📄 simple_idct_mmx.c
字号:
/*
* Simple IDCT MMX
*
* Copyright (c) 2001, 2002 Michael Niedermayer <michaelni@gmx.at>
*
* 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
*/
#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
static const uint64_t attribute_used __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000ULL;
static const uint64_t attribute_used __attribute__((aligned(8))) d40000= 0x0000000000040000ULL;
static const int16_t __attribute__((aligned(8))) 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)
{
int64_t __attribute__((aligned(8))) 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 */\
"pmaddwd %%mm6, %%mm1 \n\t" /* -C2R6+C6R2 -C2r6+C6r2 */\
"movq 48(%2), %%mm7 \n\t" /* C3 C1 C3 C1 */\
"pmaddwd %%mm2, %%mm7 \n\t" /* C3R3+C1R1 C3r3+C1r1 */\
#rounder ", %%mm4 \n\t"\
"movq %%mm4, %%mm6 \n\t" /* C4R4+C4R0 C4r4+C4r0 */\
"paddd %%mm5, %%mm4 \n\t" /* A0 a0 */\
"psubd %%mm5, %%mm6 \n\t" /* A3 a3 */\
"movq 56(%2), %%mm5 \n\t" /* C7 C5 C7 C5 */\
"pmaddwd %%mm3, %%mm5 \n\t" /* C7R7+C5R5 C7r7+C5r5 */\
#rounder ", %%mm0 \n\t"\
"paddd %%mm0, %%mm1 \n\t" /* A1 a1 */\
"paddd %%mm0, %%mm0 \n\t" \
"psubd %%mm1, %%mm0 \n\t" /* A2 a2 */\
"pmaddwd 64(%2), %%mm2 \n\t" /* -C7R3+C3R1 -C7r3+C3r1 */\
"paddd %%mm5, %%mm7 \n\t" /* B0 b0 */\
"movq 72(%2), %%mm5 \n\t" /* -C5 -C1 -C5 -C1 */\
"pmaddwd %%mm3, %%mm5 \n\t" /* -C5R7-C1R5 -C5r7-C1r5 */\
"paddd %%mm4, %%mm7 \n\t" /* A0+B0 a0+b0 */\
"paddd %%mm4, %%mm4 \n\t" /* 2A0 2a0 */\
"psubd %%mm7, %%mm4 \n\t" /* A0-B0 a0-b0 */\
"paddd %%mm2, %%mm5 \n\t" /* B1 b1 */\
"psrad $" #shift ", %%mm7 \n\t"\
"psrad $" #shift ", %%mm4 \n\t"\
"movq %%mm1, %%mm2 \n\t" /* A1 a1 */\
"paddd %%mm5, %%mm1 \n\t" /* A1+B1 a1+b1 */\
"psubd %%mm5, %%mm2 \n\t" /* A1-B1 a1-b1 */\
"psrad $" #shift ", %%mm1 \n\t"\
"psrad $" #shift ", %%mm2 \n\t"\
"packssdw %%mm1, %%mm7 \n\t" /* A1+B1 a1+b1 A0+B0 a0+b0 */\
"packssdw %%mm4, %%mm2 \n\t" /* A0-B0 a0-b0 A1-B1 a1-b1 */\
"movq %%mm7, " #dst " \n\t"\
"movq " #src1 ", %%mm1 \n\t" /* R3 R1 r3 r1 */\
"movq 80(%2), %%mm4 \n\t" /* -C1 C5 -C1 C5 */\
"movq %%mm2, 24+" #dst " \n\t"\
"pmaddwd %%mm1, %%mm4 \n\t" /* -C1R3+C5R1 -C1r3+C5r1 */\
"movq 88(%2), %%mm7 \n\t" /* C3 C7 C3 C7 */\
"pmaddwd 96(%2), %%mm1 \n\t" /* -C5R3+C7R1 -C5r3+C7r1 */\
"pmaddwd %%mm3, %%mm7 \n\t" /* C3R7+C7R5 C3r7+C7r5 */\
