git.sesse.net Git - ffmpeg/blob - libavcodec/i386/simple_idct_mmx.c

   1 /*
   2  * Simple IDCT MMX
   3  *
   4  * Copyright (c) 2001, 2002 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20 #include "../dsputil.h"
  21
  22 /*
  23 23170.475006
  24 22725.260826
  25 21406.727617
  26 19265.545870
  27 16384.000000
  28 12872.826198
  29 8866.956905
  30 4520.335430
  31 */
  32 #define C0 23170 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  33 #define C1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  34 #define C2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  35 #define C3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  36 #if 0
  37 #define C4 16384 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  38 #else
  39 #define C4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) - 0.5
  40 #endif
  41 #define C5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  42 #define C6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  43 #define C7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  44
  45 #define ROW_SHIFT 11
  46 #define COL_SHIFT 20 // 6
  47
  48 static const uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000ULL;
  49 static const uint64_t __attribute__((aligned(8))) d40000= 0x0000000000040000ULL;
  50 static int16_t __attribute__((aligned(8))) temp[64];
  51 static int16_t __attribute__((aligned(8))) coeffs[]= {
  52         1<<(ROW_SHIFT-1), 0, 1<<(ROW_SHIFT-1), 0,
  53 //      1<<(COL_SHIFT-1), 0, 1<<(COL_SHIFT-1), 0,
  54 //      0, 1<<(COL_SHIFT-1-16), 0, 1<<(COL_SHIFT-1-16),
  55         1<<(ROW_SHIFT-1), 1, 1<<(ROW_SHIFT-1), 0,
  56         // the 1 = ((1<<(COL_SHIFT-1))/C4)<<ROW_SHIFT :)
  57 //      0, 0, 0, 0,
  58 //      0, 0, 0, 0,
  59
  60  C4,  C4,  C4,  C4,
  61  C4, -C4,  C4, -C4,
  62
  63  C2,  C6,  C2,  C6,
  64  C6, -C2,  C6, -C2,
  65
  66  C1,  C3,  C1,  C3,
  67  C5,  C7,  C5,  C7,
  68
  69  C3, -C7,  C3, -C7,
  70 -C1, -C5, -C1, -C5,
  71
  72  C5, -C1,  C5, -C1,
  73  C7,  C3,  C7,  C3,
  74
  75  C7, -C5,  C7, -C5,
  76  C3, -C1,  C3, -C1
  77 };
  78
  79 #if 0
  80 static void unused_var_killer(){
  81         int a= wm1010 + d40000;
  82         temp[0]=a;
  83 }
  84
  85 static void inline idctCol (int16_t * col, int16_t *input)
  86 {
  87 #undef C0
  88 #undef C1
  89 #undef C2
  90 #undef C3
  91 #undef C4
  92 #undef C5
  93 #undef C6
  94 #undef C7
  95         int a0, a1, a2, a3, b0, b1, b2, b3;
  96         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  97         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  98         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  99         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 100         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 101         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 102         const int C6 = 8867; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 103         const int C7 = 4520; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 104 /*
 105         if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
 106                 col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
 107                         col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
 108                 return;
 109         }*/
 110
 111 col[8*0] = input[8*0 + 0];
 112 col[8*1] = input[8*2 + 0];
 113 col[8*2] = input[8*0 + 1];
 114 col[8*3] = input[8*2 + 1];
 115 col[8*4] = input[8*4 + 0];
 116 col[8*5] = input[8*6 + 0];
 117 col[8*6] = input[8*4 + 1];
 118 col[8*7] = input[8*6 + 1];
 119
 120         a0 = C4*col[8*0] + C2*col[8*2] + C4*col[8*4] + C6*col[8*6] + (1<<(COL_SHIFT-1));
 121         a1 = C4*col[8*0] + C6*col[8*2] - C4*col[8*4] - C2*col[8*6] + (1<<(COL_SHIFT-1));
 122         a2 = C4*col[8*0] - C6*col[8*2] - C4*col[8*4] + C2*col[8*6] + (1<<(COL_SHIFT-1));
 123         a3 = C4*col[8*0] - C2*col[8*2] + C4*col[8*4] - C6*col[8*6] + (1<<(COL_SHIFT-1));
 124
 125         b0 = C1*col[8*1] + C3*col[8*3] + C5*col[8*5] + C7*col[8*7];
 126         b1 = C3*col[8*1] - C7*col[8*3] - C1*col[8*5] - C5*col[8*7];
 127         b2 = C5*col[8*1] - C1*col[8*3] + C7*col[8*5] + C3*col[8*7];
 128         b3 = C7*col[8*1] - C5*col[8*3] + C3*col[8*5] - C1*col[8*7];
 129
 130         col[8*0] = (a0 + b0) >> COL_SHIFT;
 131         col[8*1] = (a1 + b1) >> COL_SHIFT;
 132         col[8*2] = (a2 + b2) >> COL_SHIFT;
 133         col[8*3] = (a3 + b3) >> COL_SHIFT;
 134         col[8*4] = (a3 - b3) >> COL_SHIFT;
 135         col[8*5] = (a2 - b2) >> COL_SHIFT;
 136         col[8*6] = (a1 - b1) >> COL_SHIFT;
 137         col[8*7] = (a0 - b0) >> COL_SHIFT;
 138 }
 139
 140 static void inline idctRow (int16_t * output, int16_t * input)
 141 {
 142         int16_t row[8];
 143
 144         int a0, a1, a2, a3, b0, b1, b2, b3;
 145         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 146         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 147         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 148         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 149         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 150         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 151         const int C6 = 8867; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 152         const int C7 = 4520; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 153
 154 row[0] = input[0];
 155 row[2] = input[1];
 156 row[4] = input[4];
 157 row[6] = input[5];
 158 row[1] = input[8];
 159 row[3] = input[9];
 160 row[5] = input[12];
 161 row[7] = input[13];
 162
 163         if( !(row[1] | row[2] |row[3] |row[4] |row[5] |row[6] | row[7]) ) {
 164                 row[0] = row[1] = row[2] = row[3] = row[4] =
 165                         row[5] = row[6] = row[7] = row[0]<<3;
 166         output[0] = row[0];
 167         output[2] = row[1];
 168         output[4] = row[2];
 169         output[6] = row[3];
 170         output[8] = row[4];
 171         output[10] = row[5];
 172         output[12] = row[6];
 173         output[14] = row[7];
 174                 return;
 175         }
 176
 177         a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + (1<<(ROW_SHIFT-1));
 178         a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + (1<<(ROW_SHIFT-1));
 179         a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + (1<<(ROW_SHIFT-1));
 180         a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + (1<<(ROW_SHIFT-1));
 181
 182         b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
 183         b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
 184         b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
 185         b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];
 186
 187         row[0] = (a0 + b0) >> ROW_SHIFT;
 188         row[1] = (a1 + b1) >> ROW_SHIFT;
 189         row[2] = (a2 + b2) >> ROW_SHIFT;
 190         row[3] = (a3 + b3) >> ROW_SHIFT;
 191         row[4] = (a3 - b3) >> ROW_SHIFT;
 192         row[5] = (a2 - b2) >> ROW_SHIFT;
 193         row[6] = (a1 - b1) >> ROW_SHIFT;
 194         row[7] = (a0 - b0) >> ROW_SHIFT;
 195
 196         output[0] = row[0];
 197         output[2] = row[1];
 198         output[4] = row[2];
 199         output[6] = row[3];
 200         output[8] = row[4];
 201         output[10] = row[5];
 202         output[12] = row[6];
 203         output[14] = row[7];
 204 }
 205 #endif
 206
 207 static inline void idct(int16_t *block)
 208 {
 209         asm volatile(
 210 #if 0 //Alternative, simpler variant
 211
 212 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 213         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 214         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 215         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 216         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 217         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 218         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 219         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 220         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 221         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 222         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 223         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 224         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 225         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 226         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 227         #rounder ", %%mm4                       \n\t"\
 228         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 229         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 230         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 231         "movq 56(%2), %%mm5                     \n\t" /* C7     C5      C7      C5 */\
 232         "pmaddwd %%mm3, %%mm5                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 233         #rounder ", %%mm0                       \n\t"\
 234         "paddd %%mm0, %%mm1                     \n\t" /* A1             a1 */\
 235         "paddd %%mm0, %%mm0                     \n\t" \
 236         "psubd %%mm1, %%mm0                     \n\t" /* A2             a2 */\
 237         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 238         "paddd %%mm5, %%mm7                     \n\t" /* B0             b0 */\
 239         "movq 72(%2), %%mm5                     \n\t" /* -C5    -C1     -C5     -C1 */\
 240         "pmaddwd %%mm3, %%mm5                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 241         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 242         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 243         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 244         "paddd %%mm2, %%mm5                     \n\t" /* B1             b1 */\
 245         "psrad $" #shift ", %%mm7               \n\t"\
 246         "psrad $" #shift ", %%mm4               \n\t"\
