FFmpeg: libswscale/ppc/swscale_altivec

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 {
00029     register int i;
00030     vector unsigned int altivec_vectorShiftInt19 =
00031         vec_add(vec_splat_u32(10), vec_splat_u32(9));
00032     if ((uintptr_t)dest % 16) {
00033         /* badly aligned store, we force store alignment */
00034         /* and will handle load misalignment on val w/ vec_perm */
00035         vector unsigned char perm1;
00036         vector signed int v1;
00037         for (i = 0 ; (i < dstW) &&
00038             (((uintptr_t)dest + i) % 16) ; i++) {
00039                 int t = val[i] >> 19;
00040                 dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
00041         }
00042         perm1 = vec_lvsl(i << 2, val);
00043         v1 = vec_ld(i << 2, val);
00044         for ( ; i < (dstW - 15); i+=16) {
00045             int offset = i << 2;
00046             vector signed int v2 = vec_ld(offset + 16, val);
00047             vector signed int v3 = vec_ld(offset + 32, val);
00048             vector signed int v4 = vec_ld(offset + 48, val);
00049             vector signed int v5 = vec_ld(offset + 64, val);
00050             vector signed int v12 = vec_perm(v1, v2, perm1);
00051             vector signed int v23 = vec_perm(v2, v3, perm1);
00052             vector signed int v34 = vec_perm(v3, v4, perm1);
00053             vector signed int v45 = vec_perm(v4, v5, perm1);
00054 
00055             vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
00056             vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
00057             vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
00058             vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
00059             vector unsigned short vs1 = vec_packsu(vA, vB);
00060             vector unsigned short vs2 = vec_packsu(vC, vD);
00061             vector unsigned char vf = vec_packsu(vs1, vs2);
00062             vec_st(vf, i, dest);
00063             v1 = v5;
00064         }
00065     } else { // dest is properly aligned, great
00066         for (i = 0; i < (dstW - 15); i+=16) {
00067             int offset = i << 2;
00068             vector signed int v1 = vec_ld(offset, val);
00069             vector signed int v2 = vec_ld(offset + 16, val);
00070             vector signed int v3 = vec_ld(offset + 32, val);
00071             vector signed int v4 = vec_ld(offset + 48, val);
00072             vector signed int v5 = vec_sra(v1, altivec_vectorShiftInt19);
00073             vector signed int v6 = vec_sra(v2, altivec_vectorShiftInt19);
00074             vector signed int v7 = vec_sra(v3, altivec_vectorShiftInt19);
00075             vector signed int v8 = vec_sra(v4, altivec_vectorShiftInt19);
00076             vector unsigned short vs1 = vec_packsu(v5, v6);
00077             vector unsigned short vs2 = vec_packsu(v7, v8);
00078             vector unsigned char vf = vec_packsu(vs1, vs2);
00079             vec_st(vf, i, dest);
00080         }
00081     }
00082     for ( ; i < dstW ; i++) {
00083         int t = val[i] >> 19;
00084         dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
00085     }
00086 }
00087 
00088 static inline void
00089 yuv2yuvX_altivec_real(const int16_t *lumFilter, const int16_t **lumSrc,
00090                       int lumFilterSize, const int16_t *chrFilter,
00091                       const int16_t **chrUSrc, const int16_t **chrVSrc,
00092                       int chrFilterSize, uint8_t *dest, uint8_t *uDest,
00093                       uint8_t *vDest, int dstW, int chrDstW)
00094 {
00095     const vector signed int vini = {(1 << 18), (1 << 18), (1 << 18), (1 << 18)};
00096     register int i, j;
00097     {
00098         DECLARE_ALIGNED(16, int, val)[dstW];
00099 
00100         for (i = 0; i < (dstW -7); i+=4) {
00101             vec_st(vini, i << 2, val);
00102         }
00103         for (; i < dstW; i++) {
00104             val[i] = (1 << 18);
00105         }
00106 
00107         for (j = 0; j < lumFilterSize; j++) {
00108             vector signed short l1, vLumFilter = vec_ld(j << 1, lumFilter);
