libswscale/swscale_altivec_template.c

   1 /*
   2  * AltiVec-enhanced yuv2yuvX
   3  *
   4  * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
   5  * based on the equivalent C code in "postproc/swscale.c"
   6  *
   7  * This file is part of FFmpeg.
   8  *
   9  * FFmpeg is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * FFmpeg is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with FFmpeg; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #ifdef CONFIG_DARWIN
  25 #define AVV(x...) (x)
  26 #else
  27 #define AVV(x...) {x}
  28 #endif
  29
  30 #define vzero vec_splat_s32(0)
  31
  32 static inline void
  33 altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW) {
  34   register int i;
  35   vector unsigned int altivec_vectorShiftInt19 =
  36     vec_add(vec_splat_u32(10),vec_splat_u32(9));
  37   if ((unsigned long)dest % 16) {
  38     /* badly aligned store, we force store alignement */
  39     /* and will handle load misalignement on val w/ vec_perm */
  40     vector unsigned char perm1;
  41     vector signed int v1;
  42     for (i = 0 ; (i < dstW) &&
  43            (((unsigned long)dest + i) % 16) ; i++) {
  44       int t = val[i] >> 19;
  45       dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
  46     }
  47     perm1 = vec_lvsl(i << 2, val);
  48     v1 = vec_ld(i << 2, val);
  49     for ( ; i < (dstW - 15); i+=16) {
  50       int offset = i << 2;
  51       vector signed int v2 = vec_ld(offset + 16, val);
  52       vector signed int v3 = vec_ld(offset + 32, val);
  53       vector signed int v4 = vec_ld(offset + 48, val);
  54       vector signed int v5 = vec_ld(offset + 64, val);
  55       vector signed int v12 = vec_perm(v1,v2,perm1);
  56       vector signed int v23 = vec_perm(v2,v3,perm1);
  57       vector signed int v34 = vec_perm(v3,v4,perm1);
  58       vector signed int v45 = vec_perm(v4,v5,perm1);
  59
  60       vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
  61       vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
  62       vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
  63       vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
  64       vector unsigned short vs1 = vec_packsu(vA, vB);
  65       vector unsigned short vs2 = vec_packsu(vC, vD);
  66       vector unsigned char vf = vec_packsu(vs1, vs2);
  67       vec_st(vf, i, dest);
  68       v1 = v5;
  69     }
  70   } else { // dest is properly aligned, great
  71     for (i = 0; i < (dstW - 15); i+=16) {
  72       int offset = i << 2;
  73       vector signed int v1 = vec_ld(offset, val);
  74       vector signed int v2 = vec_ld(offset + 16, val);
  75       vector signed int v3 = vec_ld(offset + 32, val);
  76       vector signed int v4 = vec_ld(offset + 48, val);
  77       vector signed int v5 = vec_sra(v1, altivec_vectorShiftInt19);
  78       vector signed int v6 = vec_sra(v2, altivec_vectorShiftInt19);
  79       vector signed int v7 = vec_sra(v3, altivec_vectorShiftInt19);
  80       vector signed int v8 = vec_sra(v4, altivec_vectorShiftInt19);
  81       vector unsigned short vs1 = vec_packsu(v5, v6);
  82       vector unsigned short vs2 = vec_packsu(v7, v8);
  83       vector unsigned char vf = vec_packsu(vs1, vs2);
  84       vec_st(vf, i, dest);
  85     }
  86   }
  87   for ( ; i < dstW ; i++) {
  88     int t = val[i] >> 19;
  89     dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
  90   }
  91 }
  92
  93 static inline void
  94 yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
  95                       int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
  96                       uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
  97 {
  98   const vector signed int vini = {(1 << 18), (1 << 18), (1 << 18), (1 << 18)};
  99   register int i, j;
 100   {
 101     int __attribute__ ((aligned (16))) val[dstW];
 102
 103     for (i = 0; i < (dstW -7); i+=4) {
 104       vec_st(vini, i << 2, val);
 105     }
 106     for (; i < dstW; i++) {
 107       val[i] = (1 << 18);
 108     }
 109
 110     for (j = 0; j < lumFilterSize; j++) {
