libmpcodecs/vf_tfields.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <string.h>
   4
   5 #include "config.h"
   6 #include "mp_msg.h"
   7 #include "cpudetect.h"
   8
   9 #include "img_format.h"
  10 #include "mp_image.h"
  11 #include "vf.h"
  12
  13 #include "libvo/fastmemcpy.h"
  14
  15 struct vf_priv_s {
  16         int mode;
  17         int parity;
  18         int buffered_i;
  19         mp_image_t *buffered_mpi;
  20         double buffered_pts;
  21 };
  22
  23 static void deint(unsigned char *dest, int ds, unsigned char *src, int ss, int w, int h, int field)
  24 {
  25         int x, y;
  26         src += ss;
  27         dest += ds;
  28         if (field) {
  29                 src += ss;
  30                 dest += ds;
  31                 h -= 2;
  32         }
  33         for (y=h/2; y; y--) {
  34                 for (x=0; x<w; x++) {
  35                         if (((src[x-ss] < src[x]) && (src[x+ss] < src[x])) ||
  36                                 ((src[x-ss] > src[x]) && (src[x+ss] > src[x]))) {
  37                                 //dest[x] = (src[x+ss] + src[x-ss])>>1;
  38                                 dest[x] = ((src[x+ss]<<1) + (src[x-ss]<<1)
  39                                         + src[x+ss+1] + src[x-ss+1]
  40                                         + src[x+ss-1] + src[x-ss-1])>>3;
  41                         }
  42                         else dest[x] = src[x];
  43                 }
  44                 dest += ds<<1;
  45                 src += ss<<1;
  46         }
  47 }
  48
  49 #if HAVE_AMD3DNOW
  50 static void qpel_li_3DNOW(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
  51 {
  52         int i, j, ssd=ss;
  53         long crap1, crap2;
  54         if (up) {
  55                 ssd = -ss;
  56                 fast_memcpy(d, s, w);
  57                 d += ds;
  58                 s += ss;
  59         }
  60         for (i=h-1; i; i--) {
  61                 __asm__ volatile(
  62                         "1: \n\t"
  63                         "movq (%%"REG_S"), %%mm0 \n\t"
  64                         "movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
  65                         "pavgusb %%mm0, %%mm1 \n\t"
  66                         "add $8, %%"REG_S" \n\t"
  67                         "pavgusb %%mm0, %%mm1 \n\t"
  68                         "movq %%mm1, (%%"REG_D") \n\t"
  69                         "add $8, %%"REG_D" \n\t"
  70                         "decl %%ecx \n\t"
  71                         "jnz 1b \n\t"
  72                         : "=S"(crap1), "=D"(crap2)
  73                         : "c"(w>>3), "S"(s), "D"(d), "a"((long)ssd)
  74                 );
  75                 for (j=w-(w&7); j<w; j++)
  76                         d[j] = (s[j+ssd] + 3*s[j])>>2;
  77                 d += ds;
  78                 s += ss;
  79         }
  80         if (!up) fast_memcpy(d, s, w);
  81         __asm__ volatile("emms \n\t" : : : "memory");
  82 }
  83 #endif
  84
  85 #if HAVE_MMX2
  86 static void qpel_li_MMX2(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
  87 {
  88         int i, j, ssd=ss;
  89         long crap1, crap2;
  90         if (up) {
  91                 ssd = -ss;
  92                 fast_memcpy(d, s, w);
  93                 d += ds;
  94                 s += ss;
  95         }
  96         for (i=h-1; i; i--) {
  97                 __asm__ volatile(
  98                         "pxor %%mm7, %%mm7 \n\t"
  99                         "2: \n\t"
 100                         "movq (%%"REG_S"), %%mm0 \n\t"
 101                         "movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
 102                         "pavgb %%mm0, %%mm1 \n\t"
 103                         "add $8, %%"REG_S" \n\t"
 104                         "pavgb %%mm0, %%mm1 \n\t"
 105                         "movq %%mm1, (%%"REG_D") \n\t"
 106                         "add $8, %%"REG_D" \n\t"
 107                         "decl %%ecx \n\t"
 108                         "jnz 2b \n\t"
 109                         : "=S"(crap1), "=D"(crap2)
 110                         : "c"(w>>3), "S"(s), "D"(d), "a"((long)ssd)
 111                 );
 112                 for (j=w-(w&7); j<w; j++)