00109             vector unsigned char perm, perm0 = vec_lvsl(j << 1, lumFilter);
00110             vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
00111             vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
00112 
00113             perm = vec_lvsl(0, lumSrc[j]);
00114             l1 = vec_ld(0, lumSrc[j]);
00115 
00116             for (i = 0; i < (dstW - 7); i+=8) {
00117                 int offset = i << 2;
00118                 vector signed short l2 = vec_ld((i << 1) + 16, lumSrc[j]);
00119 
00120                 vector signed int v1 = vec_ld(offset, val);
00121                 vector signed int v2 = vec_ld(offset + 16, val);
00122 
00123                 vector signed short ls = vec_perm(l1, l2, perm); // lumSrc[j][i] ... lumSrc[j][i+7]
00124 
00125                 vector signed int i1 = vec_mule(vLumFilter, ls);
00126                 vector signed int i2 = vec_mulo(vLumFilter, ls);
00127 
00128                 vector signed int vf1 = vec_mergeh(i1, i2);
00129                 vector signed int vf2 = vec_mergel(i1, i2); // lumSrc[j][i] * lumFilter[j] ... lumSrc[j][i+7] * lumFilter[j]
00130 
00131                 vector signed int vo1 = vec_add(v1, vf1);
00132                 vector signed int vo2 = vec_add(v2, vf2);
00133 
00134                 vec_st(vo1, offset, val);
00135                 vec_st(vo2, offset + 16, val);
00136 
00137                 l1 = l2;
00138             }
00139             for ( ; i < dstW; i++) {
00140                 val[i] += lumSrc[j][i] * lumFilter[j];
00141             }
00142         }
00143         altivec_packIntArrayToCharArray(val, dest, dstW);
00144     }
00145     if (uDest != 0) {
00146         DECLARE_ALIGNED(16, int, u)[chrDstW];
00147         DECLARE_ALIGNED(16, int, v)[chrDstW];
00148 
00149         for (i = 0; i < (chrDstW -7); i+=4) {
00150             vec_st(vini, i << 2, u);
00151             vec_st(vini, i << 2, v);
00152         }
00153         for (; i < chrDstW; i++) {
00154             u[i] = (1 << 18);
00155             v[i] = (1 << 18);
00156         }
00157 
00158         for (j = 0; j < chrFilterSize; j++) {
00159             vector signed short l1, l1_V, vChrFilter = vec_ld(j << 1, chrFilter);
00160             vector unsigned char perm, perm0 = vec_lvsl(j << 1, chrFilter);
00161             vChrFilter = vec_perm(vChrFilter, vChrFilter, perm0);
00162             vChrFilter = vec_splat(vChrFilter, 0); // chrFilter[j] is loaded 8 times in vChrFilter
00163 
00164             perm = vec_lvsl(0, chrUSrc[j]);
00165             l1 = vec_ld(0, chrUSrc[j]);
00166             l1_V = vec_ld(0, chrVSrc[j]);
00167 
00168             for (i = 0; i < (chrDstW - 7); i+=8) {
00169                 int offset = i << 2;
00170                 vector signed short l2 = vec_ld((i << 1) + 16, chrUSrc[j]);
00171                 vector signed short l2_V = vec_ld((i << 1) + 16, chrVSrc[j]);
00172 
00173                 vector signed int v1 = vec_ld(offset, u);
00174                 vector signed int v2 = vec_ld(offset + 16, u);
00175                 vector signed int v1_V = vec_ld(offset, v);
00176                 vector signed int v2_V = vec_ld(offset + 16, v);
00177 
00178                 vector signed short ls = vec_perm(l1, l2, perm); // chrUSrc[j][i] ... chrUSrc[j][i+7]
00179                 vector signed short ls_V = vec_perm(l1_V, l2_V, perm); // chrVSrc[j][i] ... chrVSrc[j][i]
00180 
00181                 vector signed int i1 = vec_mule(vChrFilter, ls);
00182                 vector signed int i2 = vec_mulo(vChrFilter, ls);
00183                 vector signed int i1_V = vec_mule(vChrFilter, ls_V);
00184                 vector signed int i2_V = vec_mulo(vChrFilter, ls_V);
00185 
00186                 vector signed int vf1 = vec_mergeh(i1, i2);
00187                 vector signed int vf2 = vec_mergel(i1, i2); // chrUSrc[j][i] * chrFilter[j] ... chrUSrc[j][i+7] * chrFilter[j]
00188                 vector signed int vf1_V = vec_mergeh(i1_V, i2_V);
00189                 vector signed int vf2_V = vec_mergel(i1_V, i2_V); // chrVSrc[j][i] * chrFilter[j] ... chrVSrc[j][i+7] * chrFilter[j]
