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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20 #include "../dsputil.h"
  21 #include "../simple_idct.h"
  22
  23 /*
  24 23170.475006
  25 22725.260826
  26 21406.727617
  27 19265.545870
  28 16384.000000
  29 12872.826198
  30 8866.956905
  31 4520.335430
  32 */
  33 #define C0 23170 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  34 #define C1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  35 #define C2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  36 #define C3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  37 #if 0
  38 #define C4 16384 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  39 #else
  40 #define C4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) - 0.5
  41 #endif
  42 #define C5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  43 #define C6 8867  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  44 #define C7 4520  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  45
  46 #define ROW_SHIFT 11
  47 #define COL_SHIFT 20 // 6
  48
  49 static const uint64_t attribute_used __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000ULL;
  50 static const uint64_t attribute_used __attribute__((aligned(8))) d40000= 0x0000000000040000ULL;
  51
  52 static const int16_t __attribute__((aligned(8))) coeffs[]= {
  53         1<<(ROW_SHIFT-1), 0, 1<<(ROW_SHIFT-1), 0,
  54 //        1<<(COL_SHIFT-1), 0, 1<<(COL_SHIFT-1), 0,
  55 //        0, 1<<(COL_SHIFT-1-16), 0, 1<<(COL_SHIFT-1-16),
  56         1<<(ROW_SHIFT-1), 1, 1<<(ROW_SHIFT-1), 0,
  57         // the 1 = ((1<<(COL_SHIFT-1))/C4)<<ROW_SHIFT :)
  58 //        0, 0, 0, 0,
  59 //        0, 0, 0, 0,
  60
  61  C4,  C4,  C4,  C4,
  62  C4, -C4,  C4, -C4,
  63
  64  C2,  C6,  C2,  C6,
  65  C6, -C2,  C6, -C2,
  66
  67  C1,  C3,  C1,  C3,
  68  C5,  C7,  C5,  C7,
  69
  70  C3, -C7,  C3, -C7,
  71 -C1, -C5, -C1, -C5,
  72
  73  C5, -C1,  C5, -C1,
  74  C7,  C3,  C7,  C3,
  75
  76  C7, -C5,  C7, -C5,
  77  C3, -C1,  C3, -C1
  78 };
  79
  80 #if 0
  81 static void unused_var_killer(){
  82         int a= wm1010 + d40000;
  83         temp[0]=a;
  84 }
  85
  86 static void inline idctCol (int16_t * col, int16_t *input)
  87 {
  88 #undef C0
  89 #undef C1
  90 #undef C2
  91 #undef C3
  92 #undef C4
  93 #undef C5
  94 #undef C6
  95 #undef C7
  96         int a0, a1, a2, a3, b0, b1, b2, b3;
  97         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  98         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
  99         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 100         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 101         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 102         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 103         const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 104         const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 105 /*
 106         if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
 107                 col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
 108                         col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
 109                 return;
 110         }*/
 111
 112 col[8*0] = input[8*0 + 0];
 113 col[8*1] = input[8*2 + 0];
 114 col[8*2] = input[8*0 + 1];
 115 col[8*3] = input[8*2 + 1];
 116 col[8*4] = input[8*4 + 0];
 117 col[8*5] = input[8*6 + 0];
 118 col[8*6] = input[8*4 + 1];
 119 col[8*7] = input[8*6 + 1];
 120
 121         a0 = C4*col[8*0] + C2*col[8*2] + C4*col[8*4] + C6*col[8*6] + (1<<(COL_SHIFT-1));
 122         a1 = C4*col[8*0] + C6*col[8*2] - C4*col[8*4] - C2*col[8*6] + (1<<(COL_SHIFT-1));
 123         a2 = C4*col[8*0] - C6*col[8*2] - C4*col[8*4] + C2*col[8*6] + (1<<(COL_SHIFT-1));
 124         a3 = C4*col[8*0] - C2*col[8*2] + C4*col[8*4] - C6*col[8*6] + (1<<(COL_SHIFT-1));
 125
 126         b0 = C1*col[8*1] + C3*col[8*3] + C5*col[8*5] + C7*col[8*7];
 127         b1 = C3*col[8*1] - C7*col[8*3] - C1*col[8*5] - C5*col[8*7];
 128         b2 = C5*col[8*1] - C1*col[8*3] + C7*col[8*5] + C3*col[8*7];
 129         b3 = C7*col[8*1] - C5*col[8*3] + C3*col[8*5] - C1*col[8*7];
 130
 131         col[8*0] = (a0 + b0) >> COL_SHIFT;
 132         col[8*1] = (a1 + b1) >> COL_SHIFT;
 133         col[8*2] = (a2 + b2) >> COL_SHIFT;
 134         col[8*3] = (a3 + b3) >> COL_SHIFT;
 135         col[8*4] = (a3 - b3) >> COL_SHIFT;
 136         col[8*5] = (a2 - b2) >> COL_SHIFT;
 137         col[8*6] = (a1 - b1) >> COL_SHIFT;
 138         col[8*7] = (a0 - b0) >> COL_SHIFT;
 139 }
 140
 141 static void inline idctRow (int16_t * output, int16_t * input)
 142 {
 143         int16_t row[8];
 144
 145         int a0, a1, a2, a3, b0, b1, b2, b3;
 146         const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 147         const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 148         const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 149         const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 150         const int C4 = 16383; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 151         const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 152         const int C6 = 8867;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 153         const int C7 = 4520;  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
 154
 155 row[0] = input[0];
 156 row[2] = input[1];
 157 row[4] = input[4];
 158 row[6] = input[5];
 159 row[1] = input[8];
 160 row[3] = input[9];
 161 row[5] = input[12];
 162 row[7] = input[13];
 163
 164         if( !(row[1] | row[2] |row[3] |row[4] |row[5] |row[6] | row[7]) ) {
 165                 row[0] = row[1] = row[2] = row[3] = row[4] =
 166                         row[5] = row[6] = row[7] = row[0]<<3;
 167         output[0]  = row[0];
 168         output[2]  = row[1];
 169         output[4]  = row[2];
 170         output[6]  = row[3];
 171         output[8]  = row[4];
 172         output[10] = row[5];
 173         output[12] = row[6];
 174         output[14] = row[7];
 175                 return;
 176         }
 177
 178         a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + (1<<(ROW_SHIFT-1));
 179         a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + (1<<(ROW_SHIFT-1));
 180         a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + (1<<(ROW_SHIFT-1));
 181         a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + (1<<(ROW_SHIFT-1));
 182
 183         b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
 184         b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
 185         b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
 186         b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];
 187
 188         row[0] = (a0 + b0) >> ROW_SHIFT;
 189         row[1] = (a1 + b1) >> ROW_SHIFT;
 190         row[2] = (a2 + b2) >> ROW_SHIFT;
 191         row[3] = (a3 + b3) >> ROW_SHIFT;
 192         row[4] = (a3 - b3) >> ROW_SHIFT;
 193         row[5] = (a2 - b2) >> ROW_SHIFT;
 194         row[6] = (a1 - b1) >> ROW_SHIFT;
 195         row[7] = (a0 - b0) >> ROW_SHIFT;
 196
 197         output[0]  = row[0];
 198         output[2]  = row[1];
 199         output[4]  = row[2];
 200         output[6]  = row[3];
 201         output[8]  = row[4];
 202         output[10] = row[5];
 203         output[12] = row[6];
 204         output[14] = row[7];
 205 }
 206 #endif
 207
 208 static inline void idct(int16_t *block)
 209 {
 210         int64_t __attribute__((aligned(8))) align_tmp[16];
 211         int16_t * const temp= (int16_t*)align_tmp;
 212
 213         asm volatile(
 214 #if 0 //Alternative, simpler variant
 215
 216 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 217         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 218         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 219         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 220         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 221         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 222         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 223         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 224         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 225         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 226         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 227         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 228         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 229         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 230         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 231         #rounder ", %%mm4               \n\t"\
 232         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 233         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 234         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 235         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 236         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 237         #rounder ", %%mm0               \n\t"\
 238         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 239         "paddd %%mm0, %%mm0             \n\t" \
