Statistics
| Branch: | Revision:

ffmpeg / libswscale / swscale_altivec_template.c @ 3164d25e

History | View | Annotate | Download (22 KB)

1
/*
2
 * AltiVec-enhanced yuv2yuvX
3
 *
4
 * Copyright (C) 2004 Romain Dolbeau <romain@dolbeau.org>
5
 * based on the equivalent C code in 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 Street, Fifth Floor, Boston, MA 02110-1301 USA
22
 */
23

    
24
#define vzero vec_splat_s32(0)
25

    
26
static inline void
27
altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW) {
28
    register int i;
29
    vector unsigned int altivec_vectorShiftInt19 =
30
        vec_add(vec_splat_u32(10), vec_splat_u32(9));
31
    if ((unsigned long)dest % 16) {
32
        /* badly aligned store, we force store alignment */
33
        /* and will handle load misalignment on val w/ vec_perm */
34
        vector unsigned char perm1;
35
        vector signed int v1;
36
        for (i = 0 ; (i < dstW) &&
37
            (((unsigned long)dest + i) % 16) ; i++) {
38
                int t = val[i] >> 19;
39
                dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
40
        }
41
        perm1 = vec_lvsl(i << 2, val);
42
        v1 = vec_ld(i << 2, val);
43
        for ( ; i < (dstW - 15); i+=16) {
44
            int offset = i << 2;
45
            vector signed int v2 = vec_ld(offset + 16, val);
46
            vector signed int v3 = vec_ld(offset + 32, val);
47
            vector signed int v4 = vec_ld(offset + 48, val);
48
            vector signed int v5 = vec_ld(offset + 64, val);
49
            vector signed int v12 = vec_perm(v1, v2, perm1);
50
            vector signed int v23 = vec_perm(v2, v3, perm1);
51
            vector signed int v34 = vec_perm(v3, v4, perm1);
52
            vector signed int v45 = vec_perm(v4, v5, perm1);
53

    
54
            vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
55
            vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
56
            vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
57
            vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
58
            vector unsigned short vs1 = vec_packsu(vA, vB);
59
            vector unsigned short vs2 = vec_packsu(vC, vD);
60
            vector unsigned char vf = vec_packsu(vs1, vs2);
61
            vec_st(vf, i, dest);
62
            v1 = v5;
63
        }
64
    } else { // dest is properly aligned, great
65
        for (i = 0; i < (dstW - 15); i+=16) {
66
            int offset = i << 2;
67
            vector signed int v1 = vec_ld(offset, val);
68
            vector signed int v2 = vec_ld(offset + 16, val);
69
            vector signed int v3 = vec_ld(offset + 32, val);
70
            vector signed int v4 = vec_ld(offset + 48, val);
71
            vector signed int v5 = vec_sra(v1, altivec_vectorShiftInt19);
72
            vector signed int v6 = vec_sra(v2, altivec_vectorShiftInt19);
73
            vector signed int v7 = vec_sra(v3, altivec_vectorShiftInt19);
74
            vector signed int v8 = vec_sra(v4, altivec_vectorShiftInt19);
75
            vector unsigned short vs1 = vec_packsu(v5, v6);
76
            vector unsigned short vs2 = vec_packsu(v7, v8);
77
            vector unsigned char vf = vec_packsu(vs1, vs2);
78
            vec_st(vf, i, dest);
79
        }
80
    }
81
    for ( ; i < dstW ; i++) {
82
        int t = val[i] >> 19;
83
        dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
84
    }
85
}
86

    
87
static inline void
88
yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
89
                      int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
90
                      uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
91
{
92
    const vector signed int vini = {(1 << 18), (1 << 18), (1 << 18), (1 << 18)};
93
    register int i, j;
94
    {
95
        int __attribute__ ((aligned (16))) val[dstW];
96

    
97
        for (i = 0; i < (dstW -7); i+=4) {
98
            vec_st(vini, i << 2, val);
99
        }
100
        for (; i < dstW; i++) {
101
            val[i] = (1 << 18);
102
        }
103

    
104
        for (j = 0; j < lumFilterSize; j++) {
105
            vector signed short l1, vLumFilter = vec_ld(j << 1, lumFilter);
106
            vector unsigned char perm, perm0 = vec_lvsl(j << 1, lumFilter);
107
            vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
108
            vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
109

