5;
yading@10:     FLOAT temp[64];
yading@10:     int i;
yading@10: 
yading@10:     emms_c();
yading@10: 
yading@10:     row_fdct(temp, data);
yading@10: 
yading@10:     for (i=0; i<8; i++) {
yading@10:         tmp0= temp[8*0 + i] + temp[8*7 + i];
yading@10:         tmp7= temp[8*0 + i] - temp[8*7 + i];
yading@10:         tmp1= temp[8*1 + i] + temp[8*6 + i];
yading@10:         tmp6= temp[8*1 + i] - temp[8*6 + i];
yading@10:         tmp2= temp[8*2 + i] + temp[8*5 + i];
yading@10:         tmp5= temp[8*2 + i] - temp[8*5 + i];
yading@10:         tmp3= temp[8*3 + i] + temp[8*4 + i];
yading@10:         tmp4= temp[8*3 + i] - temp[8*4 + i];
yading@10: 
yading@10:         tmp10= tmp0 + tmp3;
yading@10:         tmp13= tmp0 - tmp3;
yading@10:         tmp11= tmp1 + tmp2;
yading@10:         tmp12= tmp1 - tmp2;
yading@10: 
yading@10:         data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
yading@10:         data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
yading@10: 
yading@10:         tmp12 += tmp13;
yading@10:         tmp12 *= A1;
yading@10:         data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
yading@10:         data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
yading@10: 
yading@10:         tmp4 += tmp5;
yading@10:         tmp5 += tmp6;
yading@10:         tmp6 += tmp7;
yading@10: 
yading@10: #if 0
yading@10:         z5= (tmp4 - tmp6) * A5;
yading@10:         z2= tmp4*A2 + z5;
yading@10:         z4= tmp6*A4 + z5;
yading@10: #else
yading@10:         z2= tmp4*(A2+A5) - tmp6*A5;
yading@10:         z4= tmp6*(A4-A5) + tmp4*A5;
yading@10: #endif
yading@10:         tmp5*=A1;
yading@10: 
yading@10:         z11= tmp7 + tmp5;
yading@10:         z13= tmp7 - tmp5;
yading@10: 
yading@10:         data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
yading@10:         data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
yading@10:         data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
yading@10:         data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_faandct248(int16_t *data)
yading@10: {
yading@10:     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
yading@10:     FLOAT tmp10, tmp11, tmp12, tmp13;
yading@10:     FLOAT temp[64];
yading@10:     int i;
yading@10: 
yading@10:     emms_c();
yading@10: 
yading@10:     row_fdct(temp, data);
yading@10: 
yading@10:     for (i=0; i<8; i++) {
yading@10:         tmp0 = temp[8*0 + i] + temp[8*1 + i];
yading@10:         tmp1 = temp[8*2 + i] + temp[8*3 + i];
yading@10:         tmp2 = temp[8*4 + i] + temp[8*5 + i];
yading@10:         tmp3 = temp[8*6 + i] + temp[8*7 + i];
yading@10:         tmp4 = temp[8*0 + i] - temp[8*1 + i];
yading@10:         tmp5 = temp[8*2 + i] - temp[8*3 + i];
yading@10:         tmp6 = temp[8*4 + i] - temp[8*5 + i];
yading@10:         tmp7 = temp[8*6 + i] - temp[8*7 + i];
yading@10: 
yading@10:         tmp10 = tmp0 + tmp3;
yading@10:         tmp11 = tmp1 + tmp2;
yading@10:         tmp12 = tmp1 - tmp2;
yading@10:         tmp13 = tmp0 - tmp3;
yading@10: 
yading@10:         data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
yading@10:         data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
yading@10: 
yading@10:         tmp12 += tmp13;
yading@10:         tmp12 *= A1;
yading@10:         data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
yading@10:         data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
yading@10: 
yading@10:         tmp10 = tmp4 + tmp7;
yading@10:         tmp11 = tmp5 + tmp6;
yading@10:         tmp12 = tmp5 - tmp6;
yading@10:         tmp13 = tmp4 - tmp7;
yading@10: 
yading@10:         data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
yading@10:         data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
yading@10: 
yading@10:         tmp12 += tmp13;
yading@10:         tmp12 *= A1;
yading@10:         data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
yading@10:         data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
yading@10:     }
yading@10: }