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

   1 /*
   2  * aligned/packed access motion
   3  *
   4  * Copyright (c) 2001-2003 BERO <bero@geocities.co.jp>
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23
  24 #include "libavcodec/avcodec.h"
  25 #include "libavcodec/dsputil.h"
  26 #include "dsputil_sh4.h"
  27
  28
  29 #define         LP(p)           *(uint32_t*)(p)
  30 #define         LPC(p)          *(const uint32_t*)(p)
  31
  32
  33 #define         UNPACK(ph,pl,tt0,tt1) do { \
  34         uint32_t t0,t1; t0=tt0;t1=tt1; \
  35         ph = ( (t0 & ~BYTE_VEC32(0x03))>>2) + ( (t1 & ~BYTE_VEC32(0x03))>>2); \
  36         pl = (t0 & BYTE_VEC32(0x03)) + (t1 & BYTE_VEC32(0x03)); } while(0)
  37
  38 #define         rnd_PACK(ph,pl,nph,npl) ph + nph + (((pl + npl + BYTE_VEC32(0x02))>>2) & BYTE_VEC32(0x03))
  39 #define         no_rnd_PACK(ph,pl,nph,npl)      ph + nph + (((pl + npl + BYTE_VEC32(0x01))>>2) & BYTE_VEC32(0x03))
  40
  41 /* little-endian */
  42 #define         MERGE1(a,b,ofs) (ofs==0)?a:( ((a)>>(8*ofs))|((b)<<(32-8*ofs)) )
  43 #define         MERGE2(a,b,ofs) (ofs==3)?b:( ((a)>>(8*(ofs+1)))|((b)<<(32-8*(ofs+1))) )
  44 /* big
  45 #define         MERGE1(a,b,ofs) (ofs==0)?a:( ((a)<<(8*ofs))|((b)>>(32-8*ofs)) )
  46 #define         MERGE2(a,b,ofs) (ofs==3)?b:( ((a)<<(8+8*ofs))|((b)>>(32-8-8*ofs)) )
  47 */
  48
  49
  50 #define         put(d,s)        d = s
  51 #define         avg(d,s)        d = rnd_avg32(s,d)
  52
  53 #define         OP_C4(ofs) \
  54         ref-=ofs; \
  55         do { \
  56                 OP(LP(dest),MERGE1(LPC(ref),LPC(ref+4),ofs)); \
  57                 ref+=stride; \
  58                 dest+=stride; \
  59         } while(--height)
  60
  61 #define        OP_C40() \
  62         do { \
  63                 OP(LP(dest),LPC(ref)); \
  64                 ref+=stride; \
  65                 dest+=stride; \
  66         } while(--height)
  67
  68
  69 #define         OP      put
  70
  71 static void put_pixels4_c(uint8_t *dest,const uint8_t *ref, const int stride,int height)
  72 {
  73         switch((int)ref&3){
  74         case 0: OP_C40(); return;
  75         case 1: OP_C4(1); return;
  76         case 2: OP_C4(2); return;
  77         case 3: OP_C4(3); return;
  78         }
  79 }
  80
  81 #undef          OP
  82 #define         OP      avg
  83
  84 static void avg_pixels4_c(uint8_t *dest,const uint8_t *ref, const int stride,int height)
  85 {
  86         switch((int)ref&3){
  87         case 0: OP_C40(); return;
  88         case 1: OP_C4(1); return;
  89         case 2: OP_C4(2); return;
  90         case 3: OP_C4(3); return;
  91         }
  92 }
  93
  94 #undef          OP
  95
  96 #define         OP_C(ofs,sz,avg2) \
  97 { \
  98         ref-=ofs; \
  99         do { \
 100                 uint32_t        t0,t1; \
 101                 t0 = LPC(ref+0); \
 102                 t1 = LPC(ref+4); \
 103                 OP(LP(dest+0), MERGE1(t0,t1,ofs)); \
 104                 t0 = LPC(ref+8); \
 105                 OP(LP(dest+4), MERGE1(t1,t0,ofs)); \
 106 if (sz==16) { \
 107                 t1 = LPC(ref+12); \
 108                 OP(LP(dest+8), MERGE1(t0,t1,ofs)); \
 109                 t0 = LPC(ref+16); \
 110                 OP(LP(dest+12), MERGE1(t1,t0,ofs)); \
 111 } \
 112                 ref+=stride; \
 113                 dest+= stride; \
 114         } while(--height); \
 115 }
 116
 117 /* aligned */
 118 #define         OP_C0(sz,avg2) \
 119 { \
 120         do { \
 121                 OP(LP(dest+0), LPC(ref+0)); \
 122                 OP(LP(dest+4), LPC(ref+4)); \
 123 if (sz==16) { \
 124                 OP(LP(dest+8), LPC(ref+8)); \
 125                 OP(LP(dest+12), LPC(ref+12)); \
 126 } \
 127                 ref+=stride; \
 128                 dest+= stride; \
 129         } while(--height); \
 130 }
 131
 132 #define         OP_X(ofs,sz,avg2) \
 133 { \
 134         ref-=ofs; \
 135         do { \
 136                 uint32_t        t0,t1; \
