git.sesse.net Git - ffmpeg/blob - libavcodec/x86/simple_idct_mmx.c

   1 /*
   2  * Simple IDCT MMX
   3  *
   4  * Copyright (c) 2001, 2002 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22 #include "libavcodec/dsputil.h"
  23 #include "libavcodec/simple_idct.h"
  24
  25 /*
  26 23170.475006
  27 22725.260826
  28 21406.727617
  29 19265.545870
  30 16384.000000
  31 12872.826198
  32 8866.956905
  33 4520.335430
  34 */
  35 #define C0 23170 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  36 #define C1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  37 #define C2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  38 #define C3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  39 #if 0
  40 #define C4 16384 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  41 #else
  42 #define C4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) - 0.5
  43 #endif
  44 #define C5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  45 #define C6 8867  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  46 #define C7 4520  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  47
  48 #define ROW_SHIFT 11
  49 #define COL_SHIFT 20 // 6
  50
  51 DECLARE_ASM_CONST(8, uint64_t, wm1010)= 0xFFFF0000FFFF0000ULL;
  52 DECLARE_ASM_CONST(8, uint64_t, d40000)= 0x0000000000040000ULL;
  53
  54 DECLARE_ALIGNED(8, static const int16_t, coeffs[])= {
  55         1<<(ROW_SHIFT-1), 0, 1<<(ROW_SHIFT-1), 0,
  56 //        1<<(COL_SHIFT-1), 0, 1<<(COL_SHIFT-1), 0,
  57 //        0, 1<<(COL_SHIFT-1-16), 0, 1<<(COL_SHIFT-1-16),
  58         1<<(ROW_SHIFT-1), 1, 1<<(ROW_SHIFT-1), 0,
  59         // the 1 = ((1<<(COL_SHIFT-1))/C4)<<ROW_SHIFT :)
  60 //        0, 0, 0, 0,
  61 //        0, 0, 0, 0,
  62
  63  C4,  C4,  C4,  C4,
  64  C4, -C4,  C4, -C4,
  65
  66  C2,  C6,  C2,  C6,
  67  C6, -C2,  C6, -C2,
  68
  69  C1,  C3,  C1,  C3,
  70  C5,  C7,  C5,  C7,
  71
  72  C3, -C7,  C3, -C7,
  73 -C1, -C5, -C1, -C5,
  74
  75  C5, -C1,  C5, -C1,
  76  C7,  C3,  C7,  C3,
  77
  78  C7, -C5,  C7, -C5,
  79  C3, -C1,  C3, -C1
  80 };
  81
  82 #if 0
  83 static void unused_var_killer(void)
  84 {
  85         int a= wm1010 + d40000;
  86         temp[0]=a;
  87 }
  88
  89 static void inline idctCol (int16_t * col, int16_t *input)
  90 {
  91 #undef C0
  92 #undef C1
  93 #undef C2
  94 #undef C3
  95 #undef C4
  96 #undef C5
  97 #undef C6
  98 #undef C7
  99         int a0, a1, a2, a3, b0, b1, b2, b3;
 100         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 101         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 102         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 103         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 104         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 105         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 106         const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 107         const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 108 /*
 109         if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
 110                 col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
 111                         col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
 112                 return;
 113         }*/
 114
 115 col[8*0] = input[8*0 + 0];
 116 col[8*1] = input[8*2 + 0];
 117 col[8*2] = input[8*0 + 1];
 118 col[8*3] = input[8*2 + 1];
 119 col[8*4] = input[8*4 + 0];
 120 col[8*5] = input[8*6 + 0];
 121 col[8*6] = input[8*4 + 1];
 122 col[8*7] = input[8*6 + 1];
 123
 124         a0 = C4*col[8*0] + C2*col[8*2] + C4*col[8*4] + C6*col[8*6] + (1<<(COL_SHIFT-1));
 125         a1 = C4*col[8*0] + C6*col[8*2] - C4*col[8*4] - C2*col[8*6] + (1<<(COL_SHIFT-1));
 126         a2 = C4*col[8*0] - C6*col[8*2] - C4*col[8*4] + C2*col[8*6] + (1<<(COL_SHIFT-1));
 127         a3 = C4*col[8*0] - C2*col[8*2] + C4*col[8*4] - C6*col[8*6] + (1<<(COL_SHIFT-1));
 128
 129         b0 = C1*col[8*1] + C3*col[8*3] + C5*col[8*5] + C7*col[8*7];
 130         b1 = C3*col[8*1] - C7*col[8*3] - C1*col[8*5] - C5*col[8*7];
 131         b2 = C5*col[8*1] - C1*col[8*3] + C7*col[8*5] + C3*col[8*7];
 132         b3 = C7*col[8*1] - C5*col[8*3] + C3*col[8*5] - C1*col[8*7];
 133
 134         col[8*0] = (a0 + b0) >> COL_SHIFT;
 135         col[8*1] = (a1 + b1) >> COL_SHIFT;
 136         col[8*2] = (a2 + b2) >> COL_SHIFT;
 137         col[8*3] = (a3 + b3) >> COL_SHIFT;
 138         col[8*4] = (a3 - b3) >> COL_SHIFT;
 139         col[8*5] = (a2 - b2) >> COL_SHIFT;
 140         col[8*6] = (a1 - b1) >> COL_SHIFT;
 141         col[8*7] = (a0 - b0) >> COL_SHIFT;
 142 }
 143
 144 static void inline idctRow (int16_t * output, int16_t * input)
 145 {
 146         int16_t row[8];
 147
 148         int a0, a1, a2, a3, b0, b1, b2, b3;
 149         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 150         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 151         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 152         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 153         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 154         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 155         const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 156         const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 157
 158 row[0] = input[0];
 159 row[2] = input[1];
 160 row[4] = input[4];
 161 row[6] = input[5];
 162 row[1] = input[8];
 163 row[3] = input[9];
 164 row[5] = input[12];
 165 row[7] = input[13];
 166
 167         if( !(row[1] | row[2] |row[3] |row[4] |row[5] |row[6] | row[7]) ) {
 168                 row[0] = row[1] = row[2] = row[3] = row[4] =
 169                         row[5] = row[6] = row[7] = row[0]<<3;
 170         output[0]  = row[0];
 171         output[2]  = row[1];
 172         output[4]  = row[2];
 173         output[6]  = row[3];
 174         output[8]  = row[4];
 175         output[10] = row[5];
 176         output[12] = row[6];
 177         output[14] = row[7];
 178                 return;
 179         }
 180
 181         a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + (1<<(ROW_SHIFT-1));
 182         a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + (1<<(ROW_SHIFT-1));
 183         a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + (1<<(ROW_SHIFT-1));
 184         a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + (1<<(ROW_SHIFT-1));
 185
 186         b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
 187         b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
 188         b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
 189         b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];
 190
 191         row[0] = (a0 + b0) >> ROW_SHIFT;
 192         row[1] = (a1 + b1) >> ROW_SHIFT;
 193         row[2] = (a2 + b2) >> ROW_SHIFT;
 194         row[3] = (a3 + b3) >> ROW_SHIFT;
 195         row[4] = (a3 - b3) >> ROW_SHIFT;
 196         row[5] = (a2 - b2) >> ROW_SHIFT;
 197         row[6] = (a1 - b1) >> ROW_SHIFT;
 198         row[7] = (a0 - b0) >> ROW_SHIFT;
 199
 200         output[0]  = row[0];
 201         output[2]  = row[1];
 202         output[4]  = row[2];
 203         output[6]  = row[3];
 204         output[8]  = row[4];
 205         output[10] = row[5];
 206         output[12] = row[6];
 207         output[14] = row[7];
 208 }
 209 #endif
 210
 211 static inline void idct(int16_t *block)
 212 {
 213         DECLARE_ALIGNED(8, int64_t, align_tmp[16]);
 214         int16_t * const temp= (int16_t*)align_tmp;
 215
 216         __asm__ volatile(
 217 #if 0 //Alternative, simpler variant
 218
 219 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 220         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 221         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 222         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 223         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 224         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 225         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 226         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 227         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 228         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 229         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 230         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 231         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 232         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 233         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 234         #rounder ", %%mm4               \n\t"\
 235         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 236         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 237         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 238         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 239         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 240         #rounder ", %%mm0               \n\t"\