 240         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 241         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 242         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 243         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 244         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 245         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 246         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 247         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 248         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 249         "psrad $" #shift ", %%mm7       \n\t"\
 250         "psrad $" #shift ", %%mm4       \n\t"\
 251         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 252         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 253         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 254         "psrad $" #shift ", %%mm1       \n\t"\
 255         "psrad $" #shift ", %%mm2       \n\t"\
 256         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 257         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 258         "movq %%mm7, " #dst "           \n\t"\
 259         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 260         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 261         "movq %%mm2, 24+" #dst "        \n\t"\
 262         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 263         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 264         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 265         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 266         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 267         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 268         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 269         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 270         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 271         "psrad $" #shift ", %%mm2       \n\t"\
 272         "psrad $" #shift ", %%mm0       \n\t"\
 273         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 274         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 275         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 276         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 277         "psrad $" #shift ", %%mm6       \n\t"\
 278         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 279         "movq %%mm2, 8+" #dst "         \n\t"\
 280         "psrad $" #shift ", %%mm4       \n\t"\
 281         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 282         "movq %%mm4, 16+" #dst "        \n\t"\
 283
 284 #define COL_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 285         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 286         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 287         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 288         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 289         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 290         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 291         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 292         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 293         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 294         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 295         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 296         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 297         #rounder ", %%mm4               \n\t"\
 298         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 299         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 300         #rounder ", %%mm0               \n\t"\
 301         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 302         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 303         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 304         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 305         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 306         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 307         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 308         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 309         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 310         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 311         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 312         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 313         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 314         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 315         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 316         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 317         "psrad $" #shift ", %%mm7       \n\t"\
 318         "psrad $" #shift ", %%mm4       \n\t"\
 319         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 320         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 321         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 322         "psrad $" #shift ", %%mm0       \n\t"\
 323         "psrad $" #shift ", %%mm2       \n\t"\
 324         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 325         "movd %%mm7, " #dst "           \n\t"\
 326         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 327         "movd %%mm0, 16+" #dst "        \n\t"\
 328         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 329         "movd %%mm2, 96+" #dst "        \n\t"\
 330         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 331         "movd %%mm4, 112+" #dst "       \n\t"\
 332         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 333         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 334         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 335         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 336         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 337         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 338         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 339         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 340         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 341         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 342         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 343         "psrad $" #shift ", %%mm2       \n\t"\
 344         "psrad $" #shift ", %%mm5       \n\t"\
 345         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 346         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 347         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 348         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 349         "psrad $" #shift ", %%mm6       \n\t"\
 350         "psrad $" #shift ", %%mm4       \n\t"\
 351         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 352         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 353         "movd %%mm2, 32+" #dst "        \n\t"\
 354         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
 355         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 356         "movd %%mm6, 48+" #dst "        \n\t"\
 357         "movd %%mm4, 64+" #dst "        \n\t"\
 358         "movd %%mm5, 80+" #dst "        \n\t"\
 359
 360
 361 #define DC_COND_ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 362         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 363         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 364         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 365         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 366         "movq "MANGLE(wm1010)", %%mm4   \n\t"\
 367         "pand %%mm0, %%mm4              \n\t"\
 368         "por %%mm1, %%mm4               \n\t"\
 369         "por %%mm2, %%mm4               \n\t"\
 370         "por %%mm3, %%mm4               \n\t"\
 371         "packssdw %%mm4,%%mm4           \n\t"\
 372         "movd %%mm4, %%eax              \n\t"\
 373         "orl %%eax, %%eax               \n\t"\
 374         "jz 1f                          \n\t"\
 375         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 376         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 377         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 378         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 379         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 380         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 381         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 382         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 383         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 384         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 385         #rounder ", %%mm4               \n\t"\
 386         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 387         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 388         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 389         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 390         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 391         #rounder ", %%mm0               \n\t"\