 111       vector signed short l1, vLumFilter = vec_ld(j << 1, lumFilter);
 112       vector unsigned char perm, perm0 = vec_lvsl(j << 1, lumFilter);
 113       vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
 114       vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
 115
 116       perm = vec_lvsl(0, lumSrc[j]);
 117       l1 = vec_ld(0, lumSrc[j]);
 118
 119       for (i = 0; i < (dstW - 7); i+=8) {
 120         int offset = i << 2;
 121         vector signed short l2 = vec_ld((i << 1) + 16, lumSrc[j]);
 122
 123         vector signed int v1 = vec_ld(offset, val);
 124         vector signed int v2 = vec_ld(offset + 16, val);
 125
 126         vector signed short ls = vec_perm(l1, l2, perm); // lumSrc[j][i] ... lumSrc[j][i+7]
 127
 128         vector signed int i1 = vec_mule(vLumFilter, ls);
 129         vector signed int i2 = vec_mulo(vLumFilter, ls);
 130
 131         vector signed int vf1 = vec_mergeh(i1, i2);
 132         vector signed int vf2 = vec_mergel(i1, i2); // lumSrc[j][i] * lumFilter[j] ... lumSrc[j][i+7] * lumFilter[j]
 133
 134         vector signed int vo1 = vec_add(v1, vf1);
 135         vector signed int vo2 = vec_add(v2, vf2);
 136
 137         vec_st(vo1, offset, val);
 138         vec_st(vo2, offset + 16, val);
 139
 140         l1 = l2;
 141       }
 142       for ( ; i < dstW; i++) {
 143         val[i] += lumSrc[j][i] * lumFilter[j];
 144       }
 145     }
 146     altivec_packIntArrayToCharArray(val,dest,dstW);
 147   }
 148   if (uDest != 0) {
 149     int  __attribute__ ((aligned (16))) u[chrDstW];
 150     int  __attribute__ ((aligned (16))) v[chrDstW];
 151
 152     for (i = 0; i < (chrDstW -7); i+=4) {
 153       vec_st(vini, i << 2, u);
 154       vec_st(vini, i << 2, v);
 155     }
 156     for (; i < chrDstW; i++) {
 157       u[i] = (1 << 18);
 158       v[i] = (1 << 18);
 159     }
 160
 161     for (j = 0; j < chrFilterSize; j++) {
 162       vector signed short l1, l1_V, vChrFilter = vec_ld(j << 1, chrFilter);
 163       vector unsigned char perm, perm0 = vec_lvsl(j << 1, chrFilter);
 164       vChrFilter = vec_perm(vChrFilter, vChrFilter, perm0);
 165       vChrFilter = vec_splat(vChrFilter, 0); // chrFilter[j] is loaded 8 times in vChrFilter
 166
 167       perm = vec_lvsl(0, chrSrc[j]);
 168       l1 = vec_ld(0, chrSrc[j]);
 169       l1_V = vec_ld(2048 << 1, chrSrc[j]);
 170
 171       for (i = 0; i < (chrDstW - 7); i+=8) {
 172         int offset = i << 2;
 173         vector signed short l2 = vec_ld((i << 1) + 16, chrSrc[j]);
 174         vector signed short l2_V = vec_ld(((i + 2048) << 1) + 16, chrSrc[j]);
 175
 176         vector signed int v1 = vec_ld(offset, u);
 177         vector signed int v2 = vec_ld(offset + 16, u);
 178         vector signed int v1_V = vec_ld(offset, v);
 179         vector signed int v2_V = vec_ld(offset + 16, v);
 180
 181         vector signed short ls = vec_perm(l1, l2, perm); // chrSrc[j][i] ... chrSrc[j][i+7]
 182         vector signed short ls_V = vec_perm(l1_V, l2_V, perm); // chrSrc[j][i+2048] ... chrSrc[j][i+2055]
 183
 184         vector signed int i1 = vec_mule(vChrFilter, ls);
 185         vector signed int i2 = vec_mulo(vChrFilter, ls);
 186         vector signed int i1_V = vec_mule(vChrFilter, ls_V);
 187         vector signed int i2_V = vec_mulo(vChrFilter, ls_V);
 188
 189         vector signed int vf1 = vec_mergeh(i1, i2);
 190         vector signed int vf2 = vec_mergel(i1, i2); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
 191         vector signed int vf1_V = vec_mergeh(i1_V, i2_V);
 192         vector signed int vf2_V = vec_mergel(i1_V, i2_V); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
 193
 194         vector signed int vo1 = vec_add(v1, vf1);
 195         vector signed int vo2 = vec_add(v2, vf2);
 196         vector signed int vo1_V = vec_add(v1_V, vf1_V);
 197         vector signed int vo2_V = vec_add(v2_V, vf2_V);
 198
 199         vec_st(vo1, offset, u);
 200         vec_st(vo2, offset + 16, u);
 201         vec_st(vo1_V, offset, v);
