• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • File List
  • Globals

libswscale/swscale_altivec_template.c

Go to the documentation of this file.
00001 /*
00002  * AltiVec-enhanced yuv2yuvX
00003  *
00004  * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
00005  * based on the equivalent C code in swscale.c
00006  *
00007  * This file is part of FFmpeg.
00008  *
00009  * FFmpeg is free software; you can redistribute it and/or
00010  * modify it under the terms of the GNU Lesser General Public
00011  * License as published by the Free Software Foundation; either
00012  * version 2.1 of the License, or (at your option) any later version.
00013  *
00014  * FFmpeg is distributed in the hope that it will be useful,
00015  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00016  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00017  * Lesser General Public License for more details.
00018  *
00019  * You should have received a copy of the GNU Lesser General Public
00020  * License along with FFmpeg; if not, write to the Free Software
00021  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00022  */
00023 
00024 #define vzero vec_splat_s32(0)
00025 
00026 static inline void
00027 altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW) {
00028     register int i;
00029     vector unsigned int altivec_vectorShiftInt19 =
00030         vec_add(vec_splat_u32(10), vec_splat_u32(9));
00031     if ((unsigned long)dest % 16) {
00032         /* badly aligned store, we force store alignment */
00033         /* and will handle load misalignment on val w/ vec_perm */
00034         vector unsigned char perm1;
00035         vector signed int v1;
00036         for (i = 0 ; (i < dstW) &&
00037             (((unsigned long)dest + i) % 16) ; i++) {
00038                 int t = val[i] >> 19;
00039                 dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
00040         }
00041         perm1 = vec_lvsl(i << 2, val);
00042         v1 = vec_ld(i << 2, val);
00043         for ( ; i < (dstW - 15); i+=16) {
00044             int offset = i << 2;
00045             vector signed int v2 = vec_ld(offset + 16, val);
00046             vector signed int v3 = vec_ld(offset + 32, val);
00047             vector signed int v4 = vec_ld(offset + 48, val);
00048             vector signed int v5 = vec_ld(offset + 64, val);
00049             vector signed int v12 = vec_perm(v1, v2, perm1);
00050             vector signed int v23 = vec_perm(v2, v3, perm1);
00051             vector signed int v34 = vec_perm(v3, v4, perm1);
00052             vector signed int v45 = vec_perm(v4, v5, perm1);
00053 
00054             vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
00055             vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
00056             vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
00057             vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
00058             vector unsigned short vs1 = vec_packsu(vA, vB);
00059             vector unsigned short vs2 = vec_packsu(vC, vD);
00060             vector unsigned char vf = vec_packsu(vs1, vs2);
00061             vec_st(vf, i, dest);
00062             v1 = v5;
00063         }
00064     } else { // dest is properly aligned, great
00065         for (i = 0; i < (dstW - 15); i+=16) {
00066             int offset = i << 2;
00067             vector signed int v1 = vec_ld(offset, val);
00068             vector signed int v2 = vec_ld(offset + 16, val);
00069             vector signed int v3 = vec_ld(offset + 32, val);
00070             vector signed int v4 = vec_ld(offset + 48, val);
00071             vector signed int v5 = vec_sra(v1, altivec_vectorShiftInt19);
00072             vector signed int v6 = vec_sra(v2, altivec_vectorShiftInt19);
00073             vector signed int v7 = vec_sra(v3, altivec_vectorShiftInt19);
00074             vector signed int v8 = vec_sra(v4, altivec_vectorShiftInt19);
00075             vector unsigned short vs1 = vec_packsu(v5, v6);
00076             vector unsigned short vs2 = vec_packsu(v7, v8);
00077             vector unsigned char vf = vec_packsu(vs1, vs2);
00078             vec_st(vf, i, dest);
00079         }
00080     }
00081     for ( ; i < dstW ; i++) {
00082         int t = val[i] >> 19;
00083         dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
00084     }
00085 }
00086 