 137                 t0 = LPC(ref+0); \
 138                 t1 = LPC(ref+4); \
 139                 OP(LP(dest+0), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
 140                 t0 = LPC(ref+8); \
 141                 OP(LP(dest+4), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
 142 if (sz==16) { \
 143                 t1 = LPC(ref+12); \
 144                 OP(LP(dest+8), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
 145                 t0 = LPC(ref+16); \
 146                 OP(LP(dest+12), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
 147 } \
 148                 ref+=stride; \
 149                 dest+= stride; \
 150         } while(--height); \
 151 }
 152
 153 /* aligned */
 154 #define         OP_Y0(sz,avg2) \
 155 { \
 156         uint32_t t0,t1,t2,t3,t; \
 157 \
 158         t0 = LPC(ref+0); \
 159         t1 = LPC(ref+4); \
 160 if (sz==16) { \
 161         t2 = LPC(ref+8); \
 162         t3 = LPC(ref+12); \
 163 } \
 164         do { \
 165                 ref += stride; \
 166 \
 167                 t = LPC(ref+0); \
 168                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
 169                 t = LPC(ref+4); \
 170                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
 171 if (sz==16) { \
 172                 t = LPC(ref+8); \
 173                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
 174                 t = LPC(ref+12); \
 175                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
 176 } \
 177                 dest+= stride; \
 178         } while(--height); \
 179 }
 180
 181 #define         OP_Y(ofs,sz,avg2) \
 182 { \
 183         uint32_t t0,t1,t2,t3,t,w0,w1; \
 184 \
 185         ref-=ofs; \
 186         w0 = LPC(ref+0); \
 187         w1 = LPC(ref+4); \
 188         t0 = MERGE1(w0,w1,ofs); \
 189         w0 = LPC(ref+8); \
 190         t1 = MERGE1(w1,w0,ofs); \
 191 if (sz==16) { \
 192         w1 = LPC(ref+12); \
 193         t2 = MERGE1(w0,w1,ofs); \
 194         w0 = LPC(ref+16); \
 195         t3 = MERGE1(w1,w0,ofs); \
 196 } \
 197         do { \
 198                 ref += stride; \
 199 \
 200                 w0 = LPC(ref+0); \
 201                 w1 = LPC(ref+4); \
 202                 t = MERGE1(w0,w1,ofs); \
 203                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
 204                 w0 = LPC(ref+8); \
 205                 t = MERGE1(w1,w0,ofs); \
 206                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
 207 if (sz==16) { \
 208                 w1 = LPC(ref+12); \
 209                 t = MERGE1(w0,w1,ofs); \
 210                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
 211                 w0 = LPC(ref+16); \
 212                 t = MERGE1(w1,w0,ofs); \
 213                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
 214 } \
 215                 dest+=stride; \
 216         } while(--height); \
 217 }
 218
 219 #define OP_X0(sz,avg2) OP_X(0,sz,avg2)
 220 #define OP_XY0(sz,PACK) OP_XY(0,sz,PACK)
 221 #define         OP_XY(ofs,sz,PACK) \
 222 { \
 223         uint32_t        t2,t3,w0,w1; \
 224         uint32_t        a0,a1,a2,a3,a4,a5,a6,a7; \
 225 \
 226         ref -= ofs; \
 227         w0 = LPC(ref+0); \
 228         w1 = LPC(ref+4); \
 229         UNPACK(a0,a1,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 230         w0 = LPC(ref+8); \
 231         UNPACK(a2,a3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 232 if (sz==16) { \
 233         w1 = LPC(ref+12); \
 234         UNPACK(a4,a5,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 235         w0 = LPC(ref+16); \
 236         UNPACK(a6,a7,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 237 } \
 238         do { \
 239                 ref+=stride; \
 240                 w0 = LPC(ref+0); \
 241                 w1 = LPC(ref+4); \
 242                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 243                 OP(LP(dest+0),PACK(a0,a1,t2,t3)); \
 244                 a0 = t2; a1 = t3; \
 245                 w0 = LPC(ref+8); \
 246                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 247                 OP(LP(dest+4),PACK(a2,a3,t2,t3)); \
 248                 a2 = t2; a3 = t3; \
 249 if (sz==16) { \
 250                 w1 = LPC(ref+12); \
 251                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 252                 OP(LP(dest+8),PACK(a4,a5,t2,t3)); \
 253                 a4 = t2; a5 = t3; \
 254                 w0 = LPC(ref+16); \
 255                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 256                 OP(LP(dest+12),PACK(a6,a7,t2,t3)); \
 257                 a6 = t2; a7 = t3; \
 258 } \
 259                 dest+=stride; \
 260         } while(--height); \
 261 }
 262
 263 #define         DEFFUNC(op,rnd,xy,sz,OP_N,avgfunc) \
 264 static void op##_##rnd##_pixels##sz##_##xy (uint8_t * dest, const uint8_t * ref, \
 265                                 const int stride, int height) \
 266 { \
 267         switch((int)ref&3) { \
 268         case 0:OP_N##0(sz,rnd##_##avgfunc); return; \
 269         case 1:OP_N(1,sz,rnd##_##avgfunc); return; \
 270         case 2:OP_N(2,sz,rnd##_##avgfunc); return; \
 271         case 3:OP_N(3,sz,rnd##_##avgfunc); return; \
 272         } \
 273 }
 274
 275 #define OP put
 276
 277 DEFFUNC(put,   rnd,o,8,OP_C,avg32)
 278 DEFFUNC(put,   rnd,x,8,OP_X,avg32)
 279 DEFFUNC(put,no_rnd,x,8,OP_X,avg32)
 280 DEFFUNC(put,   rnd,y,8,OP_Y,avg32)
 281 DEFFUNC(put,no_rnd,y,8,OP_Y,avg32)
 282 DEFFUNC(put,   rnd,xy,8,OP_XY,PACK)
 283 DEFFUNC(put,no_rnd,xy,8,OP_XY,PACK)
 284 DEFFUNC(put,   rnd,o,16,OP_C,avg32)
 285 DEFFUNC(put,   rnd,x,16,OP_X,avg32)
 286 DEFFUNC(put,no_rnd,x,16,OP_X,avg32)
 287 DEFFUNC(put,   rnd,y,16,OP_Y,avg32)
 288 DEFFUNC(put,no_rnd,y,16,OP_Y,avg32)
 289 DEFFUNC(put,   rnd,xy,16,OP_XY,PACK)
 290 DEFFUNC(put,no_rnd,xy,16,OP_XY,PACK)
 291
 292 #undef OP
 293 #define OP avg
 294
 295 DEFFUNC(avg,   rnd,o,8,OP_C,avg32)
 296 DEFFUNC(avg,   rnd,x,8,OP_X,avg32)
 297 DEFFUNC(avg,   rnd,y,8,OP_Y,avg32)
 298 DEFFUNC(avg,   rnd,xy,8,OP_XY,PACK)
 299 DEFFUNC(avg,   rnd,o,16,OP_C,avg32)
 300 DEFFUNC(avg,   rnd,x,16,OP_X,avg32)
 301 DEFFUNC(avg,no_rnd,x,16,OP_X,avg32)
 302 DEFFUNC(avg,   rnd,y,16,OP_Y,avg32)
 303 DEFFUNC(avg,no_rnd,y,16,OP_Y,avg32)
 304 DEFFUNC(avg,   rnd,xy,16,OP_XY,PACK)
 305 DEFFUNC(avg,no_rnd,xy,16,OP_XY,PACK)
 306
 307 #undef OP
 308
 309 #define         put_no_rnd_pixels8_o     put_rnd_pixels8_o
 310 #define         put_no_rnd_pixels16_o    put_rnd_pixels16_o
 311 #define         avg_no_rnd_pixels16_o    avg_rnd_pixels16_o
 312
 313 #define         put_pixels8_c            put_rnd_pixels8_o
 314 #define         put_pixels16_c           put_rnd_pixels16_o
 315 #define         avg_pixels8_c            avg_rnd_pixels8_o
 316 #define         avg_pixels16_c           avg_rnd_pixels16_o
 317 #define         put_no_rnd_pixels8_c     put_rnd_pixels8_o
 318 #define         put_no_rnd_pixels16_c    put_rnd_pixels16_o
 319 #define         avg_no_rnd_pixels16_c    avg_rnd_pixels16_o
 320
 321 #define         QPEL
 322
 323 #ifdef QPEL
 324
 325 #include "qpel.c"
 326
 327 #endif
 328
 329 void ff_dsputil_init_align(DSPContext* c, AVCodecContext *avctx)
 330 {
 331         const int high_bit_depth = avctx->bits_per_raw_sample > 8;
 332
 333         if (!high_bit_depth) {
 334         c->put_pixels_tab[0][0] = put_rnd_pixels16_o;
 335         c->put_pixels_tab[0][1] = put_rnd_pixels16_x;
 336         c->put_pixels_tab[0][2] = put_rnd_pixels16_y;
 337         c->put_pixels_tab[0][3] = put_rnd_pixels16_xy;
 338         c->put_pixels_tab[1][0] = put_rnd_pixels8_o;
 339         c->put_pixels_tab[1][1] = put_rnd_pixels8_x;
 340         c->put_pixels_tab[1][2] = put_rnd_pixels8_y;
 341         c->put_pixels_tab[1][3] = put_rnd_pixels8_xy;
 342
 343         c->put_no_rnd_pixels_tab[0][0] = put_no_rnd_pixels16_o;
 344         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x;