 241         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 242         "paddd %%mm0, %%mm0             \n\t" \
 243         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 244         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 245         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 246         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 247         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 248         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 249         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 250         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 251         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 252         "psrad $" #shift ", %%mm7       \n\t"\
 253         "psrad $" #shift ", %%mm4       \n\t"\
 254         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 255         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 256         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 257         "psrad $" #shift ", %%mm1       \n\t"\
 258         "psrad $" #shift ", %%mm2       \n\t"\
 259         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 260         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 261         "movq %%mm7, " #dst "           \n\t"\
 262         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 263         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 264         "movq %%mm2, 24+" #dst "        \n\t"\
 265         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 266         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 267         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 268         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 269         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 270         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 271         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 272         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 273         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 274         "psrad $" #shift ", %%mm2       \n\t"\
 275         "psrad $" #shift ", %%mm0       \n\t"\
 276         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 277         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 278         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 279         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 280         "psrad $" #shift ", %%mm6       \n\t"\
 281         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 282         "movq %%mm2, 8+" #dst "         \n\t"\
 283         "psrad $" #shift ", %%mm4       \n\t"\
 284         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 285         "movq %%mm4, 16+" #dst "        \n\t"\
 286
 287 #define COL_IDCT(src0, src4, src1, src5, dst, shift) \
 288         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 289         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 290         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 291         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 292         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 293         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 294         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 295         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 296         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 297         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 298         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 299         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 300         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 301         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 302         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 303         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 304         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 305         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 306         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 307         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 308         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 309         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 310         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 311         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 312         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 313         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 314         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 315         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 316         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 317         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 318         "psrad $" #shift ", %%mm7       \n\t"\
 319         "psrad $" #shift ", %%mm4       \n\t"\
 320         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 321         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 322         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 323         "psrad $" #shift ", %%mm0       \n\t"\
 324         "psrad $" #shift ", %%mm2       \n\t"\
 325         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 326         "movd %%mm7, " #dst "           \n\t"\
 327         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 328         "movd %%mm0, 16+" #dst "        \n\t"\
 329         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 330         "movd %%mm2, 96+" #dst "        \n\t"\
 331         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 332         "movd %%mm4, 112+" #dst "       \n\t"\
 333         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 334         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 335         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 336         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 337         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 338         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 339         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 340         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 341         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 342         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 343         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 344         "psrad $" #shift ", %%mm2       \n\t"\
 345         "psrad $" #shift ", %%mm5       \n\t"\
 346         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 347         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 348         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 349         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 350         "psrad $" #shift ", %%mm6       \n\t"\
 351         "psrad $" #shift ", %%mm4       \n\t"\
 352         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 353         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 354         "movd %%mm2, 32+" #dst "        \n\t"\
 355         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
 356         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 357         "movd %%mm6, 48+" #dst "        \n\t"\
 358         "movd %%mm4, 64+" #dst "        \n\t"\
 359         "movd %%mm5, 80+" #dst "        \n\t"\
 360
 361
 362 #define DC_COND_ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 363         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 364         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 365         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 366         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 367         "movq "MANGLE(wm1010)", %%mm4   \n\t"\
 368         "pand %%mm0, %%mm4              \n\t"\
 369         "por %%mm1, %%mm4               \n\t"\
 370         "por %%mm2, %%mm4               \n\t"\
 371         "por %%mm3, %%mm4               \n\t"\
 372         "packssdw %%mm4,%%mm4           \n\t"\
 373         "movd %%mm4, %%eax              \n\t"\
 374         "orl %%eax, %%eax               \n\t"\
 375         "jz 1f                          \n\t"\
 376         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 377         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 378         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 379         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 380         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 381         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 382         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 383         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 384         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 385         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 386         #rounder ", %%mm4               \n\t"\