 247         "movq %%mm1, %%mm2                      \n\t" /* A1             a1 */\
 248         "paddd %%mm5, %%mm1                     \n\t" /* A1+B1          a1+b1 */\
 249         "psubd %%mm5, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 250         "psrad $" #shift ", %%mm1               \n\t"\
 251         "psrad $" #shift ", %%mm2               \n\t"\
 252         "packssdw %%mm1, %%mm7                  \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 253         "packssdw %%mm4, %%mm2                  \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 254         "movq %%mm7, " #dst "                   \n\t"\
 255         "movq " #src1 ", %%mm1                  \n\t" /* R3     R1      r3      r1 */\
 256         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 257         "movq %%mm2, 24+" #dst "                \n\t"\
 258         "pmaddwd %%mm1, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 259         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 260         "pmaddwd 96(%2), %%mm1                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 261         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 262         "movq %%mm0, %%mm2                      \n\t" /* A2             a2 */\
 263         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 264         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 265         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 266         "psubd %%mm4, %%mm0                     \n\t" /* a2-B2          a2-b2 */\
 267         "psrad $" #shift ", %%mm2               \n\t"\
 268         "psrad $" #shift ", %%mm0               \n\t"\
 269         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 270         "paddd %%mm1, %%mm3                     \n\t" /* B3             b3 */\
 271         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 272         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 273         "psrad $" #shift ", %%mm6               \n\t"\
 274         "packssdw %%mm6, %%mm2                  \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 275         "movq %%mm2, 8+" #dst "                 \n\t"\
 276         "psrad $" #shift ", %%mm4               \n\t"\
 277         "packssdw %%mm0, %%mm4                  \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 278         "movq %%mm4, 16+" #dst "                \n\t"\
 279
 280 #define COL_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 281         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 282         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 283         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 284         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 285         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 286         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 287         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 288         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 289         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 290         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 291         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 292         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 293         #rounder ", %%mm4                       \n\t"\
 294         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 295         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 296         #rounder ", %%mm0                       \n\t"\
 297         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 298         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 299         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 300         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 301         "paddd %%mm1, %%mm0                     \n\t" /* A1             a1 */\
 302         "psubd %%mm1, %%mm5                     \n\t" /* A2             a2 */\
 303         "movq 56(%2), %%mm1                     \n\t" /* C7     C5      C7      C5 */\
 304         "pmaddwd %%mm3, %%mm1                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 305         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 306         "paddd %%mm1, %%mm7                     \n\t" /* B0             b0 */\
 307         "movq 72(%2), %%mm1                     \n\t" /* -C5    -C1     -C5     -C1 */\
 308         "pmaddwd %%mm3, %%mm1                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 309         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 310         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 311         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 312         "paddd %%mm2, %%mm1                     \n\t" /* B1             b1 */\
 313         "psrad $" #shift ", %%mm7               \n\t"\
 314         "psrad $" #shift ", %%mm4               \n\t"\
 315         "movq %%mm0, %%mm2                      \n\t" /* A1             a1 */\
 316         "paddd %%mm1, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
 317         "psubd %%mm1, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 318         "psrad $" #shift ", %%mm0               \n\t"\
 319         "psrad $" #shift ", %%mm2               \n\t"\
 320         "packssdw %%mm7, %%mm7                  \n\t" /* A0+B0  a0+b0 */\
 321         "movd %%mm7, " #dst "                   \n\t"\
 322         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
 323         "movd %%mm0, 16+" #dst "                \n\t"\
 324         "packssdw %%mm2, %%mm2                  \n\t" /* A1-B1  a1-b1 */\
 325         "movd %%mm2, 96+" #dst "                \n\t"\
 326         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
 327         "movd %%mm4, 112+" #dst "               \n\t"\
 328         "movq " #src1 ", %%mm0                  \n\t" /* R3     R1      r3      r1 */\
 329         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 330         "pmaddwd %%mm0, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 331         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 332         "pmaddwd 96(%2), %%mm0                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 333         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 334         "movq %%mm5, %%mm2                      \n\t" /* A2             a2 */\
 335         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 336         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 337         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 338         "psubd %%mm4, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
 339         "psrad $" #shift ", %%mm2               \n\t"\
 340         "psrad $" #shift ", %%mm5               \n\t"\
 341         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 342         "paddd %%mm0, %%mm3                     \n\t" /* B3             b3 */\
 343         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 344         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 345         "psrad $" #shift ", %%mm6               \n\t"\
 346         "psrad $" #shift ", %%mm4               \n\t"\
 347         "packssdw %%mm2, %%mm2                  \n\t" /* A2+B2  a2+b2 */\
 348         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
 349         "movd %%mm2, 32+" #dst "                \n\t"\
 350         "packssdw %%mm4, %%mm4                  \n\t" /* A3-B3  a3-b3 */\
 351         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
 352         "movd %%mm6, 48+" #dst "                \n\t"\
 353         "movd %%mm4, 64+" #dst "                \n\t"\
 354         "movd %%mm5, 80+" #dst "                \n\t"\
 355
 356
 357 #define DC_COND_ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 358         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 359         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 360         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 361         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 362         "movq "MANGLE(wm1010)", %%mm4           \n\t"\
 363         "pand %%mm0, %%mm4                      \n\t"\
 364         "por %%mm1, %%mm4                       \n\t"\
 365         "por %%mm2, %%mm4                       \n\t"\
 366         "por %%mm3, %%mm4                       \n\t"\
 367         "packssdw %%mm4,%%mm4                   \n\t"\
 368         "movd %%mm4, %%eax                      \n\t"\
 369         "orl %%eax, %%eax                       \n\t"\
 370         "jz 1f                                  \n\t"\
 371         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 372         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 373         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 374         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 375         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 376         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 377         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 378         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 379         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 380         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 381         #rounder ", %%mm4                       \n\t"\
 382         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 383         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 384         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 385         "movq 56(%2), %%mm5                     \n\t" /* C7     C5      C7      C5 */\
 386         "pmaddwd %%mm3, %%mm5                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 387         #rounder ", %%mm0                       \n\t"\
 388         "paddd %%mm0, %%mm1                     \n\t" /* A1             a1 */\
 389         "paddd %%mm0, %%mm0                     \n\t" \
 390         "psubd %%mm1, %%mm0                     \n\t" /* A2             a2 */\