00087 static inline void
00088 yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
00089                       int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
00090                       uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
00091 {
00092     const vector signed int vini = {(1 << 18), (1 << 18), (1 << 18), (1 << 18)};
00093     register int i, j;
00094     {
00095         int __attribute__ ((aligned (16))) val[dstW];
00096 
00097         for (i = 0; i < (dstW -7); i+=4) {
00098             vec_st(vini, i << 2, val);
00099         }
00100         for (; i < dstW; i++) {
00101             val[i] = (1 << 18);
00102         }
00103 
00104         for (j = 0; j < lumFilterSize; j++) {
00105             vector signed short l1, vLumFilter = vec_ld(j << 1, lumFilter);
00106             vector unsigned char perm, perm0 = vec_lvsl(j << 1, lumFilter);
00107             vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
00108             vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
00109 
00110             perm = vec_lvsl(0, lumSrc[j]);
00111             l1 = vec_ld(0, lumSrc[j]);
00112 
00113             for (i = 0; i < (dstW - 7); i+=8) {
00114                 int offset = i << 2;
00115                 vector signed short l2 = vec_ld((i << 1) + 16, lumSrc[j]);
00116 
00117                 vector signed int v1 = vec_ld(offset, val);
00118                 vector signed int v2 = vec_ld(offset + 16, val);
00119 
00120                 vector signed short ls = vec_perm(l1, l2, perm); // lumSrc[j][i] ... lumSrc[j][i+7]
00121 
00122                 vector signed int i1 = vec_mule(vLumFilter, ls);
00123                 vector signed int i2 = vec_mulo(vLumFilter, ls);
00124 
00125                 vector signed int vf1 = vec_mergeh(i1, i2);
00126                 vector signed int vf2 = vec_mergel(i1, i2); // lumSrc[j][i] * lumFilter[j] ... lumSrc[j][i+7] * lumFilter[j]
00127 
00128                 vector signed int vo1 = vec_add(v1, vf1);
00129                 vector signed int vo2 = vec_add(v2, vf2);
00130 
00131                 vec_st(vo1, offset, val);
00132                 vec_st(vo2, offset + 16, val);
00133 
00134                 l1 = l2;
00135             }
00136             for ( ; i < dstW; i++) {
00137                 val[i] += lumSrc[j][i] * lumFilter[j];
00138             }
00139         }
00140         altivec_packIntArrayToCharArray(val, dest, dstW);
00141     }
00142     if (uDest != 0) {
00143         int  __attribute__ ((aligned (16))) u[chrDstW];
00144         int  __attribute__ ((aligned (16))) v[chrDstW];
00145 
00146         for (i = 0; i < (chrDstW -7); i+=4) {
00147             vec_st(vini, i << 2, u);
00148             vec_st(vini, i << 2, v);
00149         }
00150         for (; i < chrDstW; i++) {
00151             u[i] = (1 << 18);
00152             v[i] = (1 << 18);
00153         }
00154 
00155         for (j = 0; j < chrFilterSize; j++) {
00156             vector signed short l1, l1_V, vChrFilter = vec_ld(j << 1, chrFilter);
00157             vector unsigned char perm, perm0 = vec_lvsl(j << 1, chrFilter);
00158             vChrFilter = vec_perm(vChrFilter, vChrFilter, perm0);
00159             vChrFilter = vec_splat(vChrFilter, 0); // chrFilter[j] is loaded 8 times in vChrFilter
00160 
00161             perm = vec_lvsl(0, chrSrc[j]);
00162             l1 = vec_ld(0, chrSrc[j]);
00163             l1_V = vec_ld(2048 << 1, chrSrc[j]);
00164 
00165             for (i = 0; i < (chrDstW - 7); i+=8) {
00166                 int offset = i << 2;
00167                 vector signed short l2 = vec_ld((i << 1) + 16, chrSrc[j]);
00168                 vector signed short l2_V = vec_ld(((i + 2048) << 1) + 16, chrSrc[j]);
00169 
00170                 vector signed int v1 = vec_ld(offset, u);
00171                 vector signed int v2 = vec_ld(offset + 16, u);
00172                 vector signed int v1_V = vec_ld(offset, v);
00173                 vector signed int v2_V = vec_ld(offset + 16, v);
00174 
00175                 vector signed short ls = vec_perm(l1, l2, perm); // chrSrc[j][i] ... chrSrc[j][i+7]
00176                 vector signed short ls_V = vec_perm(l1_V, l2_V, perm); // chrSrc[j][i+2048] ... chrSrc[j][i+2055]
00177 
00178                 vector signed int i1 = vec_mule(vChrFilter, ls);
00179                 vector signed int i2 = vec_mulo(vChrFilter, ls);