 387         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 388         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 389         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 390         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 391         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 392         #rounder ", %%mm0               \n\t"\
 393         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 394         "paddd %%mm0, %%mm0             \n\t" \
 395         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 396         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 397         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 398         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 399         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 400         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 401         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 402         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 403         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 404         "psrad $" #shift ", %%mm7       \n\t"\
 405         "psrad $" #shift ", %%mm4       \n\t"\
 406         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 407         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 408         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 409         "psrad $" #shift ", %%mm1       \n\t"\
 410         "psrad $" #shift ", %%mm2       \n\t"\
 411         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 412         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 413         "movq %%mm7, " #dst "           \n\t"\
 414         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 415         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 416         "movq %%mm2, 24+" #dst "        \n\t"\
 417         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 418         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 419         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 420         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 421         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 422         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 423         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 424         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 425         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 426         "psrad $" #shift ", %%mm2       \n\t"\
 427         "psrad $" #shift ", %%mm0       \n\t"\
 428         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 429         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 430         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 431         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 432         "psrad $" #shift ", %%mm6       \n\t"\
 433         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 434         "movq %%mm2, 8+" #dst "         \n\t"\
 435         "psrad $" #shift ", %%mm4       \n\t"\
 436         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 437         "movq %%mm4, 16+" #dst "        \n\t"\
 438         "jmp 2f                         \n\t"\
 439         "1:                             \n\t"\
 440         "pslld $16, %%mm0               \n\t"\
 441         "#paddd "MANGLE(d40000)", %%mm0 \n\t"\
 442         "psrad $13, %%mm0               \n\t"\
 443         "packssdw %%mm0, %%mm0          \n\t"\
 444         "movq %%mm0, " #dst "           \n\t"\
 445         "movq %%mm0, 8+" #dst "         \n\t"\
 446         "movq %%mm0, 16+" #dst "        \n\t"\
 447         "movq %%mm0, 24+" #dst "        \n\t"\
 448         "2:                             \n\t"
 449
 450
 451 //IDCT(      src0,   src4,   src1,   src5,    dst,    rounder, shift)
 452 ROW_IDCT(    (%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 453 /*ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1), paddd (%2), 11)
 454 ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1), paddd (%2), 11)
 455 ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1), paddd (%2), 11)*/
 456
 457 DC_COND_ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11)
 458 DC_COND_ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11)
 459 DC_COND_ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11)
 460
 461
 462 //IDCT(      src0,   src4,   src1,    src5,    dst, shift)
 463 COL_IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
 464 COL_IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
 465 COL_IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
 466 COL_IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
 467
 468 #else
 469
 470 #define DC_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 471         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 472         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 473         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 474         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 475         "movq "MANGLE(wm1010)", %%mm4   \n\t"\
 476         "pand %%mm0, %%mm4              \n\t"\
 477         "por %%mm1, %%mm4               \n\t"\
 478         "por %%mm2, %%mm4               \n\t"\
 479         "por %%mm3, %%mm4               \n\t"\
 480         "packssdw %%mm4,%%mm4           \n\t"\
 481         "movd %%mm4, %%eax              \n\t"\
 482         "orl %%eax, %%eax               \n\t"\
 483         "jz 1f                          \n\t"\
 484         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 485         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 486         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 487         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 488         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 489         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 490         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 491         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 492         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 493         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 494         #rounder ", %%mm4               \n\t"\
 495         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 496         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 497         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 498         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 499         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 500         #rounder ", %%mm0               \n\t"\
 501         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 502         "paddd %%mm0, %%mm0             \n\t" \
 503         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 504         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 505         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 506         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 507         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 508         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 509         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 510         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 511         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 512         "psrad $" #shift ", %%mm7       \n\t"\
 513         "psrad $" #shift ", %%mm4       \n\t"\
 514         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 515         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 516         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 517         "psrad $" #shift ", %%mm1       \n\t"\
 518         "psrad $" #shift ", %%mm2       \n\t"\
 519         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 520         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 521         "movq %%mm7, " #dst "           \n\t"\
 522         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 523         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 524         "movq %%mm2, 24+" #dst "        \n\t"\
 525         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 526         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 527         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 528         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 529         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 530         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 531         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 532         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 533         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 534         "psrad $" #shift ", %%mm2       \n\t"\