 391         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 392         "paddd %%mm5, %%mm7                     \n\t" /* B0             b0 */\
 393         "movq 72(%2), %%mm5                     \n\t" /* -C5    -C1     -C5     -C1 */\
 394         "pmaddwd %%mm3, %%mm5                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 395         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 396         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 397         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 398         "paddd %%mm2, %%mm5                     \n\t" /* B1             b1 */\
 399         "psrad $" #shift ", %%mm7               \n\t"\
 400         "psrad $" #shift ", %%mm4               \n\t"\
 401         "movq %%mm1, %%mm2                      \n\t" /* A1             a1 */\
 402         "paddd %%mm5, %%mm1                     \n\t" /* A1+B1          a1+b1 */\
 403         "psubd %%mm5, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 404         "psrad $" #shift ", %%mm1               \n\t"\
 405         "psrad $" #shift ", %%mm2               \n\t"\
 406         "packssdw %%mm1, %%mm7                  \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 407         "packssdw %%mm4, %%mm2                  \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 408         "movq %%mm7, " #dst "                   \n\t"\
 409         "movq " #src1 ", %%mm1                  \n\t" /* R3     R1      r3      r1 */\
 410         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 411         "movq %%mm2, 24+" #dst "                \n\t"\
 412         "pmaddwd %%mm1, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 413         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 414         "pmaddwd 96(%2), %%mm1                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 415         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 416         "movq %%mm0, %%mm2                      \n\t" /* A2             a2 */\
 417         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 418         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 419         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 420         "psubd %%mm4, %%mm0                     \n\t" /* a2-B2          a2-b2 */\
 421         "psrad $" #shift ", %%mm2               \n\t"\
 422         "psrad $" #shift ", %%mm0               \n\t"\
 423         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 424         "paddd %%mm1, %%mm3                     \n\t" /* B3             b3 */\
 425         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 426         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 427         "psrad $" #shift ", %%mm6               \n\t"\
 428         "packssdw %%mm6, %%mm2                  \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 429         "movq %%mm2, 8+" #dst "                 \n\t"\
 430         "psrad $" #shift ", %%mm4               \n\t"\
 431         "packssdw %%mm0, %%mm4                  \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 432         "movq %%mm4, 16+" #dst "                \n\t"\
 433         "jmp 2f                                 \n\t"\
 434         "1:                                     \n\t"\
 435         "pslld $16, %%mm0                       \n\t"\
 436         "#paddd "MANGLE(d40000)", %%mm0         \n\t"\
 437         "psrad $13, %%mm0                       \n\t"\
 438         "packssdw %%mm0, %%mm0                  \n\t"\
 439         "movq %%mm0, " #dst "                   \n\t"\
 440         "movq %%mm0, 8+" #dst "                 \n\t"\
 441         "movq %%mm0, 16+" #dst "                \n\t"\
 442         "movq %%mm0, 24+" #dst "                \n\t"\
 443         "2:                                     \n\t"
 444
 445
 446 //IDCT(      src0,   src4,   src1,   src5,    dst,    rounder, shift)
 447 ROW_IDCT(    (%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 448 /*ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1), paddd (%2), 11)
 449 ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1), paddd (%2), 11)
 450 ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1), paddd (%2), 11)*/
 451
 452 DC_COND_ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11)
 453 DC_COND_ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11)
 454 DC_COND_ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11)
 455
 456
 457 //IDCT(      src0,   src4,   src1,    src5,    dst, rounder, shift)
 458 COL_IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 459 COL_IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 460 COL_IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 461 COL_IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 462
 463 #else
 464
 465 #define DC_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 466         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 467         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 468         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 469         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 470         "movq "MANGLE(wm1010)", %%mm4           \n\t"\
 471         "pand %%mm0, %%mm4                      \n\t"\
 472         "por %%mm1, %%mm4                       \n\t"\
 473         "por %%mm2, %%mm4                       \n\t"\
 474         "por %%mm3, %%mm4                       \n\t"\
 475         "packssdw %%mm4,%%mm4                   \n\t"\
 476         "movd %%mm4, %%eax                      \n\t"\
 477         "orl %%eax, %%eax                       \n\t"\
 478         "jz 1f                                  \n\t"\
 479         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 480         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 481         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 482         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 483         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 484         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 485         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 486         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 487         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 488         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 489         #rounder ", %%mm4                       \n\t"\
 490         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 491         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 492         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 493         "movq 56(%2), %%mm5                     \n\t" /* C7     C5      C7      C5 */\
 494         "pmaddwd %%mm3, %%mm5                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 495         #rounder ", %%mm0                       \n\t"\
 496         "paddd %%mm0, %%mm1                     \n\t" /* A1             a1 */\
 497         "paddd %%mm0, %%mm0                     \n\t" \
 498         "psubd %%mm1, %%mm0                     \n\t" /* A2             a2 */\
 499         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 500         "paddd %%mm5, %%mm7                     \n\t" /* B0             b0 */\
 501         "movq 72(%2), %%mm5                     \n\t" /* -C5    -C1     -C5     -C1 */\
 502         "pmaddwd %%mm3, %%mm5                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 503         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 504         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 505         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 506         "paddd %%mm2, %%mm5                     \n\t" /* B1             b1 */\
 507         "psrad $" #shift ", %%mm7               \n\t"\
 508         "psrad $" #shift ", %%mm4               \n\t"\
 509         "movq %%mm1, %%mm2                      \n\t" /* A1             a1 */\
 510         "paddd %%mm5, %%mm1                     \n\t" /* A1+B1          a1+b1 */\
 511         "psubd %%mm5, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 512         "psrad $" #shift ", %%mm1               \n\t"\
 513         "psrad $" #shift ", %%mm2               \n\t"\
 514         "packssdw %%mm1, %%mm7                  \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 515         "packssdw %%mm4, %%mm2                  \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 516         "movq %%mm7, " #dst "                   \n\t"\
 517         "movq " #src1 ", %%mm1                  \n\t" /* R3     R1      r3      r1 */\
 518         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 519         "movq %%mm2, 24+" #dst "                \n\t"\
 520         "pmaddwd %%mm1, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 521         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 522         "pmaddwd 96(%2), %%mm1                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 523         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 524         "movq %%mm0, %%mm2                      \n\t" /* A2             a2 */\
 525         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 526         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 527         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 528         "psubd %%mm4, %%mm0                     \n\t" /* a2-B2          a2-b2 */\
 529         "psrad $" #shift ", %%mm2               \n\t"\
 530         "psrad $" #shift ", %%mm0               \n\t"\
 531         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 532         "paddd %%mm1, %%mm3                     \n\t" /* B3             b3 */\
 533         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 534         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 535         "psrad $" #shift ", %%mm6               \n\t"\
 536         "packssdw %%mm6, %%mm2                  \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 537         "movq %%mm2, 8+" #dst "                 \n\t"\
 538         "psrad $" #shift ", %%mm4               \n\t"\
