simple_idct_mmx.c

mediastreamer2是开源的网络传输媒体流的库

/*
 * Simple IDCT MMX
 *
 * Copyright (c) 2001, 2002 Michael Niedermayer <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg 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 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
#include "dsputil.h"
#include "simple_idct.h"

/*
23170.475006
22725.260826
21406.727617
19265.545870
16384.000000
12872.826198
8866.956905
4520.335430
*/
#define C0 23170 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define C1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define C2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define C3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#if 0
#define C4 16384 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#else
#define C4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) - 0.5
#endif
#define C5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define C6 8867  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define C7 4520  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5

#define ROW_SHIFT 11
#define COL_SHIFT 20 // 6

DECLARE_ASM_CONST(8, uint64_t, wm1010)= 0xFFFF0000FFFF0000ULL;
DECLARE_ASM_CONST(8, uint64_t, d40000)= 0x0000000000040000ULL;

DECLARE_ALIGNED(8, static const int16_t, coeffs[])= {
        1<<(ROW_SHIFT-1), 0, 1<<(ROW_SHIFT-1), 0,
//        1<<(COL_SHIFT-1), 0, 1<<(COL_SHIFT-1), 0,
//        0, 1<<(COL_SHIFT-1-16), 0, 1<<(COL_SHIFT-1-16),
        1<<(ROW_SHIFT-1), 1, 1<<(ROW_SHIFT-1), 0,
        // the 1 = ((1<<(COL_SHIFT-1))/C4)<<ROW_SHIFT :)
//        0, 0, 0, 0,
//        0, 0, 0, 0,

 C4,  C4,  C4,  C4,
 C4, -C4,  C4, -C4,

 C2,  C6,  C2,  C6,
 C6, -C2,  C6, -C2,

 C1,  C3,  C1,  C3,
 C5,  C7,  C5,  C7,

 C3, -C7,  C3, -C7,
-C1, -C5, -C1, -C5,

 C5, -C1,  C5, -C1,
 C7,  C3,  C7,  C3,

 C7, -C5,  C7, -C5,
 C3, -C1,  C3, -C1
};

#if 0
static void unused_var_killer(){
        int a= wm1010 + d40000;
        temp[0]=a;
}

static void inline idctCol (int16_t * col, int16_t *input)
{
#undef C0
#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
        int a0, a1, a2, a3, b0, b1, b2, b3;
        const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
/*
        if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
                col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
                        col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
                return;
        }*/

col[8*0] = input[8*0 + 0];
col[8*1] = input[8*2 + 0];
col[8*2] = input[8*0 + 1];
col[8*3] = input[8*2 + 1];
col[8*4] = input[8*4 + 0];
col[8*5] = input[8*6 + 0];
col[8*6] = input[8*4 + 1];
col[8*7] = input[8*6 + 1];

        a0 = C4*col[8*0] + C2*col[8*2] + C4*col[8*4] + C6*col[8*6] + (1<<(COL_SHIFT-1));
        a1 = C4*col[8*0] + C6*col[8*2] - C4*col[8*4] - C2*col[8*6] + (1<<(COL_SHIFT-1));
        a2 = C4*col[8*0] - C6*col[8*2] - C4*col[8*4] + C2*col[8*6] + (1<<(COL_SHIFT-1));
        a3 = C4*col[8*0] - C2*col[8*2] + C4*col[8*4] - C6*col[8*6] + (1<<(COL_SHIFT-1));

        b0 = C1*col[8*1] + C3*col[8*3] + C5*col[8*5] + C7*col[8*7];
        b1 = C3*col[8*1] - C7*col[8*3] - C1*col[8*5] - C5*col[8*7];
        b2 = C5*col[8*1] - C1*col[8*3] + C7*col[8*5] + C3*col[8*7];
        b3 = C7*col[8*1] - C5*col[8*3] + C3*col[8*5] - C1*col[8*7];

        col[8*0] = (a0 + b0) >> COL_SHIFT;
        col[8*1] = (a1 + b1) >> COL_SHIFT;
        col[8*2] = (a2 + b2) >> COL_SHIFT;
        col[8*3] = (a3 + b3) >> COL_SHIFT;
        col[8*4] = (a3 - b3) >> COL_SHIFT;
        col[8*5] = (a2 - b2) >> COL_SHIFT;
        col[8*6] = (a1 - b1) >> COL_SHIFT;
        col[8*7] = (a0 - b0) >> COL_SHIFT;
}

static void inline idctRow (int16_t * output, int16_t * input)
{
        int16_t row[8];

        int a0, a1, a2, a3, b0, b1, b2, b3;
        const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
        const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5

row[0] = input[0];
row[2] = input[1];
row[4] = input[4];
row[6] = input[5];
row[1] = input[8];
row[3] = input[9];
row[5] = input[12];
row[7] = input[13];

        if( !(row[1] | row[2] |row[3] |row[4] |row[5] |row[6] | row[7]) ) {
                row[0] = row[1] = row[2] = row[3] = row[4] =
                        row[5] = row[6] = row[7] = row[0]<<3;
        output[0]  = row[0];
        output[2]  = row[1];
        output[4]  = row[2];
        output[6]  = row[3];
        output[8]  = row[4];
        output[10] = row[5];
        output[12] = row[6];
        output[14] = row[7];
                return;
        }

        a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + (1<<(ROW_SHIFT-1));
        a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + (1<<(ROW_SHIFT-1));
        a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + (1<<(ROW_SHIFT-1));
        a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + (1<<(ROW_SHIFT-1));

        b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
        b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
        b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
        b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];

        row[0] = (a0 + b0) >> ROW_SHIFT;
        row[1] = (a1 + b1) >> ROW_SHIFT;
        row[2] = (a2 + b2) >> ROW_SHIFT;
        row[3] = (a3 + b3) >> ROW_SHIFT;
        row[4] = (a3 - b3) >> ROW_SHIFT;
        row[5] = (a2 - b2) >> ROW_SHIFT;
        row[6] = (a1 - b1) >> ROW_SHIFT;
        row[7] = (a0 - b0) >> ROW_SHIFT;

        output[0]  = row[0];
        output[2]  = row[1];
        output[4]  = row[2];
        output[6]  = row[3];
        output[8]  = row[4];
        output[10] = row[5];
        output[12] = row[6];
        output[14] = row[7];
}
#endif

static inline void idct(int16_t *block)
{
        DECLARE_ALIGNED(8, int64_t, align_tmp[16]);
        int16_t * const temp= (int16_t*)align_tmp;

        asm volatile(
#if 0 //Alternative, simpler variant

#define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
        "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
        "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
        "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
        "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
        "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
        "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
        "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
        "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
        "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
        "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
        "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\