 113                         d[j] = (s[j+ssd] + 3*s[j])>>2;
 114                 d += ds;
 115                 s += ss;
 116         }
 117         if (!up) fast_memcpy(d, s, w);
 118         __asm__ volatile("emms \n\t" : : : "memory");
 119 }
 120 #endif
 121
 122 #if HAVE_MMX
 123 static void qpel_li_MMX(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
 124 {
 125         int i, j, ssd=ss;
 126         int crap1, crap2;
 127         if (up) {
 128                 ssd = -ss;
 129                 fast_memcpy(d, s, w);
 130                 d += ds;
 131                 s += ss;
 132         }
 133         for (i=h-1; i; i--) {
 134                 __asm__ volatile(
 135                         "pxor %%mm7, %%mm7 \n\t"
 136                         "3: \n\t"
 137                         "movq (%%"REG_S"), %%mm0 \n\t"
 138                         "movq (%%"REG_S"), %%mm1 \n\t"
 139                         "movq (%%"REG_S",%%"REG_a"), %%mm2 \n\t"
 140                         "movq (%%"REG_S",%%"REG_a"), %%mm3 \n\t"
 141                         "add $8, %%"REG_S" \n\t"
 142                         "punpcklbw %%mm7, %%mm0 \n\t"
 143                         "punpckhbw %%mm7, %%mm1 \n\t"
 144                         "punpcklbw %%mm7, %%mm2 \n\t"
 145                         "punpckhbw %%mm7, %%mm3 \n\t"
 146                         "paddw %%mm0, %%mm2 \n\t"
 147                         "paddw %%mm1, %%mm3 \n\t"
 148                         "paddw %%mm0, %%mm2 \n\t"
 149                         "paddw %%mm1, %%mm3 \n\t"
 150                         "paddw %%mm0, %%mm2 \n\t"
 151                         "paddw %%mm1, %%mm3 \n\t"
 152                         "psrlw $2, %%mm2 \n\t"
 153                         "psrlw $2, %%mm3 \n\t"
 154                         "packsswb %%mm3, %%mm2 \n\t"
 155                         "movq %%mm2, (%%"REG_D") \n\t"
 156                         "add $8, %%"REG_D" \n\t"
 157                         "decl %%ecx \n\t"
 158                         "jnz 3b \n\t"
 159                         : "=S"(crap1), "=D"(crap2)
 160                         : "c"(w>>3), "S"(s), "D"(d), "a"((long)ssd)
 161                 );
 162                 for (j=w-(w&7); j<w; j++)
 163                         d[j] = (s[j+ssd] + 3*s[j])>>2;
 164                 d += ds;
 165                 s += ss;
 166         }
 167         if (!up) fast_memcpy(d, s, w);
 168         __asm__ volatile("emms \n\t" : : : "memory");
 169 }
 170
 171 static void qpel_4tap_MMX(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
 172 {
 173         int i, j, ssd=ss;
 174         static const short filter[] = {
 175                 29, 29, 29, 29, 110, 110, 110, 110,
 176                 9, 9, 9, 9, 3, 3, 3, 3,
 177                 64, 64, 64, 64 };
 178         int crap1, crap2;
 179         if (up) {
 180                 ssd = -ss;
 181                 fast_memcpy(d, s, w);
 182                 d += ds; s += ss;
 183         }
 184         for (j=0; j<w; j++)
 185                 d[j] = (s[j+ssd] + 3*s[j])>>2;
 186         d += ds; s += ss;
 187         for (i=h-3; i; i--) {
 188                 __asm__ volatile(
 189                         "pxor %%mm0, %%mm0 \n\t"
 190                         "movq (%%"REG_d"), %%mm4 \n\t"
 191                         "movq 8(%%"REG_d"), %%mm5 \n\t"
 192                         "movq 16(%%"REG_d"), %%mm6 \n\t"
 193                         "movq 24(%%"REG_d"), %%mm7 \n\t"
 194                         "4: \n\t"
 195
 196                         "movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
 197                         "movq (%%"REG_S"), %%mm2 \n\t"
 198                         "movq (%%"REG_S",%%"REG_b"), %%mm3 \n\t"
 199                         "punpcklbw %%mm0, %%mm1 \n\t"
 200                         "punpcklbw %%mm0, %%mm2 \n\t"
 201                         "pmullw %%mm4, %%mm1 \n\t"
 202                         "punpcklbw %%mm0, %%mm3 \n\t"
 203                         "pmullw %%mm5, %%mm2 \n\t"
 204                         "paddusw %%mm2, %%mm1 \n\t"
 205                         "pmullw %%mm6, %%mm3 \n\t"
 206                         "movq (%%"REG_S",%%"REG_a",2), %%mm2 \n\t"
 207                         "psubusw %%mm3, %%mm1 \n\t"
 208                         "punpcklbw %%mm0, %%mm2 \n\t"