    
110
            perm = vec_lvsl(0, lumSrc[j]);
111
            l1 = vec_ld(0, lumSrc[j]);
112

    
113
            for (i = 0; i < (dstW - 7); i+=8) {
114
                int offset = i << 2;
115
                vector signed short l2 = vec_ld((i << 1) + 16, lumSrc[j]);
116

    
117
                vector signed int v1 = vec_ld(offset, val);
118
                vector signed int v2 = vec_ld(offset + 16, val);
119

    
120
                vector signed short ls = vec_perm(l1, l2, perm); // lumSrc[j][i] ... lumSrc[j][i+7]
121

    
122
                vector signed int i1 = vec_mule(vLumFilter, ls);
123
                vector signed int i2 = vec_mulo(vLumFilter, ls);
124

    
125
                vector signed int vf1 = vec_mergeh(i1, i2);
126
                vector signed int vf2 = vec_mergel(i1, i2); // lumSrc[j][i] * lumFilter[j] ... lumSrc[j][i+7] * lumFilter[j]
127

    
128
                vector signed int vo1 = vec_add(v1, vf1);
129
                vector signed int vo2 = vec_add(v2, vf2);
130

    
131
                vec_st(vo1, offset, val);
132
                vec_st(vo2, offset + 16, val);
133

    
134
                l1 = l2;
135
            }
136
            for ( ; i < dstW; i++) {
137
                val[i] += lumSrc[j][i] * lumFilter[j];
138
            }
139
        }
140
        altivec_packIntArrayToCharArray(val, dest, dstW);
141
    }
142
    if (uDest != 0) {
143
        int  __attribute__ ((aligned (16))) u[chrDstW];
144
        int  __attribute__ ((aligned (16))) v[chrDstW];
145

    
146
        for (i = 0; i < (chrDstW -7); i+=4) {
147
            vec_st(vini, i << 2, u);
148
            vec_st(vini, i << 2, v);
149
        }
150
        for (; i < chrDstW; i++) {
151
            u[i] = (1 << 18);
152
            v[i] = (1 << 18);
153
        }
154

    
155
        for (j = 0; j < chrFilterSize; j++) {
156
            vector signed short l1, l1_V, vChrFilter = vec_ld(j << 1, chrFilter);
157
            vector unsigned char perm, perm0 = vec_lvsl(j << 1, chrFilter);
158
            vChrFilter = vec_perm(vChrFilter, vChrFilter, perm0);
159
            vChrFilter = vec_splat(vChrFilter, 0); // chrFilter[j] is loaded 8 times in vChrFilter
160

    
161
            perm = vec_lvsl(0, chrSrc[j]);
162
            l1 = vec_ld(0, chrSrc[j]);
163
            l1_V = vec_ld(2048 << 1, chrSrc[j]);
164

    
165
            for (i = 0; i < (chrDstW - 7); i+=8) {
166
                int offset = i << 2;
167
                vector signed short l2 = vec_ld((i << 1) + 16, chrSrc[j]);
168
                vector signed short l2_V = vec_ld(((i + 2048) << 1) + 16, chrSrc[j]);
169

    
170
                vector signed int v1 = vec_ld(offset, u);
171
                vector signed int v2 = vec_ld(offset + 16, u);
172
                vector signed int v1_V = vec_ld(offset, v);
173
                vector signed int v2_V = vec_ld(offset + 16, v);
174

    
175
                vector signed short ls = vec_perm(l1, l2, perm); // chrSrc[j][i] ... chrSrc[j][i+7]
176
                vector signed short ls_V = vec_perm(l1_V, l2_V, perm); // chrSrc[j][i+2048] ... chrSrc[j][i+2055]
177

    
178
                vector signed int i1 = vec_mule(vChrFilter, ls);
179
                vector signed int i2 = vec_mulo(vChrFilter, ls);
180
                vector signed int i1_V = vec_mule(vChrFilter, ls_V);
181
                vector signed int i2_V = vec_mulo(vChrFilter, ls_V);
182

    
183
                vector signed int vf1 = vec_mergeh(i1, i2);
184
                vector signed int vf2 = vec_mergel(i1, i2); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
185
                vector signed int vf1_V = vec_mergeh(i1_V, i2_V);
186
                vector signed int vf2_V = vec_mergel(i1_V, i2_V); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
187