00180                 vector signed int i1_V = vec_mule(vChrFilter, ls_V);
00181                 vector signed int i2_V = vec_mulo(vChrFilter, ls_V);
00182 
00183                 vector signed int vf1 = vec_mergeh(i1, i2);
00184                 vector signed int vf2 = vec_mergel(i1, i2); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
00185                 vector signed int vf1_V = vec_mergeh(i1_V, i2_V);
00186                 vector signed int vf2_V = vec_mergel(i1_V, i2_V); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
00187 
00188                 vector signed int vo1 = vec_add(v1, vf1);
00189                 vector signed int vo2 = vec_add(v2, vf2);
00190                 vector signed int vo1_V = vec_add(v1_V, vf1_V);
00191                 vector signed int vo2_V = vec_add(v2_V, vf2_V);
00192 
00193                 vec_st(vo1, offset, u);
00194                 vec_st(vo2, offset + 16, u);
00195                 vec_st(vo1_V, offset, v);
00196                 vec_st(vo2_V, offset + 16, v);
00197 
00198                 l1 = l2;
00199                 l1_V = l2_V;
00200             }
00201             for ( ; i < chrDstW; i++) {
00202                 u[i] += chrSrc[j][i] * chrFilter[j];
00203                 v[i] += chrSrc[j][i + 2048] * chrFilter[j];
00204             }
00205         }
00206         altivec_packIntArrayToCharArray(u, uDest, chrDstW);
00207         altivec_packIntArrayToCharArray(v, vDest, chrDstW);
00208     }
00209 }
00210 
00211 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) {
00212     register int i;
00213     int __attribute__ ((aligned (16))) tempo[4];
00214 
00215     if (filterSize % 4) {
00216         for (i=0; i<dstW; i++) {
00217             register int j;
00218             register int srcPos = filterPos[i];
00219             register int val = 0;
00220             for (j=0; j<filterSize; j++) {
00221                 val += ((int)src[srcPos + j])*filter[filterSize*i + j];
00222             }
00223             dst[i] = FFMIN(val>>7, (1<<15)-1);
00224         }
00225     }
00226     else
00227     switch (filterSize) {
00228     case 4:
00229     {
00230     for (i=0; i<dstW; i++) {
00231         register int srcPos = filterPos[i];
00232 
00233         vector unsigned char src_v0 = vec_ld(srcPos, src);
00234         vector unsigned char src_v1, src_vF;
00235         vector signed short src_v, filter_v;
00236         vector signed int val_vEven, val_s;
00237         if ((((int)src + srcPos)% 16) > 12) {
00238             src_v1 = vec_ld(srcPos + 16, src);
00239         }
00240         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00241 
00242         src_v = // vec_unpackh sign-extends...
00243             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00244         // now put our elements in the even slots
00245         src_v = vec_mergeh(src_v, (vector signed short)vzero);
00246 
00247         filter_v = vec_ld(i << 3, filter);
00248         // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
00249 
00250         // The neat trick: We only care for half the elements,
00251         // high or low depending on (i<<3)%16 (it's 0 or 8 here),
00252         // and we're going to use vec_mule, so we choose
00253         // carefully how to "unpack" the elements into the even slots.
00254         if ((i << 3) % 16)
00255             filter_v = vec_mergel(filter_v, (vector signed short)vzero);
00256         else
00257             filter_v = vec_mergeh(filter_v, (vector signed short)vzero);
00258 
00259         val_vEven = vec_mule(src_v, filter_v);
00260         val_s = vec_sums(val_vEven, vzero);
00261         vec_st(val_s, 0, tempo);
00262         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00263     }
00264     }
00265     break;
00266 
00267     case 8:
00268     {
00269     for (i=0; i<dstW; i++) {
00270         register int srcPos = filterPos[i];
00271 
00272         vector unsigned char src_v0 = vec_ld(srcPos, src);
00273         vector unsigned char src_v1, src_vF;
00274         vector signed short src_v, filter_v;
00275         vector signed int val_v, val_s;
00276         if ((((int)src + srcPos)% 16) > 8) {
00277             src_v1 = vec_ld(srcPos + 16, src);
00278         }
00279         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00280 
00281         src_v = // vec_unpackh sign-extends...