 535         "psrad $" #shift ", %%mm0       \n\t"\
 536         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 537         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 538         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 539         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 540         "psrad $" #shift ", %%mm6       \n\t"\
 541         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 542         "movq %%mm2, 8+" #dst "         \n\t"\
 543         "psrad $" #shift ", %%mm4       \n\t"\
 544         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 545         "movq %%mm4, 16+" #dst "        \n\t"\
 546         "jmp 2f                         \n\t"\
 547         "1:                             \n\t"\
 548         "pslld $16, %%mm0               \n\t"\
 549         "paddd "MANGLE(d40000)", %%mm0  \n\t"\
 550         "psrad $13, %%mm0               \n\t"\
 551         "packssdw %%mm0, %%mm0          \n\t"\
 552         "movq %%mm0, " #dst "           \n\t"\
 553         "movq %%mm0, 8+" #dst "         \n\t"\
 554         "movq %%mm0, 16+" #dst "        \n\t"\
 555         "movq %%mm0, 24+" #dst "        \n\t"\
 556         "2:                             \n\t"
 557
 558 #define Z_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift, bt) \
 559         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 560         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 561         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 562         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 563         "movq %%mm0, %%mm4              \n\t"\
 564         "por %%mm1, %%mm4               \n\t"\
 565         "por %%mm2, %%mm4               \n\t"\
 566         "por %%mm3, %%mm4               \n\t"\
 567         "packssdw %%mm4,%%mm4           \n\t"\
 568         "movd %%mm4, %%eax              \n\t"\
 569         "orl %%eax, %%eax               \n\t"\
 570         "jz " #bt "                     \n\t"\
 571         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 572         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 573         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 574         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 575         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 576         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 577         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 578         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 579         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 580         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 581         #rounder ", %%mm4               \n\t"\
 582         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 583         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 584         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 585         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 586         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 587         #rounder ", %%mm0               \n\t"\
 588         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 589         "paddd %%mm0, %%mm0             \n\t" \
 590         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 591         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 592         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 593         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 594         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 595         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 596         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 597         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 598         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 599         "psrad $" #shift ", %%mm7       \n\t"\
 600         "psrad $" #shift ", %%mm4       \n\t"\
 601         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 602         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 603         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 604         "psrad $" #shift ", %%mm1       \n\t"\
 605         "psrad $" #shift ", %%mm2       \n\t"\
 606         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 607         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 608         "movq %%mm7, " #dst "           \n\t"\
 609         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 610         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 611         "movq %%mm2, 24+" #dst "        \n\t"\
 612         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 613         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 614         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 615         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 616         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 617         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 618         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 619         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 620         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 621         "psrad $" #shift ", %%mm2       \n\t"\
 622         "psrad $" #shift ", %%mm0       \n\t"\
 623         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 624         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 625         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 626         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 627         "psrad $" #shift ", %%mm6       \n\t"\
 628         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 629         "movq %%mm2, 8+" #dst "         \n\t"\
 630         "psrad $" #shift ", %%mm4       \n\t"\
 631         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 632         "movq %%mm4, 16+" #dst "        \n\t"\
 633
 634 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 635         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 636         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 637         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 638         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 639         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 640         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 641         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 642         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 643         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 644         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 645         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 646         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 647         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 648         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 649         #rounder ", %%mm4               \n\t"\
 650         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 651         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 652         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 653         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 654         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 655         #rounder ", %%mm0               \n\t"\
 656         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 657         "paddd %%mm0, %%mm0             \n\t" \
 658         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 659         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 660         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 661         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 662         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 663         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 664         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 665         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 666         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 667         "psrad $" #shift ", %%mm7       \n\t"\
 668         "psrad $" #shift ", %%mm4       \n\t"\
 669         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 670         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 671         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 672         "psrad $" #shift ", %%mm1       \n\t"\
 673         "psrad $" #shift ", %%mm2       \n\t"\
 674         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 675         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 676         "movq %%mm7, " #dst "           \n\t"\
 677         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 678         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 679         "movq %%mm2, 24+" #dst "        \n\t"\