 209                         "pmullw %%mm7, %%mm2 \n\t"
 210                         "psubusw %%mm2, %%mm1 \n\t"
 211                         "psrlw $7, %%mm1 \n\t"
 212
 213                         "movq (%%"REG_S",%%"REG_a"), %%mm2 \n\t"
 214                         "movq (%%"REG_S"), %%mm3 \n\t"
 215                         "punpckhbw %%mm0, %%mm2 \n\t"
 216                         "punpckhbw %%mm0, %%mm3 \n\t"
 217                         "pmullw %%mm4, %%mm2 \n\t"
 218                         "pmullw %%mm5, %%mm3 \n\t"
 219                         "paddusw %%mm3, %%mm2 \n\t"
 220                         "movq (%%"REG_S",%%"REG_b"), %%mm3 \n\t"
 221                         "punpckhbw %%mm0, %%mm3 \n\t"
 222                         "pmullw %%mm6, %%mm3 \n\t"
 223                         "psubusw %%mm3, %%mm2 \n\t"
 224                         "movq (%%"REG_S",%%"REG_a",2), %%mm3 \n\t"
 225                         "punpckhbw %%mm0, %%mm3 \n\t"
 226                         "add $8, %%"REG_S" \n\t"
 227                         "pmullw %%mm7, %%mm3 \n\t"
 228                         "psubusw %%mm3, %%mm2 \n\t"
 229                         "psrlw $7, %%mm2 \n\t"
 230
 231                         "packuswb %%mm2, %%mm1 \n\t"
 232                         "movq %%mm1, (%%"REG_D") \n\t"
 233                         "add $8, %%"REG_D" \n\t"
 234                         "decl %%ecx \n\t"
 235                         "jnz 4b \n\t"
 236                         : "=S"(crap1), "=D"(crap2)
 237                         : "c"(w>>3), "S"(s), "D"(d), "a"((long)ssd), "b"((long)-ssd), "d"(filter)
 238                 );
 239                 for (j=w-(w&7); j<w; j++)
 240                         d[j] = (-9*s[j-ssd] + 111*s[j] + 29*s[j+ssd] - 3*s[j+ssd+ssd])>>7;
 241                 d += ds;
 242                 s += ss;
 243         }
 244         for (j=0; j<w; j++)
 245                 d[j] = (s[j+ssd] + 3*s[j])>>2;
 246         d += ds; s += ss;
 247         if (!up) fast_memcpy(d, s, w);
 248         __asm__ volatile("emms \n\t" : : : "memory");
 249 }
 250 #endif
 251
 252 static inline int clamp(int a)
 253 {
 254         // If a<512, this is equivalent to:
 255         // return (a<0) ? 0 : ( (a>255) ? 255 : a);
 256         return (~(a>>31)) & (a | ((a<<23)>>31));
 257 }
 258
 259 static void qpel_li_C(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
 260 {
 261         int i, j, ssd=ss;
 262         if (up) {
 263                 ssd = -ss;
 264                 fast_memcpy(d, s, w);
 265                 d += ds;
 266                 s += ss;
 267         }
 268         for (i=h-1; i; i--) {
 269                 for (j=0; j<w; j++)
 270                         d[j] = (s[j+ssd] + 3*s[j])>>2;
 271                 d += ds;
 272                 s += ss;
 273         }
 274         if (!up) fast_memcpy(d, s, w);
 275 }
 276
 277 static void qpel_4tap_C(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up)
 278 {
 279         int i, j, ssd=ss;
 280         if (up) {
 281                 ssd = -ss;
 282                 fast_memcpy(d, s, w);
 283                 d += ds; s += ss;
 284         }
 285         for (j=0; j<w; j++)
 286                 d[j] = (s[j+ssd] + 3*s[j] + 2)>>2;
 287         d += ds; s += ss;
 288         for (i=h-3; i; i--) {
 289                 for (j=0; j<w; j++)
 290                         d[j] = clamp((-9*s[j-ssd] + 111*s[j] + 29*s[j+ssd] - 3*s[j+ssd+ssd] + 64)>>7);
 291                 d += ds; s += ss;
 292         }
 293         for (j=0; j<w; j++)
 294                 d[j] = (s[j+ssd] + 3*s[j] + 2)>>2;
 295         d += ds; s += ss;
 296         if (!up) fast_memcpy(d, s, w);
 297 }
 298
 299 static void (*qpel_li)(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up);
 300 static void (*qpel_4tap)(unsigned char *d, unsigned char *s, int w, int h, int ds, int ss, int up);
 301
 302 static int continue_buffered_image(struct vf_instance_s *);
 303 extern int correct_pts;
 304
 305 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts)
 306 {
 307         vf->priv->buffered_mpi = mpi;
 308         vf->priv->buffered_pts = pts;
 309         vf->priv->buffered_i = 0;
 310         return continue_buffered_image(vf);
 311 }
 312