 202         vec_st(vo2_V, offset + 16, v);
 203
 204         l1 = l2;
 205         l1_V = l2_V;
 206       }
 207       for ( ; i < chrDstW; i++) {
 208         u[i] += chrSrc[j][i] * chrFilter[j];
 209         v[i] += chrSrc[j][i + 2048] * chrFilter[j];
 210       }
 211     }
 212     altivec_packIntArrayToCharArray(u,uDest,chrDstW);
 213     altivec_packIntArrayToCharArray(v,vDest,chrDstW);
 214   }
 215 }
 216
 217 static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int srcW, int xInc, int16_t *filter, int16_t *filterPos, int filterSize) {
 218   register int i;
 219   int __attribute__ ((aligned (16))) tempo[4];
 220
 221   if (filterSize % 4) {
 222     for(i=0; i<dstW; i++) {
 223       register int j;
 224       register int srcPos = filterPos[i];
 225       register int val = 0;
 226       for(j=0; j<filterSize; j++) {
 227         val += ((int)src[srcPos + j])*filter[filterSize*i + j];
 228       }
 229       dst[i] = FFMIN(FFMAX(0, val>>7), (1<<15)-1);
 230     }
 231   }
 232   else
 233   switch (filterSize) {
 234   case 4:
 235     {
 236       for(i=0; i<dstW; i++) {
 237         register int srcPos = filterPos[i];
 238
 239         vector unsigned char src_v0 = vec_ld(srcPos, src);
 240         vector unsigned char src_v1, src_vF;
 241         vector signed short src_v, filter_v;
 242         vector signed int val_vEven, val_s;
 243         if ((((int)src + srcPos)% 16) > 12) {
 244           src_v1 = vec_ld(srcPos + 16, src);
 245         }
 246         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
 247
 248         src_v = // vec_unpackh sign-extends...
 249           (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
 250         // now put our elements in the even slots
 251         src_v = vec_mergeh(src_v, (vector signed short)vzero);
 252
 253         filter_v = vec_ld(i << 3, filter);
 254         // the 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2)
 255
 256         // the neat trick : we only care for half the elements,
 257         // high or low depending on (i<<3)%16 (it's 0 or 8 here),
 258         // and we're going to use vec_mule, so we chose
 259         // carefully how to "unpack" the elements into the even slots
 260         if ((i << 3) % 16)
 261           filter_v = vec_mergel(filter_v,(vector signed short)vzero);
 262         else
 263           filter_v = vec_mergeh(filter_v,(vector signed short)vzero);
 264
 265         val_vEven = vec_mule(src_v, filter_v);
 266         val_s = vec_sums(val_vEven, vzero);
 267         vec_st(val_s, 0, tempo);
 268         dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
 269       }
 270     }
 271     break;
 272
 273   case 8:
 274     {
 275       for(i=0; i<dstW; i++) {
 276         register int srcPos = filterPos[i];
 277
 278         vector unsigned char src_v0 = vec_ld(srcPos, src);
 279         vector unsigned char src_v1, src_vF;
 280         vector signed short src_v, filter_v;
 281         vector signed int val_v, val_s;
 282         if ((((int)src + srcPos)% 16) > 8) {
 283           src_v1 = vec_ld(srcPos + 16, src);
 284         }
 285         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
 286
 287         src_v = // vec_unpackh sign-extends...
 288           (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
 289         filter_v = vec_ld(i << 4, filter);
 290         // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
 291
 292         val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
 293         val_s = vec_sums(val_v, vzero);
 294         vec_st(val_s, 0, tempo);
 295         dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
 296       }
 297     }
 298     break;
 299
 300   case 16:
 301     {
 302       for(i=0; i<dstW; i++) {
 303         register int srcPos = filterPos[i];
 304
 305         vector unsigned char src_v0 = vec_ld(srcPos, src);
 306         vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
 307         vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
 308
 309         vector signed short src_vA = // vec_unpackh sign-extends...