 539         "packssdw %%mm0, %%mm4                  \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 540         "movq %%mm4, 16+" #dst "                \n\t"\
 541         "jmp 2f                                 \n\t"\
 542         "1:                                     \n\t"\
 543         "pslld $16, %%mm0                       \n\t"\
 544         "paddd "MANGLE(d40000)", %%mm0          \n\t"\
 545         "psrad $13, %%mm0                       \n\t"\
 546         "packssdw %%mm0, %%mm0                  \n\t"\
 547         "movq %%mm0, " #dst "                   \n\t"\
 548         "movq %%mm0, 8+" #dst "                 \n\t"\
 549         "movq %%mm0, 16+" #dst "                \n\t"\
 550         "movq %%mm0, 24+" #dst "                \n\t"\
 551         "2:                                     \n\t"
 552
 553 #define Z_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift, bt) \
 554         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 555         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 556         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 557         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 558         "movq %%mm0, %%mm4                      \n\t"\
 559         "por %%mm1, %%mm4                       \n\t"\
 560         "por %%mm2, %%mm4                       \n\t"\
 561         "por %%mm3, %%mm4                       \n\t"\
 562         "packssdw %%mm4,%%mm4                   \n\t"\
 563         "movd %%mm4, %%eax                      \n\t"\
 564         "orl %%eax, %%eax                       \n\t"\
 565         "jz " #bt "                             \n\t"\
 566         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 567         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 568         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 569         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 570         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 571         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 572         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 573         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 574         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 575         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 576         #rounder ", %%mm4                       \n\t"\
 577         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 578         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 579         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 580         "movq 56(%2), %%mm5                     \n\t" /* C7     C5      C7      C5 */\
 581         "pmaddwd %%mm3, %%mm5                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 582         #rounder ", %%mm0                       \n\t"\
 583         "paddd %%mm0, %%mm1                     \n\t" /* A1             a1 */\
 584         "paddd %%mm0, %%mm0                     \n\t" \
 585         "psubd %%mm1, %%mm0                     \n\t" /* A2             a2 */\
 586         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 587         "paddd %%mm5, %%mm7                     \n\t" /* B0             b0 */\
 588         "movq 72(%2), %%mm5                     \n\t" /* -C5    -C1     -C5     -C1 */\
 589         "pmaddwd %%mm3, %%mm5                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 590         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 591         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 592         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 593         "paddd %%mm2, %%mm5                     \n\t" /* B1             b1 */\
 594         "psrad $" #shift ", %%mm7               \n\t"\
 595         "psrad $" #shift ", %%mm4               \n\t"\
 596         "movq %%mm1, %%mm2                      \n\t" /* A1             a1 */\
 597         "paddd %%mm5, %%mm1                     \n\t" /* A1+B1          a1+b1 */\
 598         "psubd %%mm5, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 599         "psrad $" #shift ", %%mm1               \n\t"\
 600         "psrad $" #shift ", %%mm2               \n\t"\
 601         "packssdw %%mm1, %%mm7                  \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 602         "packssdw %%mm4, %%mm2                  \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 603         "movq %%mm7, " #dst "                   \n\t"\
 604         "movq " #src1 ", %%mm1                  \n\t" /* R3     R1      r3      r1 */\
 605         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 606         "movq %%mm2, 24+" #dst "                \n\t"\
 607         "pmaddwd %%mm1, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 608         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 609         "pmaddwd 96(%2), %%mm1                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 610         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 611         "movq %%mm0, %%mm2                      \n\t" /* A2             a2 */\
 612         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 613         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 614         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 615         "psubd %%mm4, %%mm0                     \n\t" /* a2-B2          a2-b2 */\
 616         "psrad $" #shift ", %%mm2               \n\t"\
 617         "psrad $" #shift ", %%mm0               \n\t"\
 618         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 619         "paddd %%mm1, %%mm3                     \n\t" /* B3             b3 */\
 620         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 621         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 622         "psrad $" #shift ", %%mm6               \n\t"\
 623         "packssdw %%mm6, %%mm2                  \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 624         "movq %%mm2, 8+" #dst "                 \n\t"\
 625         "psrad $" #shift ", %%mm4               \n\t"\
 626         "packssdw %%mm0, %%mm4                  \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 627         "movq %%mm4, 16+" #dst "                \n\t"\
 628
 629 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 630         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 631         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 632         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 633         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 634         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 635         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 636         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 637         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 638         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 639         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 640         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 641         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 642         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 643         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 644         #rounder ", %%mm4                       \n\t"\
 645         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 646         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 647         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 648         "movq 56(%2), %%mm5                     \n\t" /* C7     C5      C7      C5 */\
 649         "pmaddwd %%mm3, %%mm5                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 650         #rounder ", %%mm0                       \n\t"\
 651         "paddd %%mm0, %%mm1                     \n\t" /* A1             a1 */\
 652         "paddd %%mm0, %%mm0                     \n\t" \
 653         "psubd %%mm1, %%mm0                     \n\t" /* A2             a2 */\
 654         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 655         "paddd %%mm5, %%mm7                     \n\t" /* B0             b0 */\
 656         "movq 72(%2), %%mm5                     \n\t" /* -C5    -C1     -C5     -C1 */\
 657         "pmaddwd %%mm3, %%mm5                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 658         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 659         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 660         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 661         "paddd %%mm2, %%mm5                     \n\t" /* B1             b1 */\
 662         "psrad $" #shift ", %%mm7               \n\t"\
 663         "psrad $" #shift ", %%mm4               \n\t"\
 664         "movq %%mm1, %%mm2                      \n\t" /* A1             a1 */\
 665         "paddd %%mm5, %%mm1                     \n\t" /* A1+B1          a1+b1 */\
 666         "psubd %%mm5, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 667         "psrad $" #shift ", %%mm1               \n\t"\
 668         "psrad $" #shift ", %%mm2               \n\t"\
 669         "packssdw %%mm1, %%mm7                  \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 670         "packssdw %%mm4, %%mm2                  \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 671         "movq %%mm7, " #dst "                   \n\t"\
 672         "movq " #src1 ", %%mm1                  \n\t" /* R3     R1      r3      r1 */\
 673         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 674         "movq %%mm2, 24+" #dst "                \n\t"\
 675         "pmaddwd %%mm1, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 676         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 677         "pmaddwd 96(%2), %%mm1                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 678         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 679         "movq %%mm0, %%mm2                      \n\t" /* A2             a2 */\