 313 static int continue_buffered_image(struct vf_instance_s *vf)
 314 {
 315         int i=vf->priv->buffered_i;
 316         double pts = vf->priv->buffered_pts;
 317         mp_image_t *mpi = vf->priv->buffered_mpi;
 318         int ret=0;
 319         mp_image_t *dmpi;
 320         void (*qpel)(unsigned char *, unsigned char *, int, int, int, int, int);
 321         int bpp=1;
 322         int tff;
 323
 324         if (i == 0)
 325                 vf_queue_frame(vf, continue_buffered_image);
 326         pts += i * .02;  // XXX not right
 327
 328         if (!(mpi->flags & MP_IMGFLAG_PLANAR)) bpp = mpi->bpp/8;
 329         if (vf->priv->parity < 0) {
 330                 if (mpi->fields & MP_IMGFIELD_ORDERED)
 331                         tff = mpi->fields & MP_IMGFIELD_TOP_FIRST;
 332                 else
 333                         tff = 1;
 334         }
 335         else tff = (vf->priv->parity&1)^1;
 336
 337         switch (vf->priv->mode) {
 338         case 2:
 339                 qpel = qpel_li;
 340                 break;
 341         case 3:
 342                 // TODO: add 3tap filter
 343                 qpel = qpel_4tap;
 344                 break;
 345         case 4:
 346                 qpel = qpel_4tap;
 347                 break;
 348         }
 349
 350         switch (vf->priv->mode) {
 351         case 0:
 352                 for (; i<2; i++) {
 353                         dmpi = vf_get_image(vf->next, mpi->imgfmt,
 354                                 MP_IMGTYPE_EXPORT, MP_IMGFLAG_ACCEPT_STRIDE,
 355                                 mpi->width, mpi->height/2);
 356                         dmpi->planes[0] = mpi->planes[0] + (i^!tff)*mpi->stride[0];
 357                         dmpi->stride[0] = 2*mpi->stride[0];
 358                         if (mpi->flags & MP_IMGFLAG_PLANAR) {
 359                                 dmpi->planes[1] = mpi->planes[1] + (i^!tff)*mpi->stride[1];
 360                                 dmpi->planes[2] = mpi->planes[2] + (i^!tff)*mpi->stride[2];
 361                                 dmpi->stride[1] = 2*mpi->stride[1];
 362                                 dmpi->stride[2] = 2*mpi->stride[2];
 363                         }
 364                         ret |= vf_next_put_image(vf, dmpi, pts);
 365                         if (correct_pts)
 366                                 break;
 367                         else
 368                                 if (!i) vf_extra_flip(vf);
 369                 }
 370                 break;
 371         case 1:
 372                 for (; i<2; i++) {
 373                         dmpi = vf_get_image(vf->next, mpi->imgfmt,
 374                                 MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE,
 375                                 mpi->width, mpi->height);
 376                         my_memcpy_pic(dmpi->planes[0] + (i^!tff)*dmpi->stride[0],
 377                                 mpi->planes[0] + (i^!tff)*mpi->stride[0],
 378                                 mpi->w*bpp, mpi->h/2, dmpi->stride[0]*2, mpi->stride[0]*2);
 379                         deint(dmpi->planes[0], dmpi->stride[0], mpi->planes[0], mpi->stride[0], mpi->w, mpi->h, (i^!tff));
 380                         if (mpi->flags & MP_IMGFLAG_PLANAR) {
 381                                 my_memcpy_pic(dmpi->planes[1] + (i^!tff)*dmpi->stride[1],
 382                                         mpi->planes[1] + (i^!tff)*mpi->stride[1],
 383                                         mpi->chroma_width, mpi->chroma_height/2,
 384                                         dmpi->stride[1]*2, mpi->stride[1]*2);
 385                                 my_memcpy_pic(dmpi->planes[2] + (i^!tff)*dmpi->stride[2],
 386                                         mpi->planes[2] + (i^!tff)*mpi->stride[2],
 387                                         mpi->chroma_width, mpi->chroma_height/2,
 388                                         dmpi->stride[2]*2, mpi->stride[2]*2);
 389                                 deint(dmpi->planes[1], dmpi->stride[1], mpi->planes[1], mpi->stride[1],
 390                                         mpi->chroma_width, mpi->chroma_height, (i^!tff));
 391                                 deint(dmpi->planes[2], dmpi->stride[2], mpi->planes[2], mpi->stride[2],