    
188
                vector signed int vo1 = vec_add(v1, vf1);
189
                vector signed int vo2 = vec_add(v2, vf2);
190
                vector signed int vo1_V = vec_add(v1_V, vf1_V);
191
                vector signed int vo2_V = vec_add(v2_V, vf2_V);
192

    
193
                vec_st(vo1, offset, u);
194
                vec_st(vo2, offset + 16, u);
195
                vec_st(vo1_V, offset, v);
196
                vec_st(vo2_V, offset + 16, v);
197

    
198
                l1 = l2;
199
                l1_V = l2_V;
200
            }
201
            for ( ; i < chrDstW; i++) {
202
                u[i] += chrSrc[j][i] * chrFilter[j];
203
                v[i] += chrSrc[j][i + 2048] * chrFilter[j];
204
            }
205
        }
206
        altivec_packIntArrayToCharArray(u, uDest, chrDstW);
207
        altivec_packIntArrayToCharArray(v, vDest, chrDstW);
208
    }
209
}
210

    
211
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) {
212
    register int i;
213
    int __attribute__ ((aligned (16))) tempo[4];
214

    
215
    if (filterSize % 4) {
216
        for (i=0; i<dstW; i++) {
217
            register int j;
218
            register int srcPos = filterPos[i];
219
            register int val = 0;
220
            for (j=0; j<filterSize; j++) {
221
                val += ((int)src[srcPos + j])*filter[filterSize*i + j];
222
            }
223
            dst[i] = FFMIN(val>>7, (1<<15)-1);
224
        }
225
    }
226
    else
227
    switch (filterSize) {
228
    case 4:
229
    {
230
    for (i=0; i<dstW; i++) {
231
        register int srcPos = filterPos[i];
232

    
233
        vector unsigned char src_v0 = vec_ld(srcPos, src);
234
        vector unsigned char src_v1, src_vF;
235
        vector signed short src_v, filter_v;
236
        vector signed int val_vEven, val_s;
237
        if ((((int)src + srcPos)% 16) > 12) {
238
            src_v1 = vec_ld(srcPos + 16, src);
239
        }
240
        src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
241

    
242
        src_v = // vec_unpackh sign-extends...
243
            (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
244
        // now put our elements in the even slots
245
        src_v = vec_mergeh(src_v, (vector signed short)vzero);
246

    
247
        filter_v = vec_ld(i << 3, filter);
248
        // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
249

    
250
        // The neat trick: We only care for half the elements,
251
        // high or low depending on (i<<3)%16 (it's 0 or 8 here),
252
        // and we're going to use vec_mule, so we choose
253
        // carefully how to "unpack" the elements into the even slots.
254
        if ((i << 3) % 16)
255
            filter_v = vec_mergel(filter_v, (vector signed short)vzero);
256
        else
257
            filter_v = vec_mergeh(filter_v, (vector signed short)vzero);
258

    
259
        val_vEven = vec_mule(src_v, filter_v);
260
        val_s = vec_sums(val_vEven, vzero);
261
        vec_st(val_s, 0, tempo);
262
        dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
263
    }
264
    }
265
    break;
266

    
267
    case 8:
268
    {
269
    for (i=0; i<dstW; i++) {
270
        register int srcPos = filterPos[i];
271

    
272
        vector unsigned char src_v0 = vec_ld(srcPos, src);
273
        vector unsigned char src_v1, src_vF;
274
        vector signed short src_v, filter_v;
275
        vector signed int val_v, val_s;
276
        if ((((int)src + srcPos)% 16) > 8) {
277
            src_v1 = vec_ld(srcPos + 16, src);
278
        }
279
        src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
280

    
281
        src_v = // vec_unpackh sign-extends...
282
            (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
283
        filter_v = vec_ld(i << 4, filter);
284
        // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
285

    
286
        val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
287
        val_s = vec_sums(val_v, vzero);
288
        vec_st(val_s, 0, tempo);
289
        dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
290
    }
291
    }
292
    break;
293

    
294
    case 16:
295
    {
296
        for (i=0; i<dstW; i++) {
297
            register int srcPos = filterPos[i];
298