 680         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 681         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 682         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 683         "psubd %%mm4, %%mm0                     \n\t" /* a2-B2          a2-b2 */\
 684         "psrad $" #shift ", %%mm2               \n\t"\
 685         "psrad $" #shift ", %%mm0               \n\t"\
 686         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 687         "paddd %%mm1, %%mm3                     \n\t" /* B3             b3 */\
 688         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 689         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 690         "psrad $" #shift ", %%mm6               \n\t"\
 691         "packssdw %%mm6, %%mm2                  \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 692         "movq %%mm2, 8+" #dst "                 \n\t"\
 693         "psrad $" #shift ", %%mm4               \n\t"\
 694         "packssdw %%mm0, %%mm4                  \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 695         "movq %%mm4, 16+" #dst "                \n\t"\
 696
 697 //IDCT(         src0,   src4,   src1,   src5,    dst,   rounder, shift)
 698 DC_COND_IDCT(  0(%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 699 Z_COND_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11, 4f)
 700 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 2f)
 701 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 1f)
 702
 703 #undef IDCT
 704 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 705         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 706         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 707         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 708         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 709         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 710         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 711         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 712         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 713         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 714         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 715         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 716         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 717         #rounder ", %%mm4                       \n\t"\
 718         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 719         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 720         #rounder ", %%mm0                       \n\t"\
 721         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 722         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 723         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 724         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 725         "paddd %%mm1, %%mm0                     \n\t" /* A1             a1 */\
 726         "psubd %%mm1, %%mm5                     \n\t" /* A2             a2 */\
 727         "movq 56(%2), %%mm1                     \n\t" /* C7     C5      C7      C5 */\
 728         "pmaddwd %%mm3, %%mm1                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 729         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 730         "paddd %%mm1, %%mm7                     \n\t" /* B0             b0 */\
 731         "movq 72(%2), %%mm1                     \n\t" /* -C5    -C1     -C5     -C1 */\
 732         "pmaddwd %%mm3, %%mm1                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 733         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 734         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 735         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 736         "paddd %%mm2, %%mm1                     \n\t" /* B1             b1 */\
 737         "psrad $" #shift ", %%mm7               \n\t"\
 738         "psrad $" #shift ", %%mm4               \n\t"\
 739         "movq %%mm0, %%mm2                      \n\t" /* A1             a1 */\
 740         "paddd %%mm1, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
 741         "psubd %%mm1, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 742         "psrad $" #shift ", %%mm0               \n\t"\
 743         "psrad $" #shift ", %%mm2               \n\t"\
 744         "packssdw %%mm7, %%mm7                  \n\t" /* A0+B0  a0+b0 */\
 745         "movd %%mm7, " #dst "                   \n\t"\
 746         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
 747         "movd %%mm0, 16+" #dst "                \n\t"\
 748         "packssdw %%mm2, %%mm2                  \n\t" /* A1-B1  a1-b1 */\
 749         "movd %%mm2, 96+" #dst "                \n\t"\
 750         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
 751         "movd %%mm4, 112+" #dst "               \n\t"\
 752         "movq " #src1 ", %%mm0                  \n\t" /* R3     R1      r3      r1 */\
 753         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 754         "pmaddwd %%mm0, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 755         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 756         "pmaddwd 96(%2), %%mm0                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 757         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 758         "movq %%mm5, %%mm2                      \n\t" /* A2             a2 */\
 759         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 760         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 761         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 762         "psubd %%mm4, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
 763         "psrad $" #shift ", %%mm2               \n\t"\
 764         "psrad $" #shift ", %%mm5               \n\t"\
 765         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 766         "paddd %%mm0, %%mm3                     \n\t" /* B3             b3 */\
 767         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 768         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 769         "psrad $" #shift ", %%mm6               \n\t"\
 770         "psrad $" #shift ", %%mm4               \n\t"\
 771         "packssdw %%mm2, %%mm2                  \n\t" /* A2+B2  a2+b2 */\
 772         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
 773         "movd %%mm2, 32+" #dst "                \n\t"\
 774         "packssdw %%mm4, %%mm4                  \n\t" /* A3-B3  a3-b3 */\
 775         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
 776         "movd %%mm6, 48+" #dst "                \n\t"\
 777         "movd %%mm4, 64+" #dst "                \n\t"\
 778         "movd %%mm5, 80+" #dst "                \n\t"
 779
 780
 781 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 782 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 783 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 784 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 785 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 786         "jmp 9f                                 \n\t"
 787
 788         "#.balign 16                            \n\t"\
 789         "4:                                     \n\t"
 790 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 6f)
 791 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 5f)
 792
 793 #undef IDCT
 794 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 795         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 796         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
 797         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 798         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 799         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 800         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 801         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 802         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
 803         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 804         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
 805         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 806         #rounder ", %%mm4                       \n\t"\
 807         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 808         #rounder ", %%mm0                       \n\t"\
 809         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
 810         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
 811         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 812         "paddd %%mm1, %%mm0                     \n\t" /* A1             a1 */\
 813         "psubd %%mm1, %%mm5                     \n\t" /* A2             a2 */\
 814         "movq 56(%2), %%mm1                     \n\t" /* C7     C5      C7      C5 */\
 815         "pmaddwd %%mm3, %%mm1                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 816         "movq 72(%2), %%mm7                     \n\t" /* -C5    -C1     -C5     -C1 */\
 817         "pmaddwd %%mm3, %%mm7                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 818         "paddd %%mm4, %%mm1                     \n\t" /* A0+B0          a0+b0 */\
 819         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 820         "psubd %%mm1, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 821         "psrad $" #shift ", %%mm1               \n\t"\
 822         "psrad $" #shift ", %%mm4               \n\t"\
 823         "movq %%mm0, %%mm2                      \n\t" /* A1             a1 */\
 824         "paddd %%mm7, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
 825         "psubd %%mm7, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 826         "psrad $" #shift ", %%mm0               \n\t"\
 827         "psrad $" #shift ", %%mm2               \n\t"\
 828         "packssdw %%mm1, %%mm1                  \n\t" /* A0+B0  a0+b0 */\
 829         "movd %%mm1, " #dst "                   \n\t"\
 830         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
 831         "movd %%mm0, 16+" #dst "                \n\t"\