 680         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 681         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 682         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 683         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 684         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 685         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 686         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 687         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 688         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 689         "psrad $" #shift ", %%mm2       \n\t"\
 690         "psrad $" #shift ", %%mm0       \n\t"\
 691         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 692         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 693         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 694         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 695         "psrad $" #shift ", %%mm6       \n\t"\
 696         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 697         "movq %%mm2, 8+" #dst "         \n\t"\
 698         "psrad $" #shift ", %%mm4       \n\t"\
 699         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 700         "movq %%mm4, 16+" #dst "        \n\t"\
 701
 702 //IDCT(         src0,   src4,   src1,   src5,    dst,   rounder, shift)
 703 DC_COND_IDCT(  0(%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 704 Z_COND_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11, 4f)
 705 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 2f)
 706 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 1f)
 707
 708 #undef IDCT
 709 #define IDCT(src0, src4, src1, src5, dst, shift) \
 710         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 711         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 712         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 713         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 714         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 715         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 716         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 717         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 718         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 719         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 720         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 721         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 722         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 723         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 724         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 725         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 726         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 727         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 728         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 729         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 730         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 731         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 732         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 733         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 734         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 735         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 736         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 737         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 738         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 739         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 740         "psrad $" #shift ", %%mm7       \n\t"\
 741         "psrad $" #shift ", %%mm4       \n\t"\
 742         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 743         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 744         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 745         "psrad $" #shift ", %%mm0       \n\t"\
 746         "psrad $" #shift ", %%mm2       \n\t"\
 747         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 748         "movd %%mm7, " #dst "           \n\t"\
 749         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 750         "movd %%mm0, 16+" #dst "        \n\t"\
 751         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 752         "movd %%mm2, 96+" #dst "        \n\t"\
 753         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 754         "movd %%mm4, 112+" #dst "       \n\t"\
 755         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 756         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 757         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 758         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 759         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 760         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 761         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 762         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 763         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 764         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 765         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 766         "psrad $" #shift ", %%mm2       \n\t"\
 767         "psrad $" #shift ", %%mm5       \n\t"\
 768         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 769         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 770         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 771         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 772         "psrad $" #shift ", %%mm6       \n\t"\
 773         "psrad $" #shift ", %%mm4       \n\t"\
 774         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 775         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 776         "movd %%mm2, 32+" #dst "        \n\t"\
 777         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
 778         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 779         "movd %%mm6, 48+" #dst "        \n\t"\
 780         "movd %%mm4, 64+" #dst "        \n\t"\
 781         "movd %%mm5, 80+" #dst "        \n\t"
 782
 783
 784 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
 785 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
 786 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
 787 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
 788 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
 789         "jmp 9f                         \n\t"
 790
 791         "#" ASMALIGN(4)                      \
 792         "4:                             \n\t"
 793 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 6f)
 794 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 5f)
 795
 796 #undef IDCT
 797 #define IDCT(src0, src4, src1, src5, dst, shift) \
 798         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 799         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 800         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 801         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 802         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 803         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 804         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 805         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 806         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 807         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 808         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 809         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 810         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 811         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 812         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 813         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 814         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 815         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 816         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 817         "movq 72(%2), %%mm7             \n\t" /* -C5    -C1     -C5     -C1 */\
 818         "pmaddwd %%mm3, %%mm7           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 819         "paddd %%mm4, %%mm1             \n\t" /* A0+B0          a0+b0 */\
 820         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 821         "psubd %%mm1, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 822         "psrad $" #shift ", %%mm1       \n\t"\
 823         "psrad $" #shift ", %%mm4       \n\t"\