 392         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 393         "paddd %%mm0, %%mm0             \n\t" \
 394         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 395         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 396         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 397         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 398         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 399         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 400         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 401         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 402         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 403         "psrad $" #shift ", %%mm7       \n\t"\
 404         "psrad $" #shift ", %%mm4       \n\t"\
 405         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 406         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 407         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 408         "psrad $" #shift ", %%mm1       \n\t"\
 409         "psrad $" #shift ", %%mm2       \n\t"\
 410         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 411         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 412         "movq %%mm7, " #dst "           \n\t"\
 413         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 414         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 415         "movq %%mm2, 24+" #dst "        \n\t"\
 416         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 417         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 418         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 419         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 420         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 421         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 422         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 423         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 424         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 425         "psrad $" #shift ", %%mm2       \n\t"\
 426         "psrad $" #shift ", %%mm0       \n\t"\
 427         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 428         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 429         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 430         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 431         "psrad $" #shift ", %%mm6       \n\t"\
 432         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 433         "movq %%mm2, 8+" #dst "         \n\t"\
 434         "psrad $" #shift ", %%mm4       \n\t"\
 435         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 436         "movq %%mm4, 16+" #dst "        \n\t"\
 437         "jmp 2f                         \n\t"\
 438         "1:                             \n\t"\
 439         "pslld $16, %%mm0               \n\t"\
 440         "#paddd "MANGLE(d40000)", %%mm0 \n\t"\
 441         "psrad $13, %%mm0               \n\t"\
 442         "packssdw %%mm0, %%mm0          \n\t"\
 443         "movq %%mm0, " #dst "           \n\t"\
 444         "movq %%mm0, 8+" #dst "         \n\t"\
 445         "movq %%mm0, 16+" #dst "        \n\t"\
 446         "movq %%mm0, 24+" #dst "        \n\t"\
 447         "2:                             \n\t"
 448
 449
 450 //IDCT(      src0,   src4,   src1,   src5,    dst,    rounder, shift)
 451 ROW_IDCT(    (%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 452 /*ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1), paddd (%2), 11)
 453 ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1), paddd (%2), 11)
 454 ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1), paddd (%2), 11)*/
 455
 456 DC_COND_ROW_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11)
 457 DC_COND_ROW_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11)
 458 DC_COND_ROW_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11)
 459
 460
 461 //IDCT(      src0,   src4,   src1,    src5,    dst, rounder, shift)
 462 COL_IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 463 COL_IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 464 COL_IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 465 COL_IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 466
 467 #else
 468
 469 #define DC_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 470         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 471         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 472         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 473         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 474         "movq "MANGLE(wm1010)", %%mm4   \n\t"\
 475         "pand %%mm0, %%mm4              \n\t"\
 476         "por %%mm1, %%mm4               \n\t"\
 477         "por %%mm2, %%mm4               \n\t"\
 478         "por %%mm3, %%mm4               \n\t"\
 479         "packssdw %%mm4,%%mm4           \n\t"\
 480         "movd %%mm4, %%eax              \n\t"\
 481         "orl %%eax, %%eax               \n\t"\
 482         "jz 1f                          \n\t"\
 483         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 484         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 485         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 486         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 487         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 488         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 489         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 490         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 491         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 492         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 493         #rounder ", %%mm4               \n\t"\
 494         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 495         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 496         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 497         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 498         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 499         #rounder ", %%mm0               \n\t"\
 500         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 501         "paddd %%mm0, %%mm0             \n\t" \
 502         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 503         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 504         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 505         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 506         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 507         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 508         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 509         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 510         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 511         "psrad $" #shift ", %%mm7       \n\t"\
 512         "psrad $" #shift ", %%mm4       \n\t"\
 513         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 514         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 515         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 516         "psrad $" #shift ", %%mm1       \n\t"\
 517         "psrad $" #shift ", %%mm2       \n\t"\
 518         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 519         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 520         "movq %%mm7, " #dst "           \n\t"\
 521         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 522         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 523         "movq %%mm2, 24+" #dst "        \n\t"\
 524         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 525         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 526         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 527         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 528         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 529         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 530         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 531         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 532         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 533         "psrad $" #shift ", %%mm2       \n\t"\
 534         "psrad $" #shift ", %%mm0       \n\t"\
 535         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 536         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 537         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 538         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 539         "psrad $" #shift ", %%mm6       \n\t"\
 540         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 541         "movq %%mm2, 8+" #dst "         \n\t"\
 542         "psrad $" #shift ", %%mm4       \n\t"\
 543         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 544         "movq %%mm4, 16+" #dst "        \n\t"\