 832         "packssdw %%mm2, %%mm2                  \n\t" /* A1-B1  a1-b1 */\
 833         "movd %%mm2, 96+" #dst "                \n\t"\
 834         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
 835         "movd %%mm4, 112+" #dst "               \n\t"\
 836         "movq 88(%2), %%mm1                     \n\t" /* C3     C7      C3      C7 */\
 837         "pmaddwd %%mm3, %%mm1                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 838         "movq %%mm5, %%mm2                      \n\t" /* A2             a2 */\
 839         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 840         "paddd %%mm1, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 841         "psubd %%mm1, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
 842         "psrad $" #shift ", %%mm2               \n\t"\
 843         "psrad $" #shift ", %%mm5               \n\t"\
 844         "movq %%mm6, %%mm1                      \n\t" /* A3             a3 */\
 845         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 846         "psubd %%mm3, %%mm1                     \n\t" /* a3-B3          a3-b3 */\
 847         "psrad $" #shift ", %%mm6               \n\t"\
 848         "psrad $" #shift ", %%mm1               \n\t"\
 849         "packssdw %%mm2, %%mm2                  \n\t" /* A2+B2  a2+b2 */\
 850         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
 851         "movd %%mm2, 32+" #dst "                \n\t"\
 852         "packssdw %%mm1, %%mm1                  \n\t" /* A3-B3  a3-b3 */\
 853         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
 854         "movd %%mm6, 48+" #dst "                \n\t"\
 855         "movd %%mm1, 64+" #dst "                \n\t"\
 856         "movd %%mm5, 80+" #dst "                \n\t"
 857
 858 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 859 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 860 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 861 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 862 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 863         "jmp 9f                                 \n\t"
 864
 865         "#.balign 16                            \n\t"\
 866         "6:                                     \n\t"
 867 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 7f)
 868
 869 #undef IDCT
 870 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 871         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 872         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 873         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 874         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 875         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 876         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 877         #rounder ", %%mm4                       \n\t"\
 878         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 879         #rounder ", %%mm0                       \n\t"\
 880         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 881         "movq 56(%2), %%mm1                     \n\t" /* C7     C5      C7      C5 */\
 882         "pmaddwd %%mm3, %%mm1                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 883         "movq 72(%2), %%mm7                     \n\t" /* -C5    -C1     -C5     -C1 */\
 884         "pmaddwd %%mm3, %%mm7                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 885         "paddd %%mm4, %%mm1                     \n\t" /* A0+B0          a0+b0 */\
 886         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 887         "psubd %%mm1, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 888         "psrad $" #shift ", %%mm1               \n\t"\
 889         "psrad $" #shift ", %%mm4               \n\t"\
 890         "movq %%mm0, %%mm2                      \n\t" /* A1             a1 */\
 891         "paddd %%mm7, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
 892         "psubd %%mm7, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 893         "psrad $" #shift ", %%mm0               \n\t"\
 894         "psrad $" #shift ", %%mm2               \n\t"\
 895         "packssdw %%mm1, %%mm1                  \n\t" /* A0+B0  a0+b0 */\
 896         "movd %%mm1, " #dst "                   \n\t"\
 897         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
 898         "movd %%mm0, 16+" #dst "                \n\t"\
 899         "packssdw %%mm2, %%mm2                  \n\t" /* A1-B1  a1-b1 */\
 900         "movd %%mm2, 96+" #dst "                \n\t"\
 901         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
 902         "movd %%mm4, 112+" #dst "               \n\t"\
 903         "movq 88(%2), %%mm1                     \n\t" /* C3     C7      C3      C7 */\
 904         "pmaddwd %%mm3, %%mm1                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 905         "movq %%mm5, %%mm2                      \n\t" /* A2             a2 */\
 906         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 907         "paddd %%mm1, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 908         "psubd %%mm1, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
 909         "psrad $" #shift ", %%mm2               \n\t"\
 910         "psrad $" #shift ", %%mm5               \n\t"\
 911         "movq %%mm6, %%mm1                      \n\t" /* A3             a3 */\
 912         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 913         "psubd %%mm3, %%mm1                     \n\t" /* a3-B3          a3-b3 */\
 914         "psrad $" #shift ", %%mm6               \n\t"\
 915         "psrad $" #shift ", %%mm1               \n\t"\
 916         "packssdw %%mm2, %%mm2                  \n\t" /* A2+B2  a2+b2 */\
 917         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
 918         "movd %%mm2, 32+" #dst "                \n\t"\
 919         "packssdw %%mm1, %%mm1                  \n\t" /* A3-B3  a3-b3 */\
 920         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
 921         "movd %%mm6, 48+" #dst "                \n\t"\
 922         "movd %%mm1, 64+" #dst "                \n\t"\
 923         "movd %%mm5, 80+" #dst "                \n\t"
 924
 925
 926 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 927 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 928 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 929 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 930 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 931         "jmp 9f                                 \n\t"
 932
 933         "#.balign 16                            \n\t"\
 934         "2:                                     \n\t"
 935 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 3f)
 936
 937 #undef IDCT
 938 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 939         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
 940         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
 941         "movq " #src5 ", %%mm3                  \n\t" /* R7     R5      r7      r5 */\
 942         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
 943         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 944         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
 945         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 946         #rounder ", %%mm4                       \n\t"\
 947         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 948         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
 949         #rounder ", %%mm0                       \n\t"\
 950         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 951         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 952         "movq 56(%2), %%mm1                     \n\t" /* C7     C5      C7      C5 */\
 953         "pmaddwd %%mm3, %%mm1                   \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 954         "pmaddwd 64(%2), %%mm2                  \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 955         "paddd %%mm1, %%mm7                     \n\t" /* B0             b0 */\
 956         "movq 72(%2), %%mm1                     \n\t" /* -C5    -C1     -C5     -C1 */\
 957         "pmaddwd %%mm3, %%mm1                   \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 958         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
 959         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
 960         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
 961         "paddd %%mm2, %%mm1                     \n\t" /* B1             b1 */\
 962         "psrad $" #shift ", %%mm7               \n\t"\
 963         "psrad $" #shift ", %%mm4               \n\t"\
 964         "movq %%mm0, %%mm2                      \n\t" /* A1             a1 */\
 965         "paddd %%mm1, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
 966         "psubd %%mm1, %%mm2                     \n\t" /* A1-B1          a1-b1 */\
 967         "psrad $" #shift ", %%mm0               \n\t"\
 968         "psrad $" #shift ", %%mm2               \n\t"\
 969         "packssdw %%mm7, %%mm7                  \n\t" /* A0+B0  a0+b0 */\
 970         "movd %%mm7, " #dst "                   \n\t"\
 971         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
 972         "movd %%mm0, 16+" #dst "                \n\t"\
 973         "packssdw %%mm2, %%mm2                  \n\t" /* A1-B1  a1-b1 */\
 974         "movd %%mm2, 96+" #dst "                \n\t"\
 975         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
 976         "movd %%mm4, 112+" #dst "               \n\t"\
 977         "movq " #src1 ", %%mm0                  \n\t" /* R3     R1      r3      r1 */\
 978         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
 979         "pmaddwd %%mm0, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 980         "movq 88(%2), %%mm7                     \n\t" /* C3     C7      C3      C7 */\