 392                                         mpi->chroma_width, mpi->chroma_height, (i^!tff));
 393                         }
 394                         ret |= vf_next_put_image(vf, dmpi, pts);
 395                         if (correct_pts)
 396                                 break;
 397                         else
 398                                 if (!i) vf_extra_flip(vf);
 399                 }
 400                 break;
 401         case 2:
 402         case 3:
 403         case 4:
 404                 for (; i<2; i++) {
 405                         dmpi = vf_get_image(vf->next, mpi->imgfmt,
 406                                 MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE,
 407                                 mpi->width, mpi->height/2);
 408                         qpel(dmpi->planes[0], mpi->planes[0] + (i^!tff)*mpi->stride[0],
 409                                 mpi->w*bpp, mpi->h/2, dmpi->stride[0], mpi->stride[0]*2, (i^!tff));
 410                         if (mpi->flags & MP_IMGFLAG_PLANAR) {
 411                                 qpel(dmpi->planes[1],
 412                                         mpi->planes[1] + (i^!tff)*mpi->stride[1],
 413                                         mpi->chroma_width, mpi->chroma_height/2,
 414                                         dmpi->stride[1], mpi->stride[1]*2, (i^!tff));
 415                                 qpel(dmpi->planes[2],
 416                                         mpi->planes[2] + (i^!tff)*mpi->stride[2],
 417                                         mpi->chroma_width, mpi->chroma_height/2,
 418                                         dmpi->stride[2], mpi->stride[2]*2, (i^!tff));
 419                         }
 420                         ret |= vf_next_put_image(vf, dmpi, pts);
 421                         if (correct_pts)
 422                                 break;
 423                         else
 424                                 if (!i) vf_extra_flip(vf);
 425                 }
 426                 break;
 427         }
 428         vf->priv->buffered_i = 1;
 429         return ret;
 430 }
 431
 432 #if 0
 433 static int query_format(struct vf_instance_s* vf, unsigned int fmt)
 434 {
 435         /* FIXME - figure out which other formats work */
 436         switch (fmt) {
 437         case IMGFMT_YV12:
 438         case IMGFMT_IYUV:
 439         case IMGFMT_I420:
 440                 return vf_next_query_format(vf, fmt);
 441         }
 442         return 0;
 443 }
 444 #endif
 445
 446 static int config(struct vf_instance_s* vf,
 447         int width, int height, int d_width, int d_height,
 448         unsigned int flags, unsigned int outfmt)
 449 {
 450         switch (vf->priv->mode) {
 451         case 0:
 452         case 2:
 453         case 3:
 454         case 4:
 455                 return vf_next_config(vf,width,height/2,d_width,d_height,flags,outfmt);
 456         case 1:
 457                 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
 458         }
 459         return 0;
 460 }
 461
 462 static void uninit(struct vf_instance_s* vf)
 463 {
 464         free(vf->priv);
 465 }
 466
 467 static int open(vf_instance_t *vf, char* args)
 468 {
 469         struct vf_priv_s *p;
 470         vf->config = config;
 471         vf->put_image = put_image;
 472         //vf->query_format = query_format;
 473         vf->uninit = uninit;
 474         vf->default_reqs = VFCAP_ACCEPT_STRIDE;
 475         vf->priv = p = calloc(1, sizeof(struct vf_priv_s));
 476         vf->priv->mode = 4;
 477         vf->priv->parity = -1;
 478         if (args) sscanf(args, "%d:%d", &vf->priv->mode, &vf->priv->parity);
 479         qpel_li = qpel_li_C;
 480         qpel_4tap = qpel_4tap_C;
 481 #if HAVE_MMX
 482         if(gCpuCaps.hasMMX) qpel_li = qpel_li_MMX;
 483         if(gCpuCaps.hasMMX) qpel_4tap = qpel_4tap_MMX;
 484 #endif
 485 #if HAVE_MMX2
 486         if(gCpuCaps.hasMMX2) qpel_li = qpel_li_MMX2;
 487 #endif
 488 #if HAVE_AMD3DNOW
 489         if(gCpuCaps.has3DNow) qpel_li = qpel_li_3DNOW;
 490 #endif
 491         return 1;
 492 }
 493
 494 const vf_info_t vf_info_tfields = {
 495     "temporal field separation",
 496     "tfields",
 497     "Rich Felker",
 498     "",
 499     open,
 500     NULL
 501 };
 502
 503