00190 
00191                 vector signed int vo1 = vec_add(v1, vf1);
00192                 vector signed int vo2 = vec_add(v2, vf2);
00193                 vector signed int vo1_V = vec_add(v1_V, vf1_V);
00194                 vector signed int vo2_V = vec_add(v2_V, vf2_V);
00195 
00196                 vec_st(vo1, offset, u);
00197                 vec_st(vo2, offset + 16, u);
00198                 vec_st(vo1_V, offset, v);
00199                 vec_st(vo2_V, offset + 16, v);
00200 
00201                 l1 = l2;
00202                 l1_V = l2_V;
00203             }
00204             for ( ; i < chrDstW; i++) {
00205                 u[i] += chrUSrc[j][i] * chrFilter[j];
00206                 v[i] += chrVSrc[j][i] * chrFilter[j];
00207             }
00208         }
00209         altivec_packIntArrayToCharArray(u, uDest, chrDstW);
00210         altivec_packIntArrayToCharArray(v, vDest, chrDstW);
00211     }
00212 }
00213 
00214 static inline void hScale_altivec_real(int16_t *dst, int dstW,
00215                                        const uint8_t *src, int srcW,
00216                                        int xInc, const int16_t *filter,
00217                                        const int16_t *filterPos, int filterSize)
00218 {
00219     register int i;
00220     DECLARE_ALIGNED(16, int, tempo)[4];
00221 
00222     if (filterSize % 4) {
00223         for (i=0; i<dstW; i++) {
00224             register int j;
00225             register int srcPos = filterPos[i];
00226             register int val = 0;
00227             for (j=0; j<filterSize; j++) {
00228                 val += ((int)src[srcPos + j])*filter[filterSize*i + j];
00229             }
00230             dst[i] = FFMIN(val>>7, (1<<15)-1);
00231         }
00232     }
00233     else
00234     switch (filterSize) {
00235     case 4:
00236     {
00237     for (i=0; i<dstW; i++) {
00238         register int srcPos = filterPos[i];
00239 
00240         vector unsigned char src_v0 = vec_ld(srcPos, src);
00241         vector unsigned char src_v1, src_vF;
00242         vector signed short src_v, filter_v;
00243         vector signed int val_vEven, val_s;
00244         if ((((int)src + srcPos)% 16) > 12) {
00245             src_v1 = vec_ld(srcPos + 16, src);
00246         }
00247         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00248 
00249         src_v = // vec_unpackh sign-extends...
00250             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00251         // now put our elements in the even slots
00252         src_v = vec_mergeh(src_v, (vector signed short)vzero);
00253 
00254         filter_v = vec_ld(i << 3, filter);
00255         // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
00256 
00257         // The neat trick: We only care for half the elements,
00258         // high or low depending on (i<<3)%16 (it's 0 or 8 here),
00259         // and we're going to use vec_mule, so we choose
00260         // carefully how to "unpack" the elements into the even slots.
00261         if ((i << 3) % 16)
00262             filter_v = vec_mergel(filter_v, (vector signed short)vzero);
00263         else
00264             filter_v = vec_mergeh(filter_v, (vector signed short)vzero);
00265 
00266         val_vEven = vec_mule(src_v, filter_v);
00267         val_s = vec_sums(val_vEven, vzero);
00268         vec_st(val_s, 0, tempo);
00269         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00270     }
00271     }
00272     break;
00273 
00274     case 8:
00275     {
00276     for (i=0; i<dstW; i++) {
00277         register int srcPos = filterPos[i];
00278 
00279         vector unsigned char src_v0 = vec_ld(srcPos, src);
00280         vector unsigned char src_v1, src_vF;
00281         vector signed short src_v, filter_v;
00282         vector signed int val_v, val_s;
00283         if ((((int)src + srcPos)% 16) > 8) {
00284             src_v1 = vec_ld(srcPos + 16, src);
00285         }
00286         src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00287 
00288         src_v = // vec_unpackh sign-extends...