 981         "pmaddwd 96(%2), %%mm0                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 982         "pmaddwd %%mm3, %%mm7                   \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 983         "movq %%mm5, %%mm2                      \n\t" /* A2             a2 */\
 984         "pmaddwd 104(%2), %%mm3                 \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 985         "paddd %%mm7, %%mm4                     \n\t" /* B2             b2 */\
 986         "paddd %%mm4, %%mm2                     \n\t" /* A2+B2          a2+b2 */\
 987         "psubd %%mm4, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
 988         "psrad $" #shift ", %%mm2               \n\t"\
 989         "psrad $" #shift ", %%mm5               \n\t"\
 990         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
 991         "paddd %%mm0, %%mm3                     \n\t" /* B3             b3 */\
 992         "paddd %%mm3, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
 993         "psubd %%mm3, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
 994         "psrad $" #shift ", %%mm6               \n\t"\
 995         "psrad $" #shift ", %%mm4               \n\t"\
 996         "packssdw %%mm2, %%mm2                  \n\t" /* A2+B2  a2+b2 */\
 997         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
 998         "movd %%mm2, 32+" #dst "                \n\t"\
 999         "packssdw %%mm4, %%mm4                  \n\t" /* A3-B3  a3-b3 */\
1000         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
1001         "movd %%mm6, 48+" #dst "                \n\t"\
1002         "movd %%mm4, 64+" #dst "                \n\t"\
1003         "movd %%mm5, 80+" #dst "                \n\t"
1004
1005 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1006 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1007 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1008 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1009 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1010         "jmp 9f                                 \n\t"
1011
1012         "#.balign 16                            \n\t"\
1013         "3:                                     \n\t"
1014 #undef IDCT
1015 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1016         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
1017         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
1018         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
1019         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1020         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
1021         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1022         #rounder ", %%mm4                       \n\t"\
1023         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1024         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
1025         #rounder ", %%mm0                       \n\t"\
1026         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1027         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1028         "movq 64(%2), %%mm3                     \n\t"\
1029         "pmaddwd %%mm2, %%mm3                   \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1030         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
1031         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
1032         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
1033         "psrad $" #shift ", %%mm7               \n\t"\
1034         "psrad $" #shift ", %%mm4               \n\t"\
1035         "movq %%mm0, %%mm1                      \n\t" /* A1             a1 */\
1036         "paddd %%mm3, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
1037         "psubd %%mm3, %%mm1                     \n\t" /* A1-B1          a1-b1 */\
1038         "psrad $" #shift ", %%mm0               \n\t"\
1039         "psrad $" #shift ", %%mm1               \n\t"\
1040         "packssdw %%mm7, %%mm7                  \n\t" /* A0+B0  a0+b0 */\
1041         "movd %%mm7, " #dst "                   \n\t"\
1042         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
1043         "movd %%mm0, 16+" #dst "                \n\t"\
1044         "packssdw %%mm1, %%mm1                  \n\t" /* A1-B1  a1-b1 */\
1045         "movd %%mm1, 96+" #dst "                \n\t"\
1046         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
1047         "movd %%mm4, 112+" #dst "               \n\t"\
1048         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
1049         "pmaddwd %%mm2, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1050         "pmaddwd 96(%2), %%mm2                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1051         "movq %%mm5, %%mm1                      \n\t" /* A2             a2 */\
1052         "paddd %%mm4, %%mm1                     \n\t" /* A2+B2          a2+b2 */\
1053         "psubd %%mm4, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
1054         "psrad $" #shift ", %%mm1               \n\t"\
1055         "psrad $" #shift ", %%mm5               \n\t"\
1056         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
1057         "paddd %%mm2, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
1058         "psubd %%mm2, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
1059         "psrad $" #shift ", %%mm6               \n\t"\
1060         "psrad $" #shift ", %%mm4               \n\t"\
1061         "packssdw %%mm1, %%mm1                  \n\t" /* A2+B2  a2+b2 */\
1062         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
1063         "movd %%mm1, 32+" #dst "                \n\t"\
1064         "packssdw %%mm4, %%mm4                  \n\t" /* A3-B3  a3-b3 */\
1065         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
1066         "movd %%mm6, 48+" #dst "                \n\t"\
1067         "movd %%mm4, 64+" #dst "                \n\t"\
1068         "movd %%mm5, 80+" #dst "                \n\t"
1069
1070
1071 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1072 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1073 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1074 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1075 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1076         "jmp 9f                                 \n\t"
1077
1078         "#.balign 16                            \n\t"\
1079         "5:                                     \n\t"
1080 #undef IDCT
1081 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1082         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
1083         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
1084         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
1085         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1086         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
1087         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1088         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
1089         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1090         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
1091         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1092         #rounder ", %%mm4                       \n\t"\
1093         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1094         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
1095         #rounder ", %%mm0                       \n\t"\
1096         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
1097         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1098         "paddd %%mm1, %%mm0                     \n\t" /* A1             a1 */\
1099         "psubd %%mm1, %%mm5                     \n\t" /* A2             a2 */\
1100         "movq 8+" #src0 ", %%mm2                \n\t" /* R4     R0      r4      r0 */\
1101         "movq 8+" #src4 ", %%mm3                \n\t" /* R6     R2      r6      r2 */\
1102         "movq 16(%2), %%mm1                     \n\t" /* C4     C4      C4      C4 */\
1103         "pmaddwd %%mm2, %%mm1                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1104         "movq 24(%2), %%mm7                     \n\t" /* -C4    C4      -C4     C4 */\
1105         "pmaddwd %%mm7, %%mm2                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1106         "movq 32(%2), %%mm7                     \n\t" /* C6     C2      C6      C2 */\
1107         "pmaddwd %%mm3, %%mm7                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1108         "pmaddwd 40(%2), %%mm3                  \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1109         #rounder ", %%mm1                       \n\t"\
1110         "paddd %%mm1, %%mm7                     \n\t" /* A0             a0 */\
1111         "paddd %%mm1, %%mm1                     \n\t" /* 2C0            2c0 */\
1112         #rounder ", %%mm2                       \n\t"\
1113         "psubd %%mm7, %%mm1                     \n\t" /* A3             a3 */\
1114         "paddd %%mm2, %%mm3                     \n\t" /* A1             a1 */\
1115         "paddd %%mm2, %%mm2                     \n\t" /* 2C1            2c1 */\
1116         "psubd %%mm3, %%mm2                     \n\t" /* A2             a2 */\
1117         "psrad $" #shift ", %%mm4               \n\t"\
1118         "psrad $" #shift ", %%mm7               \n\t"\
1119         "psrad $" #shift ", %%mm3               \n\t"\
1120         "packssdw %%mm7, %%mm4                  \n\t" /* A0     a0 */\
1121         "movq %%mm4, " #dst "                   \n\t"\
1122         "psrad $" #shift ", %%mm0               \n\t"\
1123         "packssdw %%mm3, %%mm0                  \n\t" /* A1     a1 */\
1124         "movq %%mm0, 16+" #dst "                \n\t"\
1125         "movq %%mm0, 96+" #dst "                \n\t"\
1126         "movq %%mm4, 112+" #dst "               \n\t"\
1127         "psrad $" #shift ", %%mm5               \n\t"\
1128         "psrad $" #shift ", %%mm6               \n\t"\
1129         "psrad $" #shift ", %%mm2               \n\t"\
1130         "packssdw %%mm2, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