 545         "jmp 2f                         \n\t"\
 546         "1:                             \n\t"\
 547         "pslld $16, %%mm0               \n\t"\
 548         "paddd "MANGLE(d40000)", %%mm0  \n\t"\
 549         "psrad $13, %%mm0               \n\t"\
 550         "packssdw %%mm0, %%mm0          \n\t"\
 551         "movq %%mm0, " #dst "           \n\t"\
 552         "movq %%mm0, 8+" #dst "         \n\t"\
 553         "movq %%mm0, 16+" #dst "        \n\t"\
 554         "movq %%mm0, 24+" #dst "        \n\t"\
 555         "2:                             \n\t"
 556
 557 #define Z_COND_IDCT(src0, src4, src1, src5, dst, rounder, shift, bt) \
 558         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 559         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 560         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 561         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 562         "movq %%mm0, %%mm4              \n\t"\
 563         "por %%mm1, %%mm4               \n\t"\
 564         "por %%mm2, %%mm4               \n\t"\
 565         "por %%mm3, %%mm4               \n\t"\
 566         "packssdw %%mm4,%%mm4           \n\t"\
 567         "movd %%mm4, %%eax              \n\t"\
 568         "orl %%eax, %%eax               \n\t"\
 569         "jz " #bt "                     \n\t"\
 570         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 571         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 572         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 573         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 574         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 575         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 576         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 577         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 578         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 579         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 580         #rounder ", %%mm4               \n\t"\
 581         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 582         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 583         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 584         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 585         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 586         #rounder ", %%mm0               \n\t"\
 587         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 588         "paddd %%mm0, %%mm0             \n\t" \
 589         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 590         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 591         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 592         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 593         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 594         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 595         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 596         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 597         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 598         "psrad $" #shift ", %%mm7       \n\t"\
 599         "psrad $" #shift ", %%mm4       \n\t"\
 600         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 601         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 602         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 603         "psrad $" #shift ", %%mm1       \n\t"\
 604         "psrad $" #shift ", %%mm2       \n\t"\
 605         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 606         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 607         "movq %%mm7, " #dst "           \n\t"\
 608         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 609         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 610         "movq %%mm2, 24+" #dst "        \n\t"\
 611         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 612         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 613         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 614         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 615         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 616         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 617         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 618         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 619         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 620         "psrad $" #shift ", %%mm2       \n\t"\
 621         "psrad $" #shift ", %%mm0       \n\t"\
 622         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 623         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 624         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 625         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 626         "psrad $" #shift ", %%mm6       \n\t"\
 627         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 628         "movq %%mm2, 8+" #dst "         \n\t"\
 629         "psrad $" #shift ", %%mm4       \n\t"\
 630         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 631         "movq %%mm4, 16+" #dst "        \n\t"\
 632
 633 #define ROW_IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 634         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 635         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 636         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 637         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 638         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 639         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 640         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 641         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 642         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 643         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 644         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 645         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 646         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 647         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 648         #rounder ", %%mm4               \n\t"\
 649         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 650         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 651         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 652         "movq 56(%2), %%mm5             \n\t" /* C7     C5      C7      C5 */\
 653         "pmaddwd %%mm3, %%mm5           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 654         #rounder ", %%mm0               \n\t"\
 655         "paddd %%mm0, %%mm1             \n\t" /* A1             a1 */\
 656         "paddd %%mm0, %%mm0             \n\t" \
 657         "psubd %%mm1, %%mm0             \n\t" /* A2             a2 */\
 658         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 659         "paddd %%mm5, %%mm7             \n\t" /* B0             b0 */\
 660         "movq 72(%2), %%mm5             \n\t" /* -C5    -C1     -C5     -C1 */\
 661         "pmaddwd %%mm3, %%mm5           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 662         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 663         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 664         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 665         "paddd %%mm2, %%mm5             \n\t" /* B1             b1 */\
 666         "psrad $" #shift ", %%mm7       \n\t"\
 667         "psrad $" #shift ", %%mm4       \n\t"\
 668         "movq %%mm1, %%mm2              \n\t" /* A1             a1 */\
 669         "paddd %%mm5, %%mm1             \n\t" /* A1+B1          a1+b1 */\
 670         "psubd %%mm5, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 671         "psrad $" #shift ", %%mm1       \n\t"\
 672         "psrad $" #shift ", %%mm2       \n\t"\
 673         "packssdw %%mm1, %%mm7          \n\t" /* A1+B1  a1+b1   A0+B0   a0+b0 */\
 674         "packssdw %%mm4, %%mm2          \n\t" /* A0-B0  a0-b0   A1-B1   a1-b1 */\
 675         "movq %%mm7, " #dst "           \n\t"\
 676         "movq " #src1 ", %%mm1          \n\t" /* R3     R1      r3      r1 */\
 677         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 678         "movq %%mm2, 24+" #dst "        \n\t"\
 679         "pmaddwd %%mm1, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 680         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 681         "pmaddwd 96(%2), %%mm1          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 682         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 683         "movq %%mm0, %%mm2              \n\t" /* A2             a2 */\
 684         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 685         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 686         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 687         "psubd %%mm4, %%mm0             \n\t" /* a2-B2          a2-b2 */\
 688         "psrad $" #shift ", %%mm2       \n\t"\