00289             (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00290         filter_v = vec_ld(i << 4, filter);
00291         // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
00292 
00293         val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
00294         val_s = vec_sums(val_v, vzero);
00295         vec_st(val_s, 0, tempo);
00296         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00297     }
00298     }
00299     break;
00300 
00301     case 16:
00302     {
00303         for (i=0; i<dstW; i++) {
00304             register int srcPos = filterPos[i];
00305 
00306             vector unsigned char src_v0 = vec_ld(srcPos, src);
00307             vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
00308             vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
00309 
00310             vector signed short src_vA = // vec_unpackh sign-extends...
00311                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00312             vector signed short src_vB = // vec_unpackh sign-extends...
00313                 (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
00314 
00315             vector signed short filter_v0 = vec_ld(i << 5, filter);
00316             vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
00317             // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)
00318 
00319             vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
00320             vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);
00321 
00322             vector signed int val_s = vec_sums(val_v, vzero);
00323 
00324             vec_st(val_s, 0, tempo);
00325             dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00326         }
00327     }
00328     break;
00329 
00330     default:
00331     {
00332     for (i=0; i<dstW; i++) {
00333         register int j;
00334         register int srcPos = filterPos[i];
00335 
00336         vector signed int val_s, val_v = (vector signed int)vzero;
00337         vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
00338         vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);
00339 
00340         vector unsigned char src_v0 = vec_ld(srcPos, src);
00341         vector unsigned char permS = vec_lvsl(srcPos, src);
00342 
00343         for (j = 0 ; j < filterSize - 15; j += 16) {
00344             vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
00345             vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
00346 
00347             vector signed short src_vA = // vec_unpackh sign-extends...
00348                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00349             vector signed short src_vB = // vec_unpackh sign-extends...
00350                 (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
00351 
00352             vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
00353             vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
00354             vector signed short filter_v0  = vec_perm(filter_v0R, filter_v1R, permF);
00355             vector signed short filter_v1  = vec_perm(filter_v1R, filter_v2R, permF);
00356 
00357             vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
00358             val_v = vec_msums(src_vB, filter_v1, val_acc);
00359 
00360             filter_v0R = filter_v2R;
00361             src_v0 = src_v1;
00362         }
00363 
00364         if (j < filterSize-7) {
00365             // loading src_v0 is useless, it's already done above
00366             //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
00367             vector unsigned char src_v1, src_vF;
00368             vector signed short src_v, filter_v1R, filter_v;
00369             if ((((int)src + srcPos)% 16) > 8) {
00370                 src_v1 = vec_ld(srcPos + j + 16, src);
00371             }
00372             src_vF = vec_perm(src_v0, src_v1, permS);
00373 
00374             src_v = // vec_unpackh sign-extends...
00375                 (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
00376             // loading filter_v0R is useless, it's already done above
00377             //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
00378             filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
00379             filter_v = vec_perm(filter_v0R, filter_v1R, permF);
00380 
00381             val_v = vec_msums(src_v, filter_v, val_v);
00382         }
00383 
00384         val_s = vec_sums(val_v, vzero);
00385 
00386         vec_st(val_s, 0, tempo);
00387         dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
00388     }
00389 
00390     }
00391     }
00392 }
00393 
00394 static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
00395                                               int srcSliceH, uint8_t* dstParam[], int dstStride_a[])
00396 {
00397     uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
00398     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
00399     const uint8_t *ysrc = src[0];
00400     const uint8_t *usrc = src[1];
00401     const uint8_t *vsrc = src[2];
00402     const int width = c->srcW;
00403     const int height = srcSliceH;
00404     const int lumStride = srcStride[0];
00405     const int chromStride = srcStride[1];
00406     const int dstStride = dstStride_a[0];
00407     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00408     const int vertLumPerChroma = 2;
00409     register unsigned int y;
00410 
00411     if (width&15) {
00412         yv12toyuy2(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
00413         return srcSliceH;
00414     }
00415 
00416     /* This code assumes:
00417 
00418     1) dst is 16 bytes-aligned
00419     2) dstStride is a multiple of 16
00420     3) width is a multiple of 16
00421     4) lum & chrom stride are multiples of 8
00422     */
00423 
00424     for (y=0; y<height; y++) {
00425         int i;
00426         for (i = 0; i < width - 31; i+= 32) {
00427             const unsigned int j = i >> 1;
00428             vector unsigned char v_yA = vec_ld(i, ysrc);