00282             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00283         filter_v = vec_ld(i << 4, filter);
00284         // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
00285 
00286         val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
00287         val_s = vec_sums(val_v, vzero);
00288         vec_st(val_s, 0, tempo);
00289         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00290     }
00291     }
00292     break;
00293 
00294     case 16:
00295     {
00296         for (i=0; i<dstW; i++) {
00297             register int srcPos = filterPos[i];
00298 
00299             vector unsigned char src_v0 = vec_ld(srcPos, src);
00300             vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
00301             vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00302 
00303             vector signed short src_vA = // vec_unpackh sign-extends...
00304                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00305             vector signed short src_vB = // vec_unpackh sign-extends...
00306                 (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
00307 
00308             vector signed short filter_v0 = vec_ld(i << 5, filter);
00309             vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
00310             // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)
00311 
00312             vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
00313             vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);
00314 
00315             vector signed int val_s = vec_sums(val_v, vzero);
00316 
00317             vec_st(val_s, 0, tempo);
00318             dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00319         }
00320     }
00321     break;
00322 
00323     default:
00324     {
00325     for (i=0; i<dstW; i++) {
00326         register int j;
00327         register int srcPos = filterPos[i];
00328 
00329         vector signed int val_s, val_v = (vector signed int)vzero;
00330         vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
00331         vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);
00332 
00333         vector unsigned char src_v0 = vec_ld(srcPos, src);
00334         vector unsigned char permS = vec_lvsl(srcPos, src);
00335 
00336         for (j = 0 ; j < filterSize - 15; j += 16) {
00337             vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
00338             vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
00339 
00340             vector signed short src_vA = // vec_unpackh sign-extends...
00341                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00342             vector signed short src_vB = // vec_unpackh sign-extends...
00343                 (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
00344 
00345             vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
00346             vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
00347             vector signed short filter_v0  = vec_perm(filter_v0R, filter_v1R, permF);
00348             vector signed short filter_v1  = vec_perm(filter_v1R, filter_v2R, permF);
00349 
00350             vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
00351             val_v = vec_msums(src_vB, filter_v1, val_acc);
00352 
00353             filter_v0R = filter_v2R;
00354             src_v0 = src_v1;
00355         }
00356 
00357         if (j < filterSize-7) {
00358             // loading src_v0 is useless, it's already done above
00359             //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
00360             vector unsigned char src_v1, src_vF;
00361             vector signed short src_v, filter_v1R, filter_v;
00362             if ((((int)src + srcPos)% 16) > 8) {
00363                 src_v1 = vec_ld(srcPos + j + 16, src);
00364             }
00365             src_vF = vec_perm(src_v0, src_v1, permS);
00366 
00367             src_v = // vec_unpackh sign-extends...
00368                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00369             // loading filter_v0R is useless, it's already done above
00370             //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
00371             filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
00372             filter_v = vec_perm(filter_v0R, filter_v1R, permF);
00373 
00374             val_v = vec_msums(src_v, filter_v, val_v);
00375         }
00376 
00377         val_s = vec_sums(val_v, vzero);
00378 
00379         vec_st(val_s, 0, tempo);
00380         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00381     }
00382 
00383     }
00384     }
00385 }
00386 
00387 static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
00388                                               int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
00389     uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
00390     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
00391     uint8_t *ysrc = src[0];
00392     uint8_t *usrc = src[1];
00393     uint8_t *vsrc = src[2];
00394     const int width = c->srcW;
00395     const int height = srcSliceH;
00396     const int lumStride = srcStride[0];
00397     const int chromStride = srcStride[1];
00398     const int dstStride = dstStride_a[0];
00399     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00400     const int vertLumPerChroma = 2;
00401     register unsigned int y;
00402 
00403     if (width&15) {
00404         yv12toyuy2(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
00405         return srcSliceH;
00406     }
00407 
00408     /* This code assumes:
00409 
00410     1) dst is 16 bytes-aligned
00411     2) dstStride is a multiple of 16
00412     3) width is a multiple of 16
00413     4) lum & chrom stride are multiples of 8
00414     */
00415 
00416     for (y=0; y<height; y++) {
00417         int i;
00418         for (i = 0; i < width - 31; i+= 32) {
00419             const unsigned int j = i >> 1;
00420             vector unsigned char v_yA = vec_ld(i, ysrc);