 689         "psrad $" #shift ", %%mm0       \n\t"\
 690         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 691         "paddd %%mm1, %%mm3             \n\t" /* B3             b3 */\
 692         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 693         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 694         "psrad $" #shift ", %%mm6       \n\t"\
 695         "packssdw %%mm6, %%mm2          \n\t" /* A3+B3  a3+b3   A2+B2   a2+b2 */\
 696         "movq %%mm2, 8+" #dst "         \n\t"\
 697         "psrad $" #shift ", %%mm4       \n\t"\
 698         "packssdw %%mm0, %%mm4          \n\t" /* A2-B2  a2-b2   A3-B3   a3-b3 */\
 699         "movq %%mm4, 16+" #dst "        \n\t"\
 700
 701 //IDCT(         src0,   src4,   src1,   src5,    dst,   rounder, shift)
 702 DC_COND_IDCT(  0(%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
 703 Z_COND_IDCT(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11, 4f)
 704 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 2f)
 705 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 1f)
 706
 707 #undef IDCT
 708 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 709         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 710         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 711         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 712         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 713         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 714         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 715         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 716         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 717         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 718         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 719         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 720         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 721         #rounder ", %%mm4               \n\t"\
 722         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 723         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 724         #rounder ", %%mm0               \n\t"\
 725         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 726         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 727         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 728         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 729         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 730         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 731         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 732         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 733         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 734         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 735         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 736         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 737         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 738         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 739         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 740         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 741         "psrad $" #shift ", %%mm7       \n\t"\
 742         "psrad $" #shift ", %%mm4       \n\t"\
 743         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 744         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 745         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 746         "psrad $" #shift ", %%mm0       \n\t"\
 747         "psrad $" #shift ", %%mm2       \n\t"\
 748         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 749         "movd %%mm7, " #dst "           \n\t"\
 750         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 751         "movd %%mm0, 16+" #dst "        \n\t"\
 752         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 753         "movd %%mm2, 96+" #dst "        \n\t"\
 754         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 755         "movd %%mm4, 112+" #dst "       \n\t"\
 756         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 757         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 758         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 759         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 760         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 761         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 762         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 763         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 764         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 765         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 766         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 767         "psrad $" #shift ", %%mm2       \n\t"\
 768         "psrad $" #shift ", %%mm5       \n\t"\
 769         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 770         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 771         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 772         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 773         "psrad $" #shift ", %%mm6       \n\t"\
 774         "psrad $" #shift ", %%mm4       \n\t"\
 775         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 776         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 777         "movd %%mm2, 32+" #dst "        \n\t"\
 778         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
 779         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 780         "movd %%mm6, 48+" #dst "        \n\t"\
 781         "movd %%mm4, 64+" #dst "        \n\t"\
 782         "movd %%mm5, 80+" #dst "        \n\t"
 783
 784
 785 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 786 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 787 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 788 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 789 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 790         "jmp 9f                         \n\t"
 791
 792         "#.balign 16                    \n\t"\
 793         "4:                             \n\t"
 794 Z_COND_IDCT(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 6f)
 795 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 5f)
 796
 797 #undef IDCT
 798 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 799         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 800         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
 801         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 802         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 803         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 804         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 805         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 806         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
 807         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
 808         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
 809         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
 810         #rounder ", %%mm4               \n\t"\
 811         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 812         #rounder ", %%mm0               \n\t"\
 813         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
 814         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
 815         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 816         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
 817         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
 818         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 819         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 820         "movq 72(%2), %%mm7             \n\t" /* -C5    -C1     -C5     -C1 */\
 821         "pmaddwd %%mm3, %%mm7           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 822         "paddd %%mm4, %%mm1             \n\t" /* A0+B0          a0+b0 */\
 823         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 824         "psubd %%mm1, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 825         "psrad $" #shift ", %%mm1       \n\t"\
 826         "psrad $" #shift ", %%mm4       \n\t"\
 827         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 828         "paddd %%mm7, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 829         "psubd %%mm7, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 830         "psrad $" #shift ", %%mm0       \n\t"\
 831         "psrad $" #shift ", %%mm2       \n\t"\
 832         "packssdw %%mm1, %%mm1          \n\t" /* A0+B0  a0+b0 */\
 833         "movd %%mm1, " #dst "           \n\t"\
 834         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 835         "movd %%mm0, 16+" #dst "        \n\t"\
 836         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 837         "movd %%mm2, 96+" #dst "        \n\t"\
 838         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 839         "movd %%mm4, 112+" #dst "       \n\t"\