    
299
            vector unsigned char src_v0 = vec_ld(srcPos, src);
300
            vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
301
            vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
302

    
303
            vector signed short src_vA = // vec_unpackh sign-extends...
304
                (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
305
            vector signed short src_vB = // vec_unpackh sign-extends...
306
                (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
307

    
308
            vector signed short filter_v0 = vec_ld(i << 5, filter);
309
            vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
310
            // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)
311

    
312
            vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
313
            vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);
314

    
315
            vector signed int val_s = vec_sums(val_v, vzero);
316

    
317
            vec_st(val_s, 0, tempo);
318
            dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
319
        }
320
    }
321
    break;
322

    
323
    default:
324
    {
325
    for (i=0; i<dstW; i++) {
326
        register int j;
327
        register int srcPos = filterPos[i];
328

    
329
        vector signed int val_s, val_v = (vector signed int)vzero;
330
        vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
331
        vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);
332

    
333
        vector unsigned char src_v0 = vec_ld(srcPos, src);
334
        vector unsigned char permS = vec_lvsl(srcPos, src);
335

    
336
        for (j = 0 ; j < filterSize - 15; j += 16) {
337
            vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
338
            vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
339

    
340
            vector signed short src_vA = // vec_unpackh sign-extends...
341
                (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
342
            vector signed short src_vB = // vec_unpackh sign-extends...
343
                (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
344

    
345
            vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
346
            vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
347
            vector signed short filter_v0  = vec_perm(filter_v0R, filter_v1R, permF);
348
            vector signed short filter_v1  = vec_perm(filter_v1R, filter_v2R, permF);
349

    
350
            vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
351
            val_v = vec_msums(src_vB, filter_v1, val_acc);
352

    
353
            filter_v0R = filter_v2R;
354
            src_v0 = src_v1;
355
        }
356

    
357
        if (j < filterSize-7) {
358
            // loading src_v0 is useless, it's already done above
359
            //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
360
            vector unsigned char src_v1, src_vF;
361
            vector signed short src_v, filter_v1R, filter_v;
362
            if ((((int)src + srcPos)% 16) > 8) {
363
                src_v1 = vec_ld(srcPos + j + 16, src);
364
            }
365
            src_vF = vec_perm(src_v0, src_v1, permS);
366

    
367
            src_v = // vec_unpackh sign-extends...
368
                (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
369
            // loading filter_v0R is useless, it's already done above
370
            //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
371
            filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
372
            filter_v = vec_perm(filter_v0R, filter_v1R, permF);
373

    
374
            val_v = vec_msums(src_v, filter_v, val_v);
375
        }
376

    
377
        val_s = vec_sums(val_v, vzero);
378

    
379
        vec_st(val_s, 0, tempo);
380
        dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
381
    }
382

    
383
    }
384
    }
385
}
386

    
387
static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
388
                                              int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
389
    uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
390
    // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
391
    uint8_t *ysrc = src[0];
392
    uint8_t *usrc = src[1];
393
    uint8_t *vsrc = src[2];
394
    const int width = c->srcW;
395
    const int height = srcSliceH;
396
    const int lumStride = srcStride[0];
397
    const int chromStride = srcStride[1];
398
    const int dstStride = dstStride_a[0];
399
    const vector unsigned char yperm = vec_lvsl(0, ysrc);
400
    const int vertLumPerChroma = 2;
401
    register unsigned int y;
402

    
403
    if (width&15) {
404
        yv12toyuy2(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
405
        return srcSliceH;
406
    }
407

    
408
    /* This code assumes:
409

410
    1) dst is 16 bytes-aligned
411
    2) dstStride is a multiple of 16
412
    3) width is a multiple of 16
413
    4) lum & chrom stride are multiples of 8
414
    */
415