 824         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 825         "paddd %%mm7, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 826         "psubd %%mm7, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 827         "psrad $" #shift ", %%mm0       \n\t"\
 828         "psrad $" #shift ", %%mm2       \n\t"\
 829         "packssdw %%mm1, %%mm1          \n\t" /* A0+B0  a0+b0 */\
 830         "movd %%mm1, " #dst "           \n\t"\
 831         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 832         "movd %%mm0, 16+" #dst "        \n\t"\
 833         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 834         "movd %%mm2, 96+" #dst "        \n\t"\
 835         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 836         "movd %%mm4, 112+" #dst "       \n\t"\
 837         "movq 88(%2), %%mm1             \n\t" /* C3     C7      C3      C7 */\
 838         "pmaddwd %%mm3, %%mm1           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 839         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 840         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 841         "paddd %%mm1, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 842         "psubd %%mm1, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 843         "psrad $" #shift ", %%mm2       \n\t"\
 844         "psrad $" #shift ", %%mm5       \n\t"\
 845         "movq %%mm6, %%mm1              \n\t" /* A3             a3 */\
 846         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 847         "psubd %%mm3, %%mm1             \n\t" /* a3-B3          a3-b3 */\
 848         "psrad $" #shift ", %%mm6       \n\t"\
 849         "psrad $" #shift ", %%mm1       \n\t"\
 850         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 851         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 852         "movd %%mm2, 32+" #dst "        \n\t"\
 853         "packssdw %%mm1, %%mm1          \n\t" /* A3-B3  a3-b3 */\
 854         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 855         "movd %%mm6, 48+" #dst "        \n\t"\
 856         "movd %%mm1, 64+" #dst "        \n\t"\
 857         "movd %%mm5, 80+" #dst "        \n\t"
 858
 859 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
 860 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
 861 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
 862 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
 863 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
 864         "jmp 9f                         \n\t"
 865
 866         "#" ASMALIGN(4)                      \
 867         "6:                             \n\t"
 868 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 7f)
 869
 870 #undef IDCT
 871 #define IDCT(src0, src4, src1, src5, dst, shift) \
 872         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 873         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 874         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 875         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 876         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 877         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 878         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 879         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 880         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 881         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 882         "movq 72(%2), %%mm7             \n\t" /* -C5    -C1     -C5     -C1 */\
 883         "pmaddwd %%mm3, %%mm7           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 884         "paddd %%mm4, %%mm1             \n\t" /* A0+B0          a0+b0 */\
 885         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 886         "psubd %%mm1, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 887         "psrad $" #shift ", %%mm1       \n\t"\
 888         "psrad $" #shift ", %%mm4       \n\t"\
 889         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 890         "paddd %%mm7, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 891         "psubd %%mm7, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 892         "psrad $" #shift ", %%mm0       \n\t"\
 893         "psrad $" #shift ", %%mm2       \n\t"\
 894         "packssdw %%mm1, %%mm1          \n\t" /* A0+B0  a0+b0 */\
 895         "movd %%mm1, " #dst "           \n\t"\
 896         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 897         "movd %%mm0, 16+" #dst "        \n\t"\
 898         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 899         "movd %%mm2, 96+" #dst "        \n\t"\
 900         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 901         "movd %%mm4, 112+" #dst "       \n\t"\
 902         "movq 88(%2), %%mm1             \n\t" /* C3     C7      C3      C7 */\
 903         "pmaddwd %%mm3, %%mm1           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 904         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 905         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 906         "paddd %%mm1, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 907         "psubd %%mm1, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 908         "psrad $" #shift ", %%mm2       \n\t"\
 909         "psrad $" #shift ", %%mm5       \n\t"\
 910         "movq %%mm6, %%mm1              \n\t" /* A3             a3 */\
 911         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 912         "psubd %%mm3, %%mm1             \n\t" /* a3-B3          a3-b3 */\
 913         "psrad $" #shift ", %%mm6       \n\t"\
 914         "psrad $" #shift ", %%mm1       \n\t"\
 915         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 916         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 917         "movd %%mm2, 32+" #dst "        \n\t"\
 918         "packssdw %%mm1, %%mm1          \n\t" /* A3-B3  a3-b3 */\
 919         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 920         "movd %%mm6, 48+" #dst "        \n\t"\
 921         "movd %%mm1, 64+" #dst "        \n\t"\
 922         "movd %%mm5, 80+" #dst "        \n\t"
 923
 924
 925 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
 926 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
 927 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
 928 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
 929 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
 930         "jmp 9f                         \n\t"
 931
 932         "#" ASMALIGN(4)                      \
 933         "2:                             \n\t"
 934 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 3f)
 935
 936 #undef IDCT
 937 #define IDCT(src0, src4, src1, src5, dst, shift) \
 938         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 939         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 940         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 941         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 942         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 943         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 944         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 945         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 946         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 947         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 948         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 949         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 950         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 951         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 952         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 953         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 954         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 955         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 956         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 957         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 958         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 959         "psrad $" #shift ", %%mm7       \n\t"\
 960         "psrad $" #shift ", %%mm4       \n\t"\
 961         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 962         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 963         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 964         "psrad $" #shift ", %%mm0       \n\t"\
 965         "psrad $" #shift ", %%mm2       \n\t"\
 966         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 967         "movd %%mm7, " #dst "           \n\t"\
 968         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 969         "movd %%mm0, 16+" #dst "        \n\t"\
 970         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 971         "movd %%mm2, 96+" #dst "        \n\t"\