1131         "movq %%mm5, 32+" #dst "                \n\t"\
1132         "psrad $" #shift ", %%mm1               \n\t"\
1133         "packssdw %%mm1, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
1134         "movq %%mm6, 48+" #dst "                \n\t"\
1135         "movq %%mm6, 64+" #dst "                \n\t"\
1136         "movq %%mm5, 80+" #dst "                \n\t"
1137
1138
1139 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1140 IDCT(    0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1141 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1142 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1143 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1144         "jmp 9f                                 \n\t"
1145
1146
1147         "#.balign 16                            \n\t"\
1148         "1:                                     \n\t"
1149 #undef IDCT
1150 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1151         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
1152         "movq " #src4 ", %%mm1                  \n\t" /* R6     R2      r6      r2 */\
1153         "movq " #src1 ", %%mm2                  \n\t" /* R3     R1      r3      r1 */\
1154         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
1155         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1156         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
1157         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1158         "movq 32(%2), %%mm5                     \n\t" /* C6     C2      C6      C2 */\
1159         "pmaddwd %%mm1, %%mm5                   \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1160         "movq 40(%2), %%mm6                     \n\t" /* -C2    C6      -C2     C6 */\
1161         "pmaddwd %%mm6, %%mm1                   \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1162         #rounder ", %%mm4                       \n\t"\
1163         "movq %%mm4, %%mm6                      \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1164         "movq 48(%2), %%mm7                     \n\t" /* C3     C1      C3      C1 */\
1165         #rounder ", %%mm0                       \n\t"\
1166         "pmaddwd %%mm2, %%mm7                   \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1167         "paddd %%mm5, %%mm4                     \n\t" /* A0             a0 */\
1168         "psubd %%mm5, %%mm6                     \n\t" /* A3             a3 */\
1169         "movq %%mm0, %%mm5                      \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1170         "paddd %%mm1, %%mm0                     \n\t" /* A1             a1 */\
1171         "psubd %%mm1, %%mm5                     \n\t" /* A2             a2 */\
1172         "movq 64(%2), %%mm1                     \n\t"\
1173         "pmaddwd %%mm2, %%mm1                   \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1174         "paddd %%mm4, %%mm7                     \n\t" /* A0+B0          a0+b0 */\
1175         "paddd %%mm4, %%mm4                     \n\t" /* 2A0            2a0 */\
1176         "psubd %%mm7, %%mm4                     \n\t" /* A0-B0          a0-b0 */\
1177         "psrad $" #shift ", %%mm7               \n\t"\
1178         "psrad $" #shift ", %%mm4               \n\t"\
1179         "movq %%mm0, %%mm3                      \n\t" /* A1             a1 */\
1180         "paddd %%mm1, %%mm0                     \n\t" /* A1+B1          a1+b1 */\
1181         "psubd %%mm1, %%mm3                     \n\t" /* A1-B1          a1-b1 */\
1182         "psrad $" #shift ", %%mm0               \n\t"\
1183         "psrad $" #shift ", %%mm3               \n\t"\
1184         "packssdw %%mm7, %%mm7                  \n\t" /* A0+B0  a0+b0 */\
1185         "movd %%mm7, " #dst "                   \n\t"\
1186         "packssdw %%mm0, %%mm0                  \n\t" /* A1+B1  a1+b1 */\
1187         "movd %%mm0, 16+" #dst "                \n\t"\
1188         "packssdw %%mm3, %%mm3                  \n\t" /* A1-B1  a1-b1 */\
1189         "movd %%mm3, 96+" #dst "                \n\t"\
1190         "packssdw %%mm4, %%mm4                  \n\t" /* A0-B0  a0-b0 */\
1191         "movd %%mm4, 112+" #dst "               \n\t"\
1192         "movq 80(%2), %%mm4                     \n\t" /* -C1    C5      -C1     C5 */\
1193         "pmaddwd %%mm2, %%mm4                   \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1194         "pmaddwd 96(%2), %%mm2                  \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1195         "movq %%mm5, %%mm3                      \n\t" /* A2             a2 */\
1196         "paddd %%mm4, %%mm3                     \n\t" /* A2+B2          a2+b2 */\
1197         "psubd %%mm4, %%mm5                     \n\t" /* a2-B2          a2-b2 */\
1198         "psrad $" #shift ", %%mm3               \n\t"\
1199         "psrad $" #shift ", %%mm5               \n\t"\
1200         "movq %%mm6, %%mm4                      \n\t" /* A3             a3 */\
1201         "paddd %%mm2, %%mm6                     \n\t" /* A3+B3          a3+b3 */\
1202         "psubd %%mm2, %%mm4                     \n\t" /* a3-B3          a3-b3 */\
1203         "psrad $" #shift ", %%mm6               \n\t"\
1204         "packssdw %%mm3, %%mm3                  \n\t" /* A2+B2  a2+b2 */\
1205         "movd %%mm3, 32+" #dst "                \n\t"\
1206         "psrad $" #shift ", %%mm4               \n\t"\
1207         "packssdw %%mm6, %%mm6                  \n\t" /* A3+B3  a3+b3 */\
1208         "movd %%mm6, 48+" #dst "                \n\t"\
1209         "packssdw %%mm4, %%mm4                  \n\t" /* A3-B3  a3-b3 */\
1210         "packssdw %%mm5, %%mm5                  \n\t" /* A2-B2  a2-b2 */\
1211         "movd %%mm4, 64+" #dst "                \n\t"\
1212         "movd %%mm5, 80+" #dst "                \n\t"
1213
1214
1215 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1216 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1217 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1218 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1219 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1220         "jmp 9f                                 \n\t"
1221
1222
1223         "#.balign 16                            \n\t"
1224         "7:                                     \n\t"
1225 #undef IDCT
1226 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1227         "movq " #src0 ", %%mm0                  \n\t" /* R4     R0      r4      r0 */\
1228         "movq 16(%2), %%mm4                     \n\t" /* C4     C4      C4      C4 */\
1229         "pmaddwd %%mm0, %%mm4                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1230         "movq 24(%2), %%mm5                     \n\t" /* -C4    C4      -C4     C4 */\
1231         "pmaddwd %%mm5, %%mm0                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1232         #rounder ", %%mm4                       \n\t"\
1233         #rounder ", %%mm0                       \n\t"\
1234         "psrad $" #shift ", %%mm4               \n\t"\
1235         "psrad $" #shift ", %%mm0               \n\t"\
1236         "movq 8+" #src0 ", %%mm2                \n\t" /* R4     R0      r4      r0 */\
1237         "movq 16(%2), %%mm1                     \n\t" /* C4     C4      C4      C4 */\
1238         "pmaddwd %%mm2, %%mm1                   \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1239         "movq 24(%2), %%mm7                     \n\t" /* -C4    C4      -C4     C4 */\
1240         "pmaddwd %%mm7, %%mm2                   \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1241         "movq 32(%2), %%mm7                     \n\t" /* C6     C2      C6      C2 */\
1242         #rounder ", %%mm1                       \n\t"\
1243         #rounder ", %%mm2                       \n\t"\
1244         "psrad $" #shift ", %%mm1               \n\t"\
1245         "packssdw %%mm1, %%mm4                  \n\t" /* A0     a0 */\
1246         "movq %%mm4, " #dst "                   \n\t"\
1247         "psrad $" #shift ", %%mm2               \n\t"\
1248         "packssdw %%mm2, %%mm0                  \n\t" /* A1     a1 */\
1249         "movq %%mm0, 16+" #dst "                \n\t"\
1250         "movq %%mm0, 96+" #dst "                \n\t"\
1251         "movq %%mm4, 112+" #dst "               \n\t"\
1252         "movq %%mm0, 32+" #dst "                \n\t"\
1253         "movq %%mm4, 48+" #dst "                \n\t"\
1254         "movq %%mm4, 64+" #dst "                \n\t"\
1255         "movq %%mm0, 80+" #dst "                \n\t"
1256
1257 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1258 IDCT(   0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1259 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1260 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1261 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1262
1263
1264 #endif
1265
1266 /*
1267 Input
1268  00 40 04 44 20 60 24 64
1269  10 30 14 34 50 70 54 74
1270  01 41 03 43 21 61 23 63
1271  11 31 13 33 51 71 53 73
1272  02 42 06 46 22 62 26 66
1273  12 32 16 36 52 72 56 76
1274  05 45 07 47 25 65 27 67
1275  15 35 17 37 55 75 57 77
1276
1277 Temp
1278  00 04 10 14 20 24 30 34
1279  40 44 50 54 60 64 70 74
1280  01 03 11 13 21 23 31 33
1281  41 43 51 53 61 63 71 73
1282  02 06 12 16 22 26 32 36
1283  42 46 52 56 62 66 72 76
1284  05 07 15 17 25 27 35 37
1285  45 47 55 57 65 67 75 77
1286 */
1287
1288 "9: \n\t"
1289                 :: "r" (block), "r" (temp), "r" (coeffs)
1290                 : "%eax"
1291         );
1292 }
1293
1294 void ff_simple_idct_mmx(int16_t *block)
1295 {
1296     idct(block);
1297 }
1298
1299 //FIXME merge add/put into the idct
1300
1301 void ff_simple_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1302 {
1303     idct(block);
1304     put_pixels_clamped_mmx(block, dest, line_size);
1305 }
1306 void ff_simple_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1307 {
1308     idct(block);
1309     add_pixels_clamped_mmx(block, dest, line_size);
1310 }