 840         "movq 88(%2), %%mm1             \n\t" /* C3     C7      C3      C7 */\
 841         "pmaddwd %%mm3, %%mm1           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 842         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 843         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 844         "paddd %%mm1, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 845         "psubd %%mm1, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 846         "psrad $" #shift ", %%mm2       \n\t"\
 847         "psrad $" #shift ", %%mm5       \n\t"\
 848         "movq %%mm6, %%mm1              \n\t" /* A3             a3 */\
 849         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 850         "psubd %%mm3, %%mm1             \n\t" /* a3-B3          a3-b3 */\
 851         "psrad $" #shift ", %%mm6       \n\t"\
 852         "psrad $" #shift ", %%mm1       \n\t"\
 853         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 854         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 855         "movd %%mm2, 32+" #dst "        \n\t"\
 856         "packssdw %%mm1, %%mm1          \n\t" /* A3-B3  a3-b3 */\
 857         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 858         "movd %%mm6, 48+" #dst "        \n\t"\
 859         "movd %%mm1, 64+" #dst "        \n\t"\
 860         "movd %%mm5, 80+" #dst "        \n\t"
 861
 862 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 863 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 864 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 865 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 866 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 867         "jmp 9f                         \n\t"
 868
 869         "#.balign 16                    \n\t"\
 870         "6:                             \n\t"
 871 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 7f)
 872
 873 #undef IDCT
 874 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 875         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 876         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 877         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 878         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 879         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 880         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 881         #rounder ", %%mm4               \n\t"\
 882         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 883         #rounder ", %%mm0               \n\t"\
 884         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 885         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 886         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 887         "movq 72(%2), %%mm7             \n\t" /* -C5    -C1     -C5     -C1 */\
 888         "pmaddwd %%mm3, %%mm7           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 889         "paddd %%mm4, %%mm1             \n\t" /* A0+B0          a0+b0 */\
 890         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 891         "psubd %%mm1, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 892         "psrad $" #shift ", %%mm1       \n\t"\
 893         "psrad $" #shift ", %%mm4       \n\t"\
 894         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 895         "paddd %%mm7, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 896         "psubd %%mm7, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 897         "psrad $" #shift ", %%mm0       \n\t"\
 898         "psrad $" #shift ", %%mm2       \n\t"\
 899         "packssdw %%mm1, %%mm1          \n\t" /* A0+B0  a0+b0 */\
 900         "movd %%mm1, " #dst "           \n\t"\
 901         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 902         "movd %%mm0, 16+" #dst "        \n\t"\
 903         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 904         "movd %%mm2, 96+" #dst "        \n\t"\
 905         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 906         "movd %%mm4, 112+" #dst "       \n\t"\
 907         "movq 88(%2), %%mm1             \n\t" /* C3     C7      C3      C7 */\
 908         "pmaddwd %%mm3, %%mm1           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 909         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 910         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 911         "paddd %%mm1, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 912         "psubd %%mm1, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 913         "psrad $" #shift ", %%mm2       \n\t"\
 914         "psrad $" #shift ", %%mm5       \n\t"\
 915         "movq %%mm6, %%mm1              \n\t" /* A3             a3 */\
 916         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 917         "psubd %%mm3, %%mm1             \n\t" /* a3-B3          a3-b3 */\
 918         "psrad $" #shift ", %%mm6       \n\t"\
 919         "psrad $" #shift ", %%mm1       \n\t"\
 920         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
 921         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
 922         "movd %%mm2, 32+" #dst "        \n\t"\
 923         "packssdw %%mm1, %%mm1          \n\t" /* A3-B3  a3-b3 */\
 924         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
 925         "movd %%mm6, 48+" #dst "        \n\t"\
 926         "movd %%mm1, 64+" #dst "        \n\t"\
 927         "movd %%mm5, 80+" #dst "        \n\t"
 928
 929
 930 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
 931 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
 932 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
 933 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
 934 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
 935         "jmp 9f                         \n\t"
 936
 937         "#.balign 16                    \n\t"\
 938         "2:                             \n\t"
 939 Z_COND_IDCT(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 3f)
 940
 941 #undef IDCT
 942 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
 943         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
 944         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
 945         "movq " #src5 ", %%mm3          \n\t" /* R7     R5      r7      r5 */\
 946         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
 947         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 948         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
 949         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 950         #rounder ", %%mm4               \n\t"\
 951         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
 952         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
 953         #rounder ", %%mm0               \n\t"\
 954         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
 955         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
 956         "movq 56(%2), %%mm1             \n\t" /* C7     C5      C7      C5 */\
 957         "pmaddwd %%mm3, %%mm1           \n\t" /* C7R7+C5R5      C7r7+C5r5 */\
 958         "pmaddwd 64(%2), %%mm2          \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
 959         "paddd %%mm1, %%mm7             \n\t" /* B0             b0 */\
 960         "movq 72(%2), %%mm1             \n\t" /* -C5    -C1     -C5     -C1 */\
 961         "pmaddwd %%mm3, %%mm1           \n\t" /* -C5R7-C1R5     -C5r7-C1r5 */\
 962         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
 963         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
 964         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
 965         "paddd %%mm2, %%mm1             \n\t" /* B1             b1 */\
 966         "psrad $" #shift ", %%mm7       \n\t"\
 967         "psrad $" #shift ", %%mm4       \n\t"\
 968         "movq %%mm0, %%mm2              \n\t" /* A1             a1 */\
 969         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
 970         "psubd %%mm1, %%mm2             \n\t" /* A1-B1          a1-b1 */\
 971         "psrad $" #shift ", %%mm0       \n\t"\
 972         "psrad $" #shift ", %%mm2       \n\t"\
 973         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
 974         "movd %%mm7, " #dst "           \n\t"\
 975         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
 976         "movd %%mm0, 16+" #dst "        \n\t"\
 977         "packssdw %%mm2, %%mm2          \n\t" /* A1-B1  a1-b1 */\
 978         "movd %%mm2, 96+" #dst "        \n\t"\
 979         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
 980         "movd %%mm4, 112+" #dst "       \n\t"\
 981         "movq " #src1 ", %%mm0          \n\t" /* R3     R1      r3      r1 */\
 982         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
 983         "pmaddwd %%mm0, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
 984         "movq 88(%2), %%mm7             \n\t" /* C3     C7      C3      C7 */\
 985         "pmaddwd 96(%2), %%mm0          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
 986         "pmaddwd %%mm3, %%mm7           \n\t" /* C3R7+C7R5      C3r7+C7r5 */\
 987         "movq %%mm5, %%mm2              \n\t" /* A2             a2 */\