    
416
    for (y=0; y<height; y++) {
417
        int i;
418
        for (i = 0; i < width - 31; i+= 32) {
419
            const unsigned int j = i >> 1;
420
            vector unsigned char v_yA = vec_ld(i, ysrc);
421
            vector unsigned char v_yB = vec_ld(i + 16, ysrc);
422
            vector unsigned char v_yC = vec_ld(i + 32, ysrc);
423
            vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
424
            vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
425
            vector unsigned char v_uA = vec_ld(j, usrc);
426
            vector unsigned char v_uB = vec_ld(j + 16, usrc);
427
            vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
428
            vector unsigned char v_vA = vec_ld(j, vsrc);
429
            vector unsigned char v_vB = vec_ld(j + 16, vsrc);
430
            vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
431
            vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
432
            vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
433
            vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
434
            vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
435
            vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
436
            vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
437
            vec_st(v_yuy2_0, (i << 1), dst);
438
            vec_st(v_yuy2_1, (i << 1) + 16, dst);
439
            vec_st(v_yuy2_2, (i << 1) + 32, dst);
440
            vec_st(v_yuy2_3, (i << 1) + 48, dst);
441
        }
442
        if (i < width) {
443
            const unsigned int j = i >> 1;
444
            vector unsigned char v_y1 = vec_ld(i, ysrc);
445
            vector unsigned char v_u = vec_ld(j, usrc);
446
            vector unsigned char v_v = vec_ld(j, vsrc);
447
            vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
448
            vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
449
            vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
450
            vec_st(v_yuy2_0, (i << 1), dst);
451
            vec_st(v_yuy2_1, (i << 1) + 16, dst);
452
        }
453
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
454
            usrc += chromStride;
455
            vsrc += chromStride;
456
        }
457
        ysrc += lumStride;
458
        dst += dstStride;
459
    }
460

    
461
    return srcSliceH;
462
}
463

    
464
static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
465
                                              int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
466
    uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
467
    // yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
468
    uint8_t *ysrc = src[0];
469
    uint8_t *usrc = src[1];
470
    uint8_t *vsrc = src[2];
471
    const int width = c->srcW;
472
    const int height = srcSliceH;
473
    const int lumStride = srcStride[0];
474
    const int chromStride = srcStride[1];
475
    const int dstStride = dstStride_a[0];
476
    const int vertLumPerChroma = 2;
477
    const vector unsigned char yperm = vec_lvsl(0, ysrc);
478
    register unsigned int y;
479

    
480
    if (width&15) {
481
        yv12touyvy(ysrc, usrc, vsrc, dst, c->srcW, srcSliceH, lumStride, chromStride, dstStride);
482
        return srcSliceH;
483
    }
484

    
485
    /* This code assumes:
486

487
    1) dst is 16 bytes-aligned
488
    2) dstStride is a multiple of 16
489
    3) width is a multiple of 16
490
    4) lum & chrom stride are multiples of 8
491
    */
492

    
493
    for (y=0; y<height; y++) {
494
        int i;
495
        for (i = 0; i < width - 31; i+= 32) {
496
            const unsigned int j = i >> 1;
497
            vector unsigned char v_yA = vec_ld(i, ysrc);
498
            vector unsigned char v_yB = vec_ld(i + 16, ysrc);
499
            vector unsigned char v_yC = vec_ld(i + 32, ysrc);
500
            vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
501
            vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
502
            vector unsigned char v_uA = vec_ld(j, usrc);
503
            vector unsigned char v_uB = vec_ld(j + 16, usrc);
504
            vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
505
            vector unsigned char v_vA = vec_ld(j, vsrc);
506
            vector unsigned char v_vB = vec_ld(j + 16, vsrc);
507
            vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
508
            vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
509
            vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
510
            vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
511
            vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
512
            vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
513
            vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
514
            vec_st(v_uyvy_0, (i << 1), dst);
515
            vec_st(v_uyvy_1, (i << 1) + 16, dst);
516
            vec_st(v_uyvy_2, (i << 1) + 32, dst);
517
            vec_st(v_uyvy_3, (i << 1) + 48, dst);
518
        }
519
        if (i < width) {
520
            const unsigned int j = i >> 1;
521
            vector unsigned char v_y1 = vec_ld(i, ysrc);
522
            vector unsigned char v_u = vec_ld(j, usrc);
523
            vector unsigned char v_v = vec_ld(j, vsrc);
524
            vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
525
            vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
526
            vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
527
            vec_st(v_uyvy_0, (i << 1), dst);
528
            vec_st(v_uyvy_1, (i << 1) + 16, dst);
529
        }
530
        if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {
531
            usrc += chromStride;
532
            vsrc += chromStride;
533
        }
534
        ysrc += lumStride;
535
        dst += dstStride;
536
    }
537
    return srcSliceH;
538
}