git.sesse.net Git - ffmpeg/blob - libavcodec/sh4/idct_sh4.c

   1 /*
   2  * idct for sh4
   3  *
   4  * Copyright (c) 2001-2003 BERO <bero@geocities.co.jp>
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include "libavcodec/dsputil.h"
  24 #define c1      1.38703984532214752434  /* sqrt(2)*cos(1*pi/16) */
  25 #define c2      1.30656296487637657577  /* sqrt(2)*cos(2*pi/16) */
  26 #define c3      1.17587560241935884520  /* sqrt(2)*cos(3*pi/16) */
  27 #define c4      1.00000000000000000000  /* sqrt(2)*cos(4*pi/16) */
  28 #define c5      0.78569495838710234903  /* sqrt(2)*cos(5*pi/16) */
  29 #define c6      0.54119610014619712324  /* sqrt(2)*cos(6*pi/16) */
  30 #define c7      0.27589937928294311353  /* sqrt(2)*cos(7*pi/16) */
  31
  32 static const float even_table[] __attribute__ ((aligned(8))) = {
  33         c4, c4, c4, c4,
  34         c2, c6,-c6,-c2,
  35         c4,-c4,-c4, c4,
  36         c6,-c2, c2,-c6
  37 };
  38
  39 static const float odd_table[] __attribute__ ((aligned(8))) = {
  40         c1, c3, c5, c7,
  41         c3,-c7,-c1,-c5,
  42         c5,-c1, c7, c3,
  43         c7,-c5, c3,-c1
  44 };
  45
  46 #undef  c1
  47 #undef  c2
  48 #undef  c3
  49 #undef  c4
  50 #undef  c5
  51 #undef  c6
  52 #undef  c7
  53
  54 #if defined(__SH4_SINGLE__) || defined(__SH4_SINGLE_ONLY__)
  55
  56 #define         load_matrix(table) \
  57         __asm__ volatile( \
  58         "       fschg\n" \
  59         "       fmov   @%0+,xd0\n" \
  60         "       fmov   @%0+,xd2\n" \
  61         "       fmov   @%0+,xd4\n" \
  62         "       fmov   @%0+,xd6\n" \
  63         "       fmov   @%0+,xd8\n" \
  64         "       fmov   @%0+,xd10\n" \
  65         "       fmov   @%0+,xd12\n" \
  66         "       fmov   @%0+,xd14\n" \
  67         "       fschg\n" \
  68         :\
  69         : "r"(table)\
  70         : "0" \
  71         )
  72
  73 #define         ftrv() \
  74                 __asm__ volatile("ftrv xmtrx,fv0" \
  75                 : "=f"(fr0),"=f"(fr1),"=f"(fr2),"=f"(fr3) \
  76                 :  "0"(fr0), "1"(fr1), "2"(fr2), "3"(fr3) );
  77
  78 #define         DEFREG        \
  79         register float fr0 __asm__("fr0"); \
  80         register float fr1 __asm__("fr1"); \
  81         register float fr2 __asm__("fr2"); \
  82         register float fr3 __asm__("fr3")
  83
  84 #else
  85
  86 /* generic C code for check */
  87
  88 static void ftrv_(const float xf[],float fv[])
  89 {
  90         float f0,f1,f2,f3;
  91         f0 = fv[0];
  92         f1 = fv[1];
  93         f2 = fv[2];
  94         f3 = fv[3];
  95         fv[0] = xf[0]*f0 + xf[4]*f1 + xf[ 8]*f2 + xf[12]*f3;
  96         fv[1] = xf[1]*f0 + xf[5]*f1 + xf[ 9]*f2 + xf[13]*f3;
  97         fv[2] = xf[2]*f0 + xf[6]*f1 + xf[10]*f2 + xf[14]*f3;
  98         fv[3] = xf[3]*f0 + xf[7]*f1 + xf[11]*f2 + xf[15]*f3;
  99 }
 100
 101 static void load_matrix_(float xf[],const float table[])
 102 {
 103         int i;
 104         for(i=0;i<16;i++) xf[i]=table[i];
 105 }
 106
 107 #define         ftrv()                  ftrv_(xf,fv)
 108 #define         load_matrix(table)      load_matrix_(xf,table)
 109
 110 #define         DEFREG \
 111         float fv[4],xf[16]
 112
 113 #define         fr0     fv[0]
 114 #define         fr1     fv[1]
 115 #define         fr2     fv[2]
 116 #define         fr3     fv[3]
 117
 118 #endif
 119
 120 #if 1
 121 #define         DESCALE(x,n)    (x)*(1.0f/(1<<(n)))
 122 #else
 123 #define         DESCALE(x,n)    (((int)(x)+(1<<(n-1)))>>(n))
 124 #endif
 125
 126 /* this code work worse on gcc cvs. 3.2.3 work fine */
 127
 128
 129 #if 1
 130 //optimized
 131