 988         "pmaddwd 104(%2), %%mm3         \n\t" /* -C1R7+C3R5     -C1r7+C3r5 */\
 989         "paddd %%mm7, %%mm4             \n\t" /* B2             b2 */\
 990         "paddd %%mm4, %%mm2             \n\t" /* A2+B2          a2+b2 */\
 991         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
 992         "psrad $" #shift ", %%mm2       \n\t"\
 993         "psrad $" #shift ", %%mm5       \n\t"\
 994         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
 995         "paddd %%mm0, %%mm3             \n\t" /* B3             b3 */\
 996         "paddd %%mm3, %%mm6             \n\t" /* A3+B3          a3+b3 */\
 997         "psubd %%mm3, %%mm4             \n\t" /* a3-B3          a3-b3 */\
 998         "psrad $" #shift ", %%mm6       \n\t"\
 999         "psrad $" #shift ", %%mm4       \n\t"\
1000         "packssdw %%mm2, %%mm2          \n\t" /* A2+B2  a2+b2 */\
1001         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1002         "movd %%mm2, 32+" #dst "        \n\t"\
1003         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
1004         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1005         "movd %%mm6, 48+" #dst "        \n\t"\
1006         "movd %%mm4, 64+" #dst "        \n\t"\
1007         "movd %%mm5, 80+" #dst "        \n\t"
1008
1009 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1010 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1011 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1012 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1013 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1014         "jmp 9f                         \n\t"
1015
1016         "#.balign 16                    \n\t"\
1017         "3:                             \n\t"
1018 #undef IDCT
1019 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1020         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1021         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
1022         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1023         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1024         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1025         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1026         #rounder ", %%mm4               \n\t"\
1027         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1028         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
1029         #rounder ", %%mm0               \n\t"\
1030         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1031         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1032         "movq 64(%2), %%mm3             \n\t"\
1033         "pmaddwd %%mm2, %%mm3           \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1034         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
1035         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
1036         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
1037         "psrad $" #shift ", %%mm7       \n\t"\
1038         "psrad $" #shift ", %%mm4       \n\t"\
1039         "movq %%mm0, %%mm1              \n\t" /* A1             a1 */\
1040         "paddd %%mm3, %%mm0             \n\t" /* A1+B1          a1+b1 */\
1041         "psubd %%mm3, %%mm1             \n\t" /* A1-B1          a1-b1 */\
1042         "psrad $" #shift ", %%mm0       \n\t"\
1043         "psrad $" #shift ", %%mm1       \n\t"\
1044         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
1045         "movd %%mm7, " #dst "           \n\t"\
1046         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
1047         "movd %%mm0, 16+" #dst "        \n\t"\
1048         "packssdw %%mm1, %%mm1          \n\t" /* A1-B1  a1-b1 */\
1049         "movd %%mm1, 96+" #dst "        \n\t"\
1050         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
1051         "movd %%mm4, 112+" #dst "       \n\t"\
1052         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
1053         "pmaddwd %%mm2, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1054         "pmaddwd 96(%2), %%mm2          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1055         "movq %%mm5, %%mm1              \n\t" /* A2             a2 */\
1056         "paddd %%mm4, %%mm1             \n\t" /* A2+B2          a2+b2 */\
1057         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
1058         "psrad $" #shift ", %%mm1       \n\t"\
1059         "psrad $" #shift ", %%mm5       \n\t"\
1060         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
1061         "paddd %%mm2, %%mm6             \n\t" /* A3+B3          a3+b3 */\
1062         "psubd %%mm2, %%mm4             \n\t" /* a3-B3          a3-b3 */\
1063         "psrad $" #shift ", %%mm6       \n\t"\
1064         "psrad $" #shift ", %%mm4       \n\t"\
1065         "packssdw %%mm1, %%mm1          \n\t" /* A2+B2  a2+b2 */\
1066         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1067         "movd %%mm1, 32+" #dst "        \n\t"\
1068         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
1069         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1070         "movd %%mm6, 48+" #dst "        \n\t"\
1071         "movd %%mm4, 64+" #dst "        \n\t"\
1072         "movd %%mm5, 80+" #dst "        \n\t"
1073
1074
1075 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1076 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1077 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1078 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1079 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1080         "jmp 9f                         \n\t"
1081
1082         "#.balign 16                    \n\t"\
1083         "5:                             \n\t"
1084 #undef IDCT
1085 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1086         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1087         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
1088         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1089         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1090         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1091         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1092         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
1093         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1094         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
1095         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1096         #rounder ", %%mm4               \n\t"\
1097         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1098         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
1099         #rounder ", %%mm0               \n\t"\
1100         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
1101         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1102         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
1103         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
1104         "movq 8+" #src0 ", %%mm2        \n\t" /* R4     R0      r4      r0 */\
1105         "movq 8+" #src4 ", %%mm3        \n\t" /* R6     R2      r6      r2 */\
1106         "movq 16(%2), %%mm1             \n\t" /* C4     C4      C4      C4 */\
1107         "pmaddwd %%mm2, %%mm1           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1108         "movq 24(%2), %%mm7             \n\t" /* -C4    C4      -C4     C4 */\
1109         "pmaddwd %%mm7, %%mm2           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1110         "movq 32(%2), %%mm7             \n\t" /* C6     C2      C6      C2 */\
1111         "pmaddwd %%mm3, %%mm7           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1112         "pmaddwd 40(%2), %%mm3          \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1113         #rounder ", %%mm1               \n\t"\
1114         "paddd %%mm1, %%mm7             \n\t" /* A0             a0 */\
1115         "paddd %%mm1, %%mm1             \n\t" /* 2C0            2c0 */\
1116         #rounder ", %%mm2               \n\t"\
1117         "psubd %%mm7, %%mm1             \n\t" /* A3             a3 */\
1118         "paddd %%mm2, %%mm3             \n\t" /* A1             a1 */\
1119         "paddd %%mm2, %%mm2             \n\t" /* 2C1            2c1 */\
1120         "psubd %%mm3, %%mm2             \n\t" /* A2             a2 */\
1121         "psrad $" #shift ", %%mm4       \n\t"\
1122         "psrad $" #shift ", %%mm7       \n\t"\
1123         "psrad $" #shift ", %%mm3       \n\t"\
1124         "packssdw %%mm7, %%mm4          \n\t" /* A0     a0 */\
1125         "movq %%mm4, " #dst "           \n\t"\
1126         "psrad $" #shift ", %%mm0       \n\t"\
1127         "packssdw %%mm3, %%mm0          \n\t" /* A1     a1 */\
1128         "movq %%mm0, 16+" #dst "        \n\t"\
1129         "movq %%mm0, 96+" #dst "        \n\t"\
1130         "movq %%mm4, 112+" #dst "       \n\t"\
1131         "psrad $" #shift ", %%mm5       \n\t"\
1132         "psrad $" #shift ", %%mm6       \n\t"\
1133         "psrad $" #shift ", %%mm2       \n\t"\
1134         "packssdw %%mm2, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1135         "movq %%mm5, 32+" #dst "        \n\t"\