 972         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 973         "movd %%mm4, 112+" #dst "       \n\t"\
 974         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 975         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 976         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 977         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 978         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 979         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 980         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 981         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 982         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 983         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 984         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 985         "psrad $" #shift ", %%mm2       \n\t"\
 986         "psrad $" #shift ", %%mm5       \n\t"\
 987         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 988         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 989         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 990         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 991         "psrad $" #shift ", %%mm6       \n\t"\
 992         "psrad $" #shift ", %%mm4       \n\t"\
 993         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 994         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 995         "movd %%mm2, 32+" #dst "        \n\t"\
 996         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
 997         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 998         "movd %%mm6, 48+" #dst "        \n\t"\
 999         "movd %%mm4, 64+" #dst "        \n\t"\
1000         "movd %%mm5, 80+" #dst "        \n\t"
1001
1002 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
1003 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
1004 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
1005 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
1006 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
1007         "jmp 9f                         \n\t"
1008
1009         "#" ASMALIGN(4)                      \
1010         "3:                             \n\t"
1011 #undef IDCT
1012 #define IDCT(src0, src4, src1, src5, dst, shift) \
1013         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1014         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
1015         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1016         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1017         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1018         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1019         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1020         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
1021         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1022         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1023         "movq 64(%2), %%mm3             \n\t"\
1024         "pmaddwd %%mm2, %%mm3           \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1025         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
1026         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
1027         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
1028         "psrad $" #shift ", %%mm7       \n\t"\
1029         "psrad $" #shift ", %%mm4       \n\t"\
1030         "movq %%mm0, %%mm1              \n\t" /* A1             a1 */\
1031         "paddd %%mm3, %%mm0             \n\t" /* A1+B1          a1+b1 */\
1032         "psubd %%mm3, %%mm1             \n\t" /* A1-B1          a1-b1 */\
1033         "psrad $" #shift ", %%mm0       \n\t"\
1034         "psrad $" #shift ", %%mm1       \n\t"\
1035         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
1036         "movd %%mm7, " #dst "           \n\t"\
1037         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
1038         "movd %%mm0, 16+" #dst "        \n\t"\
1039         "packssdw %%mm1, %%mm1          \n\t" /* A1-B1  a1-b1 */\
1040         "movd %%mm1, 96+" #dst "        \n\t"\
1041         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
1042         "movd %%mm4, 112+" #dst "       \n\t"\
1043         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
1044         "pmaddwd %%mm2, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1045         "pmaddwd 96(%2), %%mm2          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1046         "movq %%mm5, %%mm1              \n\t" /* A2             a2 */\
1047         "paddd %%mm4, %%mm1             \n\t" /* A2+B2          a2+b2 */\
1048         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
1049         "psrad $" #shift ", %%mm1       \n\t"\
1050         "psrad $" #shift ", %%mm5       \n\t"\
1051         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
1052         "paddd %%mm2, %%mm6             \n\t" /* A3+B3          a3+b3 */\
1053         "psubd %%mm2, %%mm4             \n\t" /* a3-B3          a3-b3 */\
1054         "psrad $" #shift ", %%mm6       \n\t"\
1055         "psrad $" #shift ", %%mm4       \n\t"\
1056         "packssdw %%mm1, %%mm1          \n\t" /* A2+B2  a2+b2 */\
1057         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1058         "movd %%mm1, 32+" #dst "        \n\t"\
1059         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
1060         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1061         "movd %%mm6, 48+" #dst "        \n\t"\
1062         "movd %%mm4, 64+" #dst "        \n\t"\
1063         "movd %%mm5, 80+" #dst "        \n\t"
1064
1065
1066 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
1067 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
1068 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
1069 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
1070 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
1071         "jmp 9f                         \n\t"
1072
1073         "#" ASMALIGN(4)                      \
1074         "5:                             \n\t"
1075 #undef IDCT
1076 #define IDCT(src0, src4, src1, src5, dst, shift) \
1077         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1078         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
1079         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1080         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1081         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1082         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1083         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
1084         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1085         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
1086         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1087         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1088         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
1089         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
1090         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1091         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
1092         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
1093         "movq 8+" #src0 ", %%mm2        \n\t" /* R4     R0      r4      r0 */\
1094         "movq 8+" #src4 ", %%mm3        \n\t" /* R6     R2      r6      r2 */\
1095         "movq 16(%2), %%mm1             \n\t" /* C4     C4      C4      C4 */\
1096         "pmaddwd %%mm2, %%mm1           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1097         "movq 24(%2), %%mm7             \n\t" /* -C4    C4      -C4     C4 */\
1098         "pmaddwd %%mm7, %%mm2           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1099         "movq 32(%2), %%mm7             \n\t" /* C6     C2      C6      C2 */\
1100         "pmaddwd %%mm3, %%mm7           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1101         "pmaddwd 40(%2), %%mm3          \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1102         "paddd %%mm1, %%mm7             \n\t" /* A0             a0 */\
1103         "paddd %%mm1, %%mm1             \n\t" /* 2C0            2c0 */\
1104         "psubd %%mm7, %%mm1             \n\t" /* A3             a3 */\
1105         "paddd %%mm2, %%mm3             \n\t" /* A1             a1 */\
1106         "paddd %%mm2, %%mm2             \n\t" /* 2C1            2c1 */\
1107         "psubd %%mm3, %%mm2             \n\t" /* A2             a2 */\
1108         "psrad $" #shift ", %%mm4       \n\t"\
1109         "psrad $" #shift ", %%mm7       \n\t"\
1110         "psrad $" #shift ", %%mm3       \n\t"\
1111         "packssdw %%mm7, %%mm4          \n\t" /* A0     a0 */\
1112         "movq %%mm4, " #dst "           \n\t"\
1113         "psrad $" #shift ", %%mm0       \n\t"\