"movq %%mm0, %%mm2 \n\t" /* A2 a2 */\
"pmaddwd 104(%2), %%mm3 \n\t" /* -C1R7+C3R5 -C1r7+C3r5 */\
"paddd %%mm7, %%mm4 \n\t" /* B2 b2 */\
"paddd %%mm4, %%mm2 \n\t" /* A2+B2 a2+b2 */\
"psubd %%mm4, %%mm0 \n\t" /* a2-B2 a2-b2 */\
"psrad $" #shift ", %%mm2 \n\t"\
"psrad $" #shift ", %%mm0 \n\t"\
"movq %%mm6, %%mm4 \n\t" /* A3 a3 */\
"paddd %%mm1, %%mm3 \n\t" /* B3 b3 */\
"paddd %%mm3, %%mm6 \n\t" /* A3+B3 a3+b3 */\
"psubd %%mm3, %%mm4 \n\t" /* a3-B3 a3-b3 */\
"psrad $" #shift ", %%mm6 \n\t"\
"packssdw %%mm6, %%mm2 \n\t" /* A3+B3 a3+b3 A2+B2 a2+b2 */\
"movq %%mm2, 8+" #dst " \n\t"\
"psrad $" #shift ", %%mm4 \n\t"\
"packssdw %%mm0, %%mm4 \n\t" /* A2-B2 a2-b2 A3-B3 a3-b3 */\
"movq %%mm4, 16+" #dst " \n\t"\
#define COL_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 */\
"pmaddwd %%mm6, %%mm1 \n\t" /* -C2R6+C6R2 -C2r6+C6r2 */\
#rounder ", %%mm4 \n\t"\
"movq %%mm4, %%mm6 \n\t" /* C4R4+C4R0 C4r4+C4r0 */\
"movq 48(%2), %%mm7 \n\t" /* C3 C1 C3 C1 */\
#rounder ", %%mm0 \n\t"\
"pmaddwd %%mm2, %%mm7 \n\t" /* C3R3+C1R1 C3r3+C1r1 */\
"paddd %%mm5, %%mm4 \n\t" /* A0 a0 */\
"psubd %%mm5, %%mm6 \n\t" /* A3 a3 */\
"movq %%mm0, %%mm5 \n\t" /* -C4R4+C4R0 -C4r4+C4r0 */\
"paddd %%mm1, %%mm0 \n\t" /* A1 a1 */\
"psubd %%mm1, %%mm5 \n\t" /* A2 a2 */\
"movq 56(%2), %%mm1 \n\t" /* C7 C5 C7 C5 */\
"pmaddwd %%mm3, %%mm1 \n\t" /* C7R7+C5R5 C7r7+C5r5 */\
"pmaddwd 64(%2), %%mm2 \n\t" /* -C7R3+C3R1 -C7r3+C3r1 */\
"paddd %%mm1, %%mm7 \n\t" /* B0 b0 */\
"movq 72(%2), %%mm1 \n\t" /* -C5 -C1 -C5 -C1 */\
"pmaddwd %%mm3, %%mm1 \n\t" /* -C5R7-C1R5 -C5r7-C1r5 */\
"paddd %%mm4, %%mm7 \n\t" /* A0+B0 a0+b0 */\
"paddd %%mm4, %%mm4 \n\t" /* 2A0 2a0 */\
"psubd %%mm7, %%mm4 \n\t" /* A0-B0 a0-b0 */\
"paddd %%mm2, %%mm1 \n\t" /* B1 b1 */\
"psrad $" #shift ", %%mm7 \n\t"\
"psrad $" #shift ", %%mm4 \n\t"\
"movq %%mm0, %%mm2 \n\t" /* A1 a1 */\
"paddd %%mm1, %%mm0 \n\t" /* A1+B1 a1+b1 */\
"psubd %%mm1, %%mm2 \n\t" /* A1-B1 a1-b1 */\
"psrad $" #shift ", %%mm0 \n\t"\
"psrad $" #shift ", %%mm2 \n\t"\
"packssdw %%mm7, %%mm7 \n\t" /* A0+B0 a0+b0 */\
"movd %%mm7, " #dst " \n\t"\
"packssdw %%mm0, %%mm0 \n\t" /* A1+B1 a1+b1 */\
"movd %%mm0, 16+" #dst " \n\t"\
"packssdw %%mm2, %%mm2 \n\t" /* A1-B1 a1-b1 */\
"movd %%mm2, 96+" #dst " \n\t"\
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -