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 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 "libavutil/attributes.h"
  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 #define         OP_C(ofs,sz,avg2) \
  69 { \
  70         ref-=ofs; \
  71         do { \
  72                 uint32_t        t0,t1; \
  73                 t0 = LPC(ref+0); \
  74                 t1 = LPC(ref+4); \
  75                 OP(LP(dest+0), MERGE1(t0,t1,ofs)); \
  76                 t0 = LPC(ref+8); \
  77                 OP(LP(dest+4), MERGE1(t1,t0,ofs)); \
  78 if (sz==16) { \
  79                 t1 = LPC(ref+12); \
  80                 OP(LP(dest+8), MERGE1(t0,t1,ofs)); \
  81                 t0 = LPC(ref+16); \
  82                 OP(LP(dest+12), MERGE1(t1,t0,ofs)); \
  83 } \
  84                 ref+=stride; \
  85                 dest+= stride; \
  86         } while(--height); \
  87 }
  88
  89 /* aligned */
  90 #define         OP_C0(sz,avg2) \
  91 { \
  92         do { \
  93                 OP(LP(dest+0), LPC(ref+0)); \
  94                 OP(LP(dest+4), LPC(ref+4)); \
  95 if (sz==16) { \
  96                 OP(LP(dest+8), LPC(ref+8)); \
  97                 OP(LP(dest+12), LPC(ref+12)); \
  98 } \
  99                 ref+=stride; \
 100                 dest+= stride; \
 101         } while(--height); \
 102 }
 103
 104 #define         OP_X(ofs,sz,avg2) \
 105 { \
 106         ref-=ofs; \
 107         do { \
 108                 uint32_t        t0,t1; \
 109                 t0 = LPC(ref+0); \
 110                 t1 = LPC(ref+4); \
 111                 OP(LP(dest+0), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
 112                 t0 = LPC(ref+8); \
 113                 OP(LP(dest+4), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
 114 if (sz==16) { \
 115                 t1 = LPC(ref+12); \
 116                 OP(LP(dest+8), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
 117                 t0 = LPC(ref+16); \
 118                 OP(LP(dest+12), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
 119 } \
 120                 ref+=stride; \
 121                 dest+= stride; \
 122         } while(--height); \
 123 }
 124
 125 /* aligned */
 126 #define         OP_Y0(sz,avg2) \
 127 { \
 128         uint32_t t0,t1,t2,t3,t; \
 129 \
 130         t0 = LPC(ref+0); \
 131         t1 = LPC(ref+4); \
 132 if (sz==16) { \
 133         t2 = LPC(ref+8); \
 134         t3 = LPC(ref+12); \
 135 } \
 136         do { \
 137                 ref += stride; \
 138 \
 139                 t = LPC(ref+0); \
 140                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
 141                 t = LPC(ref+4); \
 142                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
 143 if (sz==16) { \
 144                 t = LPC(ref+8); \
 145                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
 146                 t = LPC(ref+12); \
 147                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
 148 } \
 149                 dest+= stride; \
 150         } while(--height); \
 151 }
 152
 153 #define         OP_Y(ofs,sz,avg2) \
 154 { \
 155         uint32_t t0,t1,t2,t3,t,w0,w1; \
 156 \
 157         ref-=ofs; \
 158         w0 = LPC(ref+0); \
 159         w1 = LPC(ref+4); \
 160         t0 = MERGE1(w0,w1,ofs); \
 161         w0 = LPC(ref+8); \
 162         t1 = MERGE1(w1,w0,ofs); \
 163 if (sz==16) { \
 164         w1 = LPC(ref+12); \
 165         t2 = MERGE1(w0,w1,ofs); \
 166         w0 = LPC(ref+16); \
 167         t3 = MERGE1(w1,w0,ofs); \
 168 } \
 169         do { \
 170                 ref += stride; \
 171 \
 172                 w0 = LPC(ref+0); \
 173                 w1 = LPC(ref+4); \
 174                 t = MERGE1(w0,w1,ofs); \
 175                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
 176                 w0 = LPC(ref+8); \
 177                 t = MERGE1(w1,w0,ofs); \
 178                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
 179 if (sz==16) { \
 180                 w1 = LPC(ref+12); \
 181                 t = MERGE1(w0,w1,ofs); \
 182                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
 183                 w0 = LPC(ref+16); \
 184                 t = MERGE1(w1,w0,ofs); \
 185                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
 186 } \
 187                 dest+=stride; \
 188         } while(--height); \
 189 }
 190
 191 #define OP_X0(sz,avg2) OP_X(0,sz,avg2)
 192 #define OP_XY0(sz,PACK) OP_XY(0,sz,PACK)
 193 #define         OP_XY(ofs,sz,PACK) \
 194 { \
 195         uint32_t        t2,t3,w0,w1; \
 196         uint32_t        a0,a1,a2,a3,a4,a5,a6,a7; \
 197 \
 198         ref -= ofs; \
 199         w0 = LPC(ref+0); \
 200         w1 = LPC(ref+4); \
 201         UNPACK(a0,a1,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 202         w0 = LPC(ref+8); \
 203         UNPACK(a2,a3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 204 if (sz==16) { \
 205         w1 = LPC(ref+12); \
 206         UNPACK(a4,a5,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 207         w0 = LPC(ref+16); \
 208         UNPACK(a6,a7,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 209 } \
 210         do { \
 211                 ref+=stride; \
 212                 w0 = LPC(ref+0); \
 213                 w1 = LPC(ref+4); \
 214                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 215                 OP(LP(dest+0),PACK(a0,a1,t2,t3)); \
 216                 a0 = t2; a1 = t3; \
 217                 w0 = LPC(ref+8); \
 218                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 219                 OP(LP(dest+4),PACK(a2,a3,t2,t3)); \
 220                 a2 = t2; a3 = t3; \
 221 if (sz==16) { \
 222                 w1 = LPC(ref+12); \
 223                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
 224                 OP(LP(dest+8),PACK(a4,a5,t2,t3)); \
 225                 a4 = t2; a5 = t3; \
 226                 w0 = LPC(ref+16); \
 227                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
 228                 OP(LP(dest+12),PACK(a6,a7,t2,t3)); \
 229                 a6 = t2; a7 = t3; \
 230 } \
 231                 dest+=stride; \
 232         } while(--height); \
 233 }
 234
 235 #define         DEFFUNC(op,rnd,xy,sz,OP_N,avgfunc) \
 236 static void op##_##rnd##_pixels##sz##_##xy (uint8_t * dest, const uint8_t * ref, \
 237                                 const ptrdiff_t stride, int height) \
 238 { \
 239         switch((int)ref&3) { \
 240         case 0:OP_N##0(sz,rnd##_##avgfunc); return; \
 241         case 1:OP_N(1,sz,rnd##_##avgfunc); return; \
 242         case 2:OP_N(2,sz,rnd##_##avgfunc); return; \
 243         case 3:OP_N(3,sz,rnd##_##avgfunc); return; \
 244         } \
 245 }
 246
 247 #define OP put
 248
 249 DEFFUNC(put,   rnd,o,8,OP_C,avg32)
 250 DEFFUNC(put,   rnd,x,8,OP_X,avg32)
 251 DEFFUNC(put,no_rnd,x,8,OP_X,avg32)
 252 DEFFUNC(put,   rnd,y,8,OP_Y,avg32)
 253 DEFFUNC(put,no_rnd,y,8,OP_Y,avg32)
 254 DEFFUNC(put,   rnd,xy,8,OP_XY,PACK)
 255 DEFFUNC(put,no_rnd,xy,8,OP_XY,PACK)
 256 DEFFUNC(put,   rnd,o,16,OP_C,avg32)
 257 DEFFUNC(put,   rnd,x,16,OP_X,avg32)
 258 DEFFUNC(put,no_rnd,x,16,OP_X,avg32)
 259 DEFFUNC(put,   rnd,y,16,OP_Y,avg32)
 260 DEFFUNC(put,no_rnd,y,16,OP_Y,avg32)
 261 DEFFUNC(put,   rnd,xy,16,OP_XY,PACK)
 262 DEFFUNC(put,no_rnd,xy,16,OP_XY,PACK)
 263
 264 #undef OP
 265 #define OP avg
 266
 267 DEFFUNC(avg,   rnd,o,8,OP_C,avg32)
 268 DEFFUNC(avg,   rnd,x,8,OP_X,avg32)
 269 DEFFUNC(avg,   rnd,y,8,OP_Y,avg32)
 270 DEFFUNC(avg,   rnd,xy,8,OP_XY,PACK)
 271 DEFFUNC(avg,   rnd,o,16,OP_C,avg32)
 272 DEFFUNC(avg,   rnd,x,16,OP_X,avg32)
 273 DEFFUNC(avg,no_rnd,x,16,OP_X,avg32)
 274 DEFFUNC(avg,   rnd,y,16,OP_Y,avg32)
 275 DEFFUNC(avg,no_rnd,y,16,OP_Y,avg32)
 276 DEFFUNC(avg,   rnd,xy,16,OP_XY,PACK)
 277 DEFFUNC(avg,no_rnd,xy,16,OP_XY,PACK)
 278
 279 #undef OP
 280
 281 #define         put_no_rnd_pixels8_o     put_rnd_pixels8_o
 282 #define         put_no_rnd_pixels16_o    put_rnd_pixels16_o
 283 #define         avg_no_rnd_pixels16_o    avg_rnd_pixels16_o
 284
 285 #define         put_pixels8_c            put_rnd_pixels8_o
 286 #define         put_pixels16_c           put_rnd_pixels16_o
 287 #define         avg_pixels8_c            avg_rnd_pixels8_o
 288 #define         avg_pixels16_c           avg_rnd_pixels16_o
 289 #define         put_no_rnd_pixels8_c     put_rnd_pixels8_o
 290 #define         put_no_rnd_pixels16_c    put_rnd_pixels16_o
 291 #define         avg_no_rnd_pixels16_c    avg_rnd_pixels16_o
 292
 293 #define         QPEL
 294
 295 #ifdef QPEL
 296
 297 #include "qpel.c"
 298
 299 #endif
 300
 301 av_cold void ff_dsputil_init_align(DSPContext *c, AVCodecContext *avctx)
 302 {
 303         const int high_bit_depth = avctx->bits_per_raw_sample > 8;
 304
 305         if (!high_bit_depth) {
 306         c->put_pixels_tab[0][0] = put_rnd_pixels16_o;
 307         c->put_pixels_tab[0][1] = put_rnd_pixels16_x;
 308         c->put_pixels_tab[0][2] = put_rnd_pixels16_y;
 309         c->put_pixels_tab[0][3] = put_rnd_pixels16_xy;
 310         c->put_pixels_tab[1][0] = put_rnd_pixels8_o;
 311         c->put_pixels_tab[1][1] = put_rnd_pixels8_x;
 312         c->put_pixels_tab[1][2] = put_rnd_pixels8_y;
 313         c->put_pixels_tab[1][3] = put_rnd_pixels8_xy;
 314
 315         c->put_no_rnd_pixels_tab[0][0] = put_no_rnd_pixels16_o;
 316         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x;
 317         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y;
 318         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy;
 319         c->put_no_rnd_pixels_tab[1][0] = put_no_rnd_pixels8_o;
 320         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x;
 321         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y;
 322         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy;
 323
 324         c->avg_pixels_tab[0][0] = avg_rnd_pixels16_o;
 325         c->avg_pixels_tab[0][1] = avg_rnd_pixels16_x;
 326         c->avg_pixels_tab[0][2] = avg_rnd_pixels16_y;
 327         c->avg_pixels_tab[0][3] = avg_rnd_pixels16_xy;
 328         c->avg_pixels_tab[1][0] = avg_rnd_pixels8_o;
 329         c->avg_pixels_tab[1][1] = avg_rnd_pixels8_x;
 330         c->avg_pixels_tab[1][2] = avg_rnd_pixels8_y;
 331         c->avg_pixels_tab[1][3] = avg_rnd_pixels8_xy;
 332
 333         c->avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels16_o;
 334         c->avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels16_x;
 335         c->avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels16_y;
 336         c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy;
 337         }
 338
 339 #ifdef QPEL
 340
 341 #define dspfunc(PFX, IDX, NUM) \
 342     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_sh4; \
 343     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_sh4; \
 344     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_sh4; \
 345     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_sh4; \
 346     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_sh4; \
 347     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_sh4; \
 348     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_sh4; \
 349     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_sh4; \
 350     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_sh4; \
 351     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_sh4; \
 352     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_sh4; \
 353     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_sh4; \
 354     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_sh4; \
 355     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_sh4; \
 356     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_sh4; \
 357     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_sh4
 358
 359     dspfunc(put_qpel, 0, 16);
 360     dspfunc(put_no_rnd_qpel, 0, 16);
 361
 362     dspfunc(avg_qpel, 0, 16);
 363     /* dspfunc(avg_no_rnd_qpel, 0, 16); */
 364
 365     dspfunc(put_qpel, 1, 8);
 366     dspfunc(put_no_rnd_qpel, 1, 8);
 367
 368     dspfunc(avg_qpel, 1, 8);
 369     /* dspfunc(avg_no_rnd_qpel, 1, 8); */
 370
 371 #undef dspfunc
 372
 373     c->put_mspel_pixels_tab[0]= put_mspel8_mc00_sh4;
 374     c->put_mspel_pixels_tab[1]= put_mspel8_mc10_sh4;
 375     c->put_mspel_pixels_tab[2]= put_mspel8_mc20_sh4;
 376     c->put_mspel_pixels_tab[3]= put_mspel8_mc30_sh4;
 377     c->put_mspel_pixels_tab[4]= put_mspel8_mc02_sh4;
 378     c->put_mspel_pixels_tab[5]= put_mspel8_mc12_sh4;
 379     c->put_mspel_pixels_tab[6]= put_mspel8_mc22_sh4;
 380     c->put_mspel_pixels_tab[7]= put_mspel8_mc32_sh4;
 381
 382     c->gmc1 = gmc1_c;
 383
 384 #endif
 385 }