 132 void idct_sh4(DCTELEM *block)
 133 {
 134         DEFREG;
 135
 136         int i;
 137         float        tblock[8*8],*fblock;
 138         int ofs1,ofs2,ofs3;
 139
 140 #if defined(__SH4__)
 141         __asm__ ("fschg");
 142 #endif
 143
 144         /* row */
 145
 146         /* even part */
 147         load_matrix(even_table);
 148
 149         fblock = tblock+4;
 150         i = 8;
 151         do {
 152                 fr0 = block[0];
 153                 fr1 = block[2];
 154                 fr2 = block[4];
 155                 fr3 = block[6];
 156                 block+=8;
 157                 ftrv();
 158                 *--fblock = fr3;
 159                 *--fblock = fr2;
 160                 *--fblock = fr1;
 161                 *--fblock = fr0;
 162                 fblock+=8+4;
 163         } while(--i);
 164         block-=8*8;
 165         fblock-=8*8+4;
 166
 167         load_matrix(odd_table);
 168
 169         i = 8;
 170
 171 //        ofs1 = sizeof(float)*1;
 172 //        ofs2 = sizeof(float)*2;
 173 //        ofs3 = sizeof(float)*3;
 174
 175         do {
 176                 float t0,t1,t2,t3;
 177                 fr0 = block[1];
 178                 fr1 = block[3];
 179                 fr2 = block[5];
 180                 fr3 = block[7];
 181                 block+=8;
 182                 ftrv();
 183                 t0 = *fblock++;
 184                 t1 = *fblock++;
 185                 t2 = *fblock++;
 186                 t3 = *fblock++;
 187                 fblock+=4;
 188                 *--fblock = t0 - fr0;
 189                 *--fblock = t1 - fr1;
 190                 *--fblock = t2 - fr2;
 191                 *--fblock = t3 - fr3;
 192                 *--fblock = t3 + fr3;
 193                 *--fblock = t2 + fr2;
 194                 *--fblock = t1 + fr1;
 195                 *--fblock = t0 + fr0;
 196                 fblock+=8;
 197         } while(--i);
 198         block-=8*8;
 199         fblock-=8*8;
 200
 201         /* col */
 202
 203         /* even part */
 204         load_matrix(even_table);
 205
 206         ofs1 = sizeof(float)*2*8;
 207         ofs2 = sizeof(float)*4*8;
 208         ofs3 = sizeof(float)*6*8;
 209
 210         i = 8;
 211
 212 #define        OA(fblock,ofs)   *(float*)((char*)fblock + ofs)
 213
 214         do {
 215                 fr0 = OA(fblock,   0);
 216                 fr1 = OA(fblock,ofs1);
 217                 fr2 = OA(fblock,ofs2);
 218                 fr3 = OA(fblock,ofs3);
 219                 ftrv();
 220                 OA(fblock,0   ) = fr0;
 221                 OA(fblock,ofs1) = fr1;
 222                 OA(fblock,ofs2) = fr2;
 223                 OA(fblock,ofs3) = fr3;
 224                 fblock++;
 225         } while(--i);
 226         fblock-=8;
 227
 228         load_matrix(odd_table);
 229
 230         i=8;
 231         do {
 232                 float t0,t1,t2,t3;
 233                 t0 = OA(fblock,   0); /* [8*0] */
 234                 t1 = OA(fblock,ofs1); /* [8*2] */
 235                 t2 = OA(fblock,ofs2); /* [8*4] */
 236                 t3 = OA(fblock,ofs3); /* [8*6] */
 237                 fblock+=8;
 238                 fr0 = OA(fblock,   0); /* [8*1] */
 239                 fr1 = OA(fblock,ofs1); /* [8*3] */
 240                 fr2 = OA(fblock,ofs2); /* [8*5] */
 241                 fr3 = OA(fblock,ofs3); /* [8*7] */
 242                 fblock+=-8+1;
 243                 ftrv();
 244                 block[8*0] = DESCALE(t0 + fr0,3);
 245                 block[8*7] = DESCALE(t0 - fr0,3);
 246                 block[8*1] = DESCALE(t1 + fr1,3);
 247                 block[8*6] = DESCALE(t1 - fr1,3);