 310           (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
 311         vector signed short src_vB = // vec_unpackh sign-extends...
 312           (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
 313
 314         vector signed short filter_v0 = vec_ld(i << 5, filter);
 315         vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
 316         // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)
 317
 318         vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
 319         vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);
 320
 321         vector signed int val_s = vec_sums(val_v, vzero);
 322
 323         vec_st(val_s, 0, tempo);
 324         dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
 325       }
 326     }
 327     break;
 328
 329   default:
 330     {
 331       for(i=0; i<dstW; i++) {
 332         register int j;
 333         register int srcPos = filterPos[i];
 334
 335         vector signed int val_s, val_v = (vector signed int)vzero;
 336         vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
 337         vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);
 338
 339         vector unsigned char src_v0 = vec_ld(srcPos, src);
 340         vector unsigned char permS = vec_lvsl(srcPos, src);
 341
 342         for (j = 0 ; j < filterSize - 15; j += 16) {
 343           vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
 344           vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
 345
 346           vector signed short src_vA = // vec_unpackh sign-extends...
 347             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
 348           vector signed short src_vB = // vec_unpackh sign-extends...
 349             (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
 350
 351           vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
 352           vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
 353           vector signed short filter_v0 = vec_perm(filter_v0R, filter_v1R, permF);
 354           vector signed short filter_v1 = vec_perm(filter_v1R, filter_v2R, permF);
 355
 356           vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
 357           val_v = vec_msums(src_vB, filter_v1, val_acc);
 358
 359           filter_v0R = filter_v2R;
 360           src_v0 = src_v1;
 361         }
 362
 363         if (j < (filterSize-7)) {
 364           // loading src_v0 is useless, it's already done above
 365           //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
 366           vector unsigned char src_v1, src_vF;
 367           vector signed short src_v, filter_v1R, filter_v;
 368           if ((((int)src + srcPos)% 16) > 8) {
 369             src_v1 = vec_ld(srcPos + j + 16, src);
 370           }
 371           src_vF = vec_perm(src_v0, src_v1, permS);
 372
 373           src_v = // vec_unpackh sign-extends...
 374             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
 375           // loading filter_v0R is useless, it's already done above
 376           //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
 377           filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
 378           filter_v = vec_perm(filter_v0R, filter_v1R, permF);
 379
 380           val_v = vec_msums(src_v, filter_v, val_v);
 381         }
 382
 383         val_s = vec_sums(val_v, vzero);
 384
 385         vec_st(val_s, 0, tempo);
 386         dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
 387       }
 388
 389     }
 390   }
 391 }
 392
 393 static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 394      int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
 395   uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
 396   // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
 397   uint8_t *ysrc = src[0];
 398   uint8_t *usrc = src[1];
 399   uint8_t *vsrc = src[2];
 400   const int width = c->srcW;
 401   const int height = srcSliceH;
 402   const int lumStride = srcStride[0];
 403   const int chromStride = srcStride[1];
 404   const int dstStride = dstStride_a[0];
 405   const vector unsigned char yperm = vec_lvsl(0, ysrc);
 406   const int vertLumPerChroma = 2;
 407   register unsigned int y;
 408
 409   if(width&15){
 410     yv12toyuy2( ysrc, usrc, vsrc, dst,c->srcW,srcSliceH, lumStride, chromStride, dstStride);
 411     return srcSliceH;
 412   }
 413
 414   /* this code assume:
 415
 416   1) dst is 16 bytes-aligned
 417   2) dstStride is a multiple of 16
 418   3) width is a multiple of 16
 419   4) lum&chrom stride are multiple of 8
 420   */
 421
 422   for(y=0; y<height; y++)
 423     {
 424       int i;
 425       for (i = 0; i < width - 31; i+= 32) {
 426         const unsigned int j = i >> 1;
 427         vector unsigned char v_yA = vec_ld(i, ysrc);
 428         vector unsigned char v_yB = vec_ld(i + 16, ysrc);
 429         vector unsigned char v_yC = vec_ld(i + 32, ysrc);
 430         vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
 431         vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
 432         vector unsigned char v_uA = vec_ld(j, usrc);