00421             vector unsigned char v_yB = vec_ld(i + 16, ysrc);
00422             vector unsigned char v_yC = vec_ld(i + 32, ysrc);
00423             vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
00424             vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
00425             vector unsigned char v_uA = vec_ld(j, usrc);
00426             vector unsigned char v_uB = vec_ld(j + 16, usrc);
00427             vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00428             vector unsigned char v_vA = vec_ld(j, vsrc);
00429             vector unsigned char v_vB = vec_ld(j + 16, vsrc);
00430             vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00431             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00432             vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
00433             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00434             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00435             vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
00436             vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
00437             vec_st(v_yuy2_0, (i << 1), dst);
00438             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00439             vec_st(v_yuy2_2, (i << 1) + 32, dst);
00440             vec_st(v_yuy2_3, (i << 1) + 48, dst);
00441         }
00442         if (i < width) {
00443             const unsigned int j = i >> 1;
00444             vector unsigned char v_y1 = vec_ld(i, ysrc);
00445             vector unsigned char v_u = vec_ld(j, usrc);
00446             vector unsigned char v_v = vec_ld(j, vsrc);
00447             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00448             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00449             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00450             vec_st(v_yuy2_0, (i << 1), dst);
00451             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00452         }
00453         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
00454             usrc += chromStride;
00455             vsrc += chromStride;
00456         }
00457         ysrc += lumStride;
00458         dst += dstStride;
00459     }
00460 
00461     return srcSliceH;
00462 }
00463 
00464 static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
00465                                               int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
00466     uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
00467     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
00468     uint8_t *ysrc = src[0];
00469     uint8_t *usrc = src[1];
00470     uint8_t *vsrc = src[2];
00471     const int width = c->srcW;
00472     const int height = srcSliceH;
00473     const int lumStride = srcStride[0];
00474     const int chromStride = srcStride[1];
00475     const int dstStride = dstStride_a[0];
00476     const int vertLumPerChroma = 2;
00477     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00478     register unsigned int y;
00479 
00480     if (width&15) {
00481         yv12touyvy(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
00482         return srcSliceH;
00483     }
00484 
00485     /* This code assumes:
00486 
00487     1) dst is 16 bytes-aligned
00488     2) dstStride is a multiple of 16
00489     3) width is a multiple of 16
00490     4) lum & chrom stride are multiples of 8
00491     */
00492 
00493     for (y=0; y<height; y++) {
00494         int i;
00495         for (i = 0; i < width - 31; i+= 32) {
00496             const unsigned int j = i >> 1;
00497             vector unsigned char v_yA = vec_ld(i, ysrc);
00498             vector unsigned char v_yB = vec_ld(i + 16, ysrc);
00499             vector unsigned char v_yC = vec_ld(i + 32, ysrc);
00500             vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
00501             vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
00502             vector unsigned char v_uA = vec_ld(j, usrc);
00503             vector unsigned char v_uB = vec_ld(j + 16, usrc);
00504             vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00505             vector unsigned char v_vA = vec_ld(j, vsrc);
00506             vector unsigned char v_vB = vec_ld(j + 16, vsrc);
00507             vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00508             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00509             vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
00510             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00511             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00512             vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
00513             vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
00514             vec_st(v_uyvy_0, (i << 1), dst);
00515             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00516             vec_st(v_uyvy_2, (i << 1) + 32, dst);
00517             vec_st(v_uyvy_3, (i << 1) + 48, dst);
00518         }
00519         if (i < width) {
00520             const unsigned int j = i >> 1;
00521             vector unsigned char v_y1 = vec_ld(i, ysrc);
00522             vector unsigned char v_u = vec_ld(j, usrc);
00523             vector unsigned char v_v = vec_ld(j, vsrc);
00524             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00525             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00526             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00527             vec_st(v_uyvy_0, (i << 1), dst);
00528             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00529         }
00530         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
00531             usrc += chromStride;
00532             vsrc += chromStride;
00533         }
00534         ysrc += lumStride;
00535         dst += dstStride;
00536     }
00537     return srcSliceH;
00538 }

Generated on Sat Feb 16 2013 09:23:15 for ffmpeg by  doxygen 1.7.1