00429             vector unsigned char v_yB = vec_ld(i + 16, ysrc);
00430             vector unsigned char v_yC = vec_ld(i + 32, ysrc);
00431             vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
00432             vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
00433             vector unsigned char v_uA = vec_ld(j, usrc);
00434             vector unsigned char v_uB = vec_ld(j + 16, usrc);
00435             vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00436             vector unsigned char v_vA = vec_ld(j, vsrc);
00437             vector unsigned char v_vB = vec_ld(j + 16, vsrc);
00438             vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00439             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00440             vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
00441             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00442             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00443             vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
00444             vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
00445             vec_st(v_yuy2_0, (i << 1), dst);
00446             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00447             vec_st(v_yuy2_2, (i << 1) + 32, dst);
00448             vec_st(v_yuy2_3, (i << 1) + 48, dst);
00449         }
00450         if (i < width) {
00451             const unsigned int j = i >> 1;
00452             vector unsigned char v_y1 = vec_ld(i, ysrc);
00453             vector unsigned char v_u = vec_ld(j, usrc);
00454             vector unsigned char v_v = vec_ld(j, vsrc);
00455             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00456             vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
00457             vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
00458             vec_st(v_yuy2_0, (i << 1), dst);
00459             vec_st(v_yuy2_1, (i << 1) + 16, dst);
00460         }
00461         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
00462             usrc += chromStride;
00463             vsrc += chromStride;
00464         }
00465         ysrc += lumStride;
00466         dst += dstStride;
00467     }
00468 
00469     return srcSliceH;
00470 }
00471 
00472 static inline int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
00473                                               int srcSliceH, uint8_t* dstParam[], int dstStride_a[])
00474 {
00475     uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
00476     // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
00477     const uint8_t *ysrc = src[0];
00478     const uint8_t *usrc = src[1];
00479     const uint8_t *vsrc = src[2];
00480     const int width = c->srcW;
00481     const int height = srcSliceH;
00482     const int lumStride = srcStride[0];
00483     const int chromStride = srcStride[1];
00484     const int dstStride = dstStride_a[0];
00485     const int vertLumPerChroma = 2;
00486     const vector unsigned char yperm = vec_lvsl(0, ysrc);
00487     register unsigned int y;
00488 
00489     if (width&15) {
00490         yv12touyvy(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
00491         return srcSliceH;
00492     }
00493 
00494     /* This code assumes:
00495 
00496     1) dst is 16 bytes-aligned
00497     2) dstStride is a multiple of 16
00498     3) width is a multiple of 16
00499     4) lum & chrom stride are multiples of 8
00500     */
00501 
00502     for (y=0; y<height; y++) {
00503         int i;
00504         for (i = 0; i < width - 31; i+= 32) {
00505             const unsigned int j = i >> 1;
00506             vector unsigned char v_yA = vec_ld(i, ysrc);
00507             vector unsigned char v_yB = vec_ld(i + 16, ysrc);
00508             vector unsigned char v_yC = vec_ld(i + 32, ysrc);
00509             vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
00510             vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
00511             vector unsigned char v_uA = vec_ld(j, usrc);
00512             vector unsigned char v_uB = vec_ld(j + 16, usrc);
00513             vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
00514             vector unsigned char v_vA = vec_ld(j, vsrc);
00515             vector unsigned char v_vB = vec_ld(j + 16, vsrc);
00516             vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
00517             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00518             vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
00519             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00520             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00521             vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
00522             vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
00523             vec_st(v_uyvy_0, (i << 1), dst);
00524             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00525             vec_st(v_uyvy_2, (i << 1) + 32, dst);
00526             vec_st(v_uyvy_3, (i << 1) + 48, dst);
00527         }
00528         if (i < width) {
00529             const unsigned int j = i >> 1;
00530             vector unsigned char v_y1 = vec_ld(i, ysrc);
00531             vector unsigned char v_u = vec_ld(j, usrc);
00532             vector unsigned char v_v = vec_ld(j, vsrc);
00533             vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
00534             vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
00535             vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
00536             vec_st(v_uyvy_0, (i << 1), dst);
00537             vec_st(v_uyvy_1, (i << 1) + 16, dst);
00538         }
00539         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
00540             usrc += chromStride;
00541             vsrc += chromStride;
00542         }
00543         ysrc += lumStride;
00544         dst += dstStride;
00545     }
00546     return srcSliceH;
00547 }
libswscale/ppc/swscale_altivec_template.c