 433         vector unsigned char v_uB = vec_ld(j + 16, usrc);
 434         vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
 435         vector unsigned char v_vA = vec_ld(j, vsrc);
 436         vector unsigned char v_vB = vec_ld(j + 16, vsrc);
 437         vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
 438         vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
 439         vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
 440         vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
 441         vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
 442         vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
 443         vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
 444         vec_st(v_yuy2_0, (i << 1), dst);
 445         vec_st(v_yuy2_1, (i << 1) + 16, dst);
 446         vec_st(v_yuy2_2, (i << 1) + 32, dst);
 447         vec_st(v_yuy2_3, (i << 1) + 48, dst);
 448       }
 449       if (i < width) {
 450         const unsigned int j = i >> 1;
 451         vector unsigned char v_y1 = vec_ld(i, ysrc);
 452         vector unsigned char v_u = vec_ld(j, usrc);
 453         vector unsigned char v_v = vec_ld(j, vsrc);
 454         vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
 455         vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
 456         vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
 457         vec_st(v_yuy2_0, (i << 1), dst);
 458         vec_st(v_yuy2_1, (i << 1) + 16, dst);
 459       }
 460       if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
 461         {
 462           usrc += chromStride;
 463           vsrc += chromStride;
 464         }
 465       ysrc += lumStride;
 466       dst += dstStride;
 467     }
 468
 469   return srcSliceH;
 470 }
 471
 472 static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
 473      int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
 474   uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
 475   // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
 476   uint8_t *ysrc = src[0];
 477   uint8_t *usrc = src[1];
 478   uint8_t *vsrc = src[2];
 479   const int width = c->srcW;
 480   const int height = srcSliceH;
 481   const int lumStride = srcStride[0];
 482   const int chromStride = srcStride[1];
 483   const int dstStride = dstStride_a[0];
 484   const int vertLumPerChroma = 2;
 485   const vector unsigned char yperm = vec_lvsl(0, ysrc);
 486   register unsigned int y;
 487
 488   if(width&15){
 489     yv12touyvy( ysrc, usrc, vsrc, dst,c->srcW,srcSliceH, lumStride, chromStride, dstStride);
 490     return srcSliceH;
 491   }
 492
 493   /* this code assume:
 494
 495   1) dst is 16 bytes-aligned
 496   2) dstStride is a multiple of 16
 497   3) width is a multiple of 16
 498   4) lum&chrom stride are multiple of 8
 499   */
 500
 501   for(y=0; y<height; y++)
 502     {
 503       int i;
 504       for (i = 0; i < width - 31; i+= 32) {
 505         const unsigned int j = i >> 1;
 506         vector unsigned char v_yA = vec_ld(i, ysrc);
 507         vector unsigned char v_yB = vec_ld(i + 16, ysrc);
 508         vector unsigned char v_yC = vec_ld(i + 32, ysrc);
 509         vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
 510         vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
 511         vector unsigned char v_uA = vec_ld(j, usrc);
 512         vector unsigned char v_uB = vec_ld(j + 16, usrc);
 513         vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
 514         vector unsigned char v_vA = vec_ld(j, vsrc);
 515         vector unsigned char v_vB = vec_ld(j + 16, vsrc);
 516         vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
 517         vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
 518         vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
 519         vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
 520         vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
 521         vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
 522         vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
 523         vec_st(v_uyvy_0, (i << 1), dst);
 524         vec_st(v_uyvy_1, (i << 1) + 16, dst);
 525         vec_st(v_uyvy_2, (i << 1) + 32, dst);
 526         vec_st(v_uyvy_3, (i << 1) + 48, dst);
 527       }
 528       if (i < width) {
 529         const unsigned int j = i >> 1;
 530         vector unsigned char v_y1 = vec_ld(i, ysrc);
 531         vector unsigned char v_u = vec_ld(j, usrc);
 532         vector unsigned char v_v = vec_ld(j, vsrc);
 533         vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
 534         vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
 535         vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
 536         vec_st(v_uyvy_0, (i << 1), dst);
 537         vec_st(v_uyvy_1, (i << 1) + 16, dst);
 538       }
 539       if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
 540         {
 541           usrc += chromStride;
 542           vsrc += chromStride;
 543         }
 544       ysrc += lumStride;
 545       dst += dstStride;
 546     }
 547   return srcSliceH;
 548 }