 345         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y;
 346         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy;
 347         c->put_no_rnd_pixels_tab[1][0] = put_no_rnd_pixels8_o;
 348         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x;
 349         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y;
 350         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy;
 351
 352         c->avg_pixels_tab[0][0] = avg_rnd_pixels16_o;
 353         c->avg_pixels_tab[0][1] = avg_rnd_pixels16_x;
 354         c->avg_pixels_tab[0][2] = avg_rnd_pixels16_y;
 355         c->avg_pixels_tab[0][3] = avg_rnd_pixels16_xy;
 356         c->avg_pixels_tab[1][0] = avg_rnd_pixels8_o;
 357         c->avg_pixels_tab[1][1] = avg_rnd_pixels8_x;
 358         c->avg_pixels_tab[1][2] = avg_rnd_pixels8_y;
 359         c->avg_pixels_tab[1][3] = avg_rnd_pixels8_xy;
 360
 361         c->avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels16_o;
 362         c->avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels16_x;
 363         c->avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels16_y;
 364         c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy;
 365         }
 366
 367 #ifdef QPEL
 368
 369 #define dspfunc(PFX, IDX, NUM) \
 370     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_sh4; \
 371     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_sh4; \
 372     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_sh4; \
 373     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_sh4; \
 374     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_sh4; \
 375     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_sh4; \
 376     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_sh4; \
 377     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_sh4; \
 378     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_sh4; \
 379     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_sh4; \
 380     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_sh4; \
 381     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_sh4; \
 382     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_sh4; \
 383     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_sh4; \
 384     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_sh4; \
 385     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_sh4
 386
 387     dspfunc(put_qpel, 0, 16);
 388     dspfunc(put_no_rnd_qpel, 0, 16);
 389
 390     dspfunc(avg_qpel, 0, 16);
 391     /* dspfunc(avg_no_rnd_qpel, 0, 16); */
 392
 393     dspfunc(put_qpel, 1, 8);
 394     dspfunc(put_no_rnd_qpel, 1, 8);
 395
 396     dspfunc(avg_qpel, 1, 8);
 397     /* dspfunc(avg_no_rnd_qpel, 1, 8); */
 398
 399     if (!high_bit_depth) {
 400     dspfunc(put_h264_qpel, 0, 16);
 401     dspfunc(put_h264_qpel, 1, 8);
 402     dspfunc(put_h264_qpel, 2, 4);
 403     dspfunc(avg_h264_qpel, 0, 16);
 404     dspfunc(avg_h264_qpel, 1, 8);
 405     dspfunc(avg_h264_qpel, 2, 4);
 406     }
 407
 408 #undef dspfunc
 409     if (!high_bit_depth) {
 410     c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_sh4;
 411     c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_sh4;
 412     c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_sh4;
 413     c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_sh4;
 414     c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_sh4;
 415     c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_sh4;
 416     }
 417
 418     c->put_mspel_pixels_tab[0]= put_mspel8_mc00_sh4;
 419     c->put_mspel_pixels_tab[1]= put_mspel8_mc10_sh4;
 420     c->put_mspel_pixels_tab[2]= put_mspel8_mc20_sh4;
 421     c->put_mspel_pixels_tab[3]= put_mspel8_mc30_sh4;
 422     c->put_mspel_pixels_tab[4]= put_mspel8_mc02_sh4;
 423     c->put_mspel_pixels_tab[5]= put_mspel8_mc12_sh4;
 424     c->put_mspel_pixels_tab[6]= put_mspel8_mc22_sh4;
 425     c->put_mspel_pixels_tab[7]= put_mspel8_mc32_sh4;
 426
 427     c->gmc1 = gmc1_c;
 428
 429 #endif
 430 }