1114         "packssdw %%mm3, %%mm0          \n\t" /* A1     a1 */\
1115         "movq %%mm0, 16+" #dst "        \n\t"\
1116         "movq %%mm0, 96+" #dst "        \n\t"\
1117         "movq %%mm4, 112+" #dst "       \n\t"\
1118         "psrad $" #shift ", %%mm5       \n\t"\
1119         "psrad $" #shift ", %%mm6       \n\t"\
1120         "psrad $" #shift ", %%mm2       \n\t"\
1121         "packssdw %%mm2, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1122         "movq %%mm5, 32+" #dst "        \n\t"\
1123         "psrad $" #shift ", %%mm1       \n\t"\
1124         "packssdw %%mm1, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1125         "movq %%mm6, 48+" #dst "        \n\t"\
1126         "movq %%mm6, 64+" #dst "        \n\t"\
1127         "movq %%mm5, 80+" #dst "        \n\t"
1128
1129
1130 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
1131 IDCT(    0(%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
1132 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
1133 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
1134 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
1135         "jmp 9f                         \n\t"
1136
1137
1138         "#" ASMALIGN(4)                      \
1139         "1:                             \n\t"
1140 #undef IDCT
1141 #define IDCT(src0, src4, src1, src5, dst, shift) \
1142         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1143         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
1144         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
1145         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1146         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1147         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1148         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1149         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
1150         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1151         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
1152         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1153         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1154         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
1155         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1156         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
1157         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
1158         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1159         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
1160         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
1161         "movq 64(%2), %%mm1             \n\t"\
1162         "pmaddwd %%mm2, %%mm1           \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1163         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
1164         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
1165         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
1166         "psrad $" #shift ", %%mm7       \n\t"\
1167         "psrad $" #shift ", %%mm4       \n\t"\
1168         "movq %%mm0, %%mm3              \n\t" /* A1             a1 */\
1169         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
1170         "psubd %%mm1, %%mm3             \n\t" /* A1-B1          a1-b1 */\
1171         "psrad $" #shift ", %%mm0       \n\t"\
1172         "psrad $" #shift ", %%mm3       \n\t"\
1173         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
1174         "movd %%mm7, " #dst "           \n\t"\
1175         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
1176         "movd %%mm0, 16+" #dst "        \n\t"\
1177         "packssdw %%mm3, %%mm3          \n\t" /* A1-B1  a1-b1 */\
1178         "movd %%mm3, 96+" #dst "        \n\t"\
1179         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
1180         "movd %%mm4, 112+" #dst "       \n\t"\
1181         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
1182         "pmaddwd %%mm2, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1183         "pmaddwd 96(%2), %%mm2          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1184         "movq %%mm5, %%mm3              \n\t" /* A2             a2 */\
1185         "paddd %%mm4, %%mm3             \n\t" /* A2+B2          a2+b2 */\
1186         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
1187         "psrad $" #shift ", %%mm3       \n\t"\
1188         "psrad $" #shift ", %%mm5       \n\t"\
1189         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
1190         "paddd %%mm2, %%mm6             \n\t" /* A3+B3          a3+b3 */\
1191         "psubd %%mm2, %%mm4             \n\t" /* a3-B3          a3-b3 */\
1192         "psrad $" #shift ", %%mm6       \n\t"\
1193         "packssdw %%mm3, %%mm3          \n\t" /* A2+B2  a2+b2 */\
1194         "movd %%mm3, 32+" #dst "        \n\t"\
1195         "psrad $" #shift ", %%mm4       \n\t"\
1196         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1197         "movd %%mm6, 48+" #dst "        \n\t"\
1198         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
1199         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1200         "movd %%mm4, 64+" #dst "        \n\t"\
1201         "movd %%mm5, 80+" #dst "        \n\t"
1202
1203
1204 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
1205 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
1206 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
1207 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
1208 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
1209         "jmp 9f                         \n\t"
1210
1211
1212         "#" ASMALIGN(4)
1213         "7:                             \n\t"
1214 #undef IDCT
1215 #define IDCT(src0, src4, src1, src5, dst, shift) \
1216         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1217         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1218         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1219         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1220         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1221         "psrad $" #shift ", %%mm4       \n\t"\
1222         "psrad $" #shift ", %%mm0       \n\t"\
1223         "movq 8+" #src0 ", %%mm2        \n\t" /* R4     R0      r4      r0 */\
1224         "movq 16(%2), %%mm1             \n\t" /* C4     C4      C4      C4 */\
1225         "pmaddwd %%mm2, %%mm1           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1226         "movq 24(%2), %%mm7             \n\t" /* -C4    C4      -C4     C4 */\
1227         "pmaddwd %%mm7, %%mm2           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1228         "movq 32(%2), %%mm7             \n\t" /* C6     C2      C6      C2 */\
1229         "psrad $" #shift ", %%mm1       \n\t"\
1230         "packssdw %%mm1, %%mm4          \n\t" /* A0     a0 */\
1231         "movq %%mm4, " #dst "           \n\t"\
1232         "psrad $" #shift ", %%mm2       \n\t"\
1233         "packssdw %%mm2, %%mm0          \n\t" /* A1     a1 */\
1234         "movq %%mm0, 16+" #dst "        \n\t"\
1235         "movq %%mm0, 96+" #dst "        \n\t"\
1236         "movq %%mm4, 112+" #dst "       \n\t"\
1237         "movq %%mm0, 32+" #dst "        \n\t"\
1238         "movq %%mm4, 48+" #dst "        \n\t"\
1239         "movq %%mm4, 64+" #dst "        \n\t"\
1240         "movq %%mm0, 80+" #dst "        \n\t"
1241
1242 //IDCT(  src0,   src4,   src1,    src5,    dst, shift)
1243 IDCT(   0(%1), 64(%1), 32(%1),  96(%1),  0(%0), 20)
1244 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0), 20)
1245 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0), 20)
1246 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0), 20)
1247
1248
1249 #endif
1250
1251 /*
1252 Input
1253  00 40 04 44 20 60 24 64
1254  10 30 14 34 50 70 54 74
1255  01 41 03 43 21 61 23 63
1256  11 31 13 33 51 71 53 73
1257  02 42 06 46 22 62 26 66
1258  12 32 16 36 52 72 56 76
1259  05 45 07 47 25 65 27 67
1260  15 35 17 37 55 75 57 77
1261
1262 Temp
1263  00 04 10 14 20 24 30 34
1264  40 44 50 54 60 64 70 74
1265  01 03 11 13 21 23 31 33
1266  41 43 51 53 61 63 71 73
1267  02 06 12 16 22 26 32 36
1268  42 46 52 56 62 66 72 76
1269  05 07 15 17 25 27 35 37
1270  45 47 55 57 65 67 75 77
1271 */
1272
1273 "9: \n\t"
1274                 :: "r" (block), "r" (temp), "r" (coeffs)
1275                 : "%eax"
1276         );
1277 }
1278
1279 void ff_simple_idct_mmx(int16_t *block)
1280 {
1281     idct(block);
1282 }
1283
1284 //FIXME merge add/put into the idct
1285
1286 void ff_simple_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1287 {
1288     idct(block);
1289     put_pixels_clamped_mmx(block, dest, line_size);
1290 }
1291 void ff_simple_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1292 {
1293     idct(block);
1294     add_pixels_clamped_mmx(block, dest, line_size);
1295 }