 248                 block[8*2] = DESCALE(t2 + fr2,3);
 249                 block[8*5] = DESCALE(t2 - fr2,3);
 250                 block[8*3] = DESCALE(t3 + fr3,3);
 251                 block[8*4] = DESCALE(t3 - fr3,3);
 252                 block++;
 253         } while(--i);
 254
 255 #if defined(__SH4__)
 256         __asm__ ("fschg");
 257 #endif
 258 }
 259 #else
 260 void idct_sh4(DCTELEM *block)
 261 {
 262         DEFREG;
 263
 264         int i;
 265         float   tblock[8*8],*fblock;
 266
 267         /* row */
 268
 269         /* even part */
 270         load_matrix(even_table);
 271
 272         fblock = tblock;
 273         i = 8;
 274         do {
 275                 fr0 = block[0];
 276                 fr1 = block[2];
 277                 fr2 = block[4];
 278                 fr3 = block[6];
 279                 block+=8;
 280                 ftrv();
 281                 fblock[0] = fr0;
 282                 fblock[2] = fr1;
 283                 fblock[4] = fr2;
 284                 fblock[6] = fr3;
 285                 fblock+=8;
 286         } while(--i);
 287         block-=8*8;
 288         fblock-=8*8;
 289
 290         load_matrix(odd_table);
 291
 292         i = 8;
 293
 294         do {
 295                 float t0,t1,t2,t3;
 296                 fr0 = block[1];
 297                 fr1 = block[3];
 298                 fr2 = block[5];
 299                 fr3 = block[7];
 300                 block+=8;
 301                 ftrv();
 302                 t0 = fblock[0];
 303                 t1 = fblock[2];
 304                 t2 = fblock[4];
 305                 t3 = fblock[6];
 306                 fblock[0] = t0 + fr0;
 307                 fblock[7] = t0 - fr0;
 308                 fblock[1] = t1 + fr1;
 309                 fblock[6] = t1 - fr1;
 310                 fblock[2] = t2 + fr2;
 311                 fblock[5] = t2 - fr2;
 312                 fblock[3] = t3 + fr3;
 313                 fblock[4] = t3 - fr3;
 314                 fblock+=8;
 315         } while(--i);
 316         block-=8*8;
 317         fblock-=8*8;
 318
 319         /* col */
 320
 321         /* even part */
 322         load_matrix(even_table);
 323
 324         i = 8;
 325
 326         do {
 327                 fr0 = fblock[8*0];
 328                 fr1 = fblock[8*2];
 329                 fr2 = fblock[8*4];
 330                 fr3 = fblock[8*6];
 331                 ftrv();
 332                 fblock[8*0] = fr0;
 333                 fblock[8*2] = fr1;
 334                 fblock[8*4] = fr2;
 335                 fblock[8*6] = fr3;
 336                 fblock++;
 337         } while(--i);
 338         fblock-=8;
 339
 340         load_matrix(odd_table);
 341
 342         i=8;
 343         do {
 344                 float t0,t1,t2,t3;
 345                 fr0 = fblock[8*1];
 346                 fr1 = fblock[8*3];
 347                 fr2 = fblock[8*5];
 348                 fr3 = fblock[8*7];
 349                 ftrv();
 350                 t0 = fblock[8*0];
 351                 t1 = fblock[8*2];
 352                 t2 = fblock[8*4];
 353                 t3 = fblock[8*6];
 354                 fblock++;
 355                 block[8*0] = DESCALE(t0 + fr0,3);
 356                 block[8*7] = DESCALE(t0 - fr0,3);
 357                 block[8*1] = DESCALE(t1 + fr1,3);
 358                 block[8*6] = DESCALE(t1 - fr1,3);
 359                 block[8*2] = DESCALE(t2 + fr2,3);
 360                 block[8*5] = DESCALE(t2 - fr2,3);
 361                 block[8*3] = DESCALE(t3 + fr3,3);
 362                 block[8*4] = DESCALE(t3 - fr3,3);
 363                 block++;
 364         } while(--i);
 365 }
 366 #endif