1136         "psrad $" #shift ", %%mm1       \n\t"\
1137         "packssdw %%mm1, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1138         "movq %%mm6, 48+" #dst "        \n\t"\
1139         "movq %%mm6, 64+" #dst "        \n\t"\
1140         "movq %%mm5, 80+" #dst "        \n\t"
1141
1142
1143 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1144 IDCT(    0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1145 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1146 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1147 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1148         "jmp 9f                         \n\t"
1149
1150
1151         "#.balign 16                    \n\t"\
1152         "1:                             \n\t"
1153 #undef IDCT
1154 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1155         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1156         "movq " #src4 ", %%mm1          \n\t" /* R6     R2      r6      r2 */\
1157         "movq " #src1 ", %%mm2          \n\t" /* R3     R1      r3      r1 */\
1158         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1159         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1160         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1161         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1162         "movq 32(%2), %%mm5             \n\t" /* C6     C2      C6      C2 */\
1163         "pmaddwd %%mm1, %%mm5           \n\t" /* C6R6+C2R2      C6r6+C2r2 */\
1164         "movq 40(%2), %%mm6             \n\t" /* -C2    C6      -C2     C6 */\
1165         "pmaddwd %%mm6, %%mm1           \n\t" /* -C2R6+C6R2     -C2r6+C6r2 */\
1166         #rounder ", %%mm4               \n\t"\
1167         "movq %%mm4, %%mm6              \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1168         "movq 48(%2), %%mm7             \n\t" /* C3     C1      C3      C1 */\
1169         #rounder ", %%mm0               \n\t"\
1170         "pmaddwd %%mm2, %%mm7           \n\t" /* C3R3+C1R1      C3r3+C1r1 */\
1171         "paddd %%mm5, %%mm4             \n\t" /* A0             a0 */\
1172         "psubd %%mm5, %%mm6             \n\t" /* A3             a3 */\
1173         "movq %%mm0, %%mm5              \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1174         "paddd %%mm1, %%mm0             \n\t" /* A1             a1 */\
1175         "psubd %%mm1, %%mm5             \n\t" /* A2             a2 */\
1176         "movq 64(%2), %%mm1             \n\t"\
1177         "pmaddwd %%mm2, %%mm1           \n\t" /* -C7R3+C3R1     -C7r3+C3r1 */\
1178         "paddd %%mm4, %%mm7             \n\t" /* A0+B0          a0+b0 */\
1179         "paddd %%mm4, %%mm4             \n\t" /* 2A0            2a0 */\
1180         "psubd %%mm7, %%mm4             \n\t" /* A0-B0          a0-b0 */\
1181         "psrad $" #shift ", %%mm7       \n\t"\
1182         "psrad $" #shift ", %%mm4       \n\t"\
1183         "movq %%mm0, %%mm3              \n\t" /* A1             a1 */\
1184         "paddd %%mm1, %%mm0             \n\t" /* A1+B1          a1+b1 */\
1185         "psubd %%mm1, %%mm3             \n\t" /* A1-B1          a1-b1 */\
1186         "psrad $" #shift ", %%mm0       \n\t"\
1187         "psrad $" #shift ", %%mm3       \n\t"\
1188         "packssdw %%mm7, %%mm7          \n\t" /* A0+B0  a0+b0 */\
1189         "movd %%mm7, " #dst "           \n\t"\
1190         "packssdw %%mm0, %%mm0          \n\t" /* A1+B1  a1+b1 */\
1191         "movd %%mm0, 16+" #dst "        \n\t"\
1192         "packssdw %%mm3, %%mm3          \n\t" /* A1-B1  a1-b1 */\
1193         "movd %%mm3, 96+" #dst "        \n\t"\
1194         "packssdw %%mm4, %%mm4          \n\t" /* A0-B0  a0-b0 */\
1195         "movd %%mm4, 112+" #dst "       \n\t"\
1196         "movq 80(%2), %%mm4             \n\t" /* -C1    C5      -C1     C5 */\
1197         "pmaddwd %%mm2, %%mm4           \n\t" /* -C1R3+C5R1     -C1r3+C5r1 */\
1198         "pmaddwd 96(%2), %%mm2          \n\t" /* -C5R3+C7R1     -C5r3+C7r1 */\
1199         "movq %%mm5, %%mm3              \n\t" /* A2             a2 */\
1200         "paddd %%mm4, %%mm3             \n\t" /* A2+B2          a2+b2 */\
1201         "psubd %%mm4, %%mm5             \n\t" /* a2-B2          a2-b2 */\
1202         "psrad $" #shift ", %%mm3       \n\t"\
1203         "psrad $" #shift ", %%mm5       \n\t"\
1204         "movq %%mm6, %%mm4              \n\t" /* A3             a3 */\
1205         "paddd %%mm2, %%mm6             \n\t" /* A3+B3          a3+b3 */\
1206         "psubd %%mm2, %%mm4             \n\t" /* a3-B3          a3-b3 */\
1207         "psrad $" #shift ", %%mm6       \n\t"\
1208         "packssdw %%mm3, %%mm3          \n\t" /* A2+B2  a2+b2 */\
1209         "movd %%mm3, 32+" #dst "        \n\t"\
1210         "psrad $" #shift ", %%mm4       \n\t"\
1211         "packssdw %%mm6, %%mm6          \n\t" /* A3+B3  a3+b3 */\
1212         "movd %%mm6, 48+" #dst "        \n\t"\
1213         "packssdw %%mm4, %%mm4          \n\t" /* A3-B3  a3-b3 */\
1214         "packssdw %%mm5, %%mm5          \n\t" /* A2-B2  a2-b2 */\
1215         "movd %%mm4, 64+" #dst "        \n\t"\
1216         "movd %%mm5, 80+" #dst "        \n\t"
1217
1218
1219 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1220 IDCT(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1221 IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1222 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1223 IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1224         "jmp 9f                         \n\t"
1225
1226
1227         "#.balign 16                    \n\t"
1228         "7:                             \n\t"
1229 #undef IDCT
1230 #define IDCT(src0, src4, src1, src5, dst, rounder, shift) \
1231         "movq " #src0 ", %%mm0          \n\t" /* R4     R0      r4      r0 */\
1232         "movq 16(%2), %%mm4             \n\t" /* C4     C4      C4      C4 */\
1233         "pmaddwd %%mm0, %%mm4           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1234         "movq 24(%2), %%mm5             \n\t" /* -C4    C4      -C4     C4 */\
1235         "pmaddwd %%mm5, %%mm0           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1236         #rounder ", %%mm4               \n\t"\
1237         #rounder ", %%mm0               \n\t"\
1238         "psrad $" #shift ", %%mm4       \n\t"\
1239         "psrad $" #shift ", %%mm0       \n\t"\
1240         "movq 8+" #src0 ", %%mm2        \n\t" /* R4     R0      r4      r0 */\
1241         "movq 16(%2), %%mm1             \n\t" /* C4     C4      C4      C4 */\
1242         "pmaddwd %%mm2, %%mm1           \n\t" /* C4R4+C4R0      C4r4+C4r0 */\
1243         "movq 24(%2), %%mm7             \n\t" /* -C4    C4      -C4     C4 */\
1244         "pmaddwd %%mm7, %%mm2           \n\t" /* -C4R4+C4R0     -C4r4+C4r0 */\
1245         "movq 32(%2), %%mm7             \n\t" /* C6     C2      C6      C2 */\
1246         #rounder ", %%mm1               \n\t"\
1247         #rounder ", %%mm2               \n\t"\
1248         "psrad $" #shift ", %%mm1       \n\t"\
1249         "packssdw %%mm1, %%mm4          \n\t" /* A0     a0 */\
1250         "movq %%mm4, " #dst "           \n\t"\
1251         "psrad $" #shift ", %%mm2       \n\t"\
1252         "packssdw %%mm2, %%mm0          \n\t" /* A1     a1 */\
1253         "movq %%mm0, 16+" #dst "        \n\t"\
1254         "movq %%mm0, 96+" #dst "        \n\t"\
1255         "movq %%mm4, 112+" #dst "       \n\t"\
1256         "movq %%mm0, 32+" #dst "        \n\t"\
1257         "movq %%mm4, 48+" #dst "        \n\t"\
1258         "movq %%mm4, 64+" #dst "        \n\t"\
1259         "movq %%mm0, 80+" #dst "        \n\t"
1260
1261 //IDCT(  src0,   src4,   src1,    src5,    dst, rounder, shift)
1262 IDCT(   0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
1263 //IDCT(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
1264 IDCT(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
1265 //IDCT(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
1266
1267
1268 #endif
1269
1270 /*
1271 Input
1272  00 40 04 44 20 60 24 64
1273  10 30 14 34 50 70 54 74
1274  01 41 03 43 21 61 23 63
1275  11 31 13 33 51 71 53 73
1276  02 42 06 46 22 62 26 66
1277  12 32 16 36 52 72 56 76
1278  05 45 07 47 25 65 27 67
1279  15 35 17 37 55 75 57 77
1280
1281 Temp
1282  00 04 10 14 20 24 30 34
1283  40 44 50 54 60 64 70 74
1284  01 03 11 13 21 23 31 33
1285  41 43 51 53 61 63 71 73
1286  02 06 12 16 22 26 32 36
1287  42 46 52 56 62 66 72 76
1288  05 07 15 17 25 27 35 37
1289  45 47 55 57 65 67 75 77
1290 */
1291
1292 "9: \n\t"
1293                 :: "r" (block), "r" (temp), "r" (coeffs)
1294                 : "%eax"
1295         );
1296 }
1297
1298 void ff_simple_idct_mmx(int16_t *block)
1299 {
1300     idct(block);
1301 }
1302
1303 //FIXME merge add/put into the idct
1304
1305 void ff_simple_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1306 {
1307     idct(block);
1308     put_pixels_clamped_mmx(block, dest, line_size);
1309 }
1310 void ff_simple_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
1311 {
1312     idct(block);
1313     add_pixels_clamped_mmx(block, dest, line_size);
1314 }