FFmpeg  4.3.9
swscale_vsx.c
Go to the documentation of this file.
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
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (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 GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License 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 #include <inttypes.h>
25 
26 #include "config.h"
27 #include "libswscale/swscale.h"
29 #include "libavutil/attributes.h"
30 #include "libavutil/cpu.h"
31 #include "yuv2rgb_altivec.h"
33 
34 #if HAVE_VSX
35 #define vzero vec_splat_s32(0)
36 
37 #if !HAVE_BIGENDIAN
38 #define GET_LS(a,b,c,s) {\
39  ls = a;\
40  a = vec_vsx_ld(((b) << 1) + 16, s);\
41  }
42 
43 #define yuv2planeX_8(d1, d2, l1, src, x, perm, filter) do {\
44  vector signed short ls;\
45  vector signed int vf1, vf2, i1, i2;\
46  GET_LS(l1, x, perm, src);\
47  i1 = vec_mule(filter, ls);\
48  i2 = vec_mulo(filter, ls);\
49  vf1 = vec_mergeh(i1, i2);\
50  vf2 = vec_mergel(i1, i2);\
51  d1 = vec_add(d1, vf1);\
52  d2 = vec_add(d2, vf2);\
53  } while (0)
54 
55 #define LOAD_FILTER(vf,f) {\
56  vf = vec_vsx_ld(joffset, f);\
57 }
58 #define LOAD_L1(ll1,s,p){\
59  ll1 = vec_vsx_ld(xoffset, s);\
60 }
61 
62 // The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
63 
64 // The neat trick: We only care for half the elements,
65 // high or low depending on (i<<3)%16 (it's 0 or 8 here),
66 // and we're going to use vec_mule, so we choose
67 // carefully how to "unpack" the elements into the even slots.
68 #define GET_VF4(a, vf, f) {\
69  vf = (vector signed short)vec_vsx_ld(a << 3, f);\
70  vf = vec_mergeh(vf, (vector signed short)vzero);\
71 }
72 #define FIRST_LOAD(sv, pos, s, per) {}
73 #define UPDATE_PTR(s0, d0, s1, d1) {}
74 #define LOAD_SRCV(pos, a, s, per, v0, v1, vf) {\
75  vf = vec_vsx_ld(pos + a, s);\
76 }
77 #define LOAD_SRCV8(pos, a, s, per, v0, v1, vf) LOAD_SRCV(pos, a, s, per, v0, v1, vf)
78 #define GET_VFD(a, b, f, vf0, vf1, per, vf, off) {\
79  vf = vec_vsx_ld((a * 2 * filterSize) + (b * 2) + off, f);\
80 }
81 
82 #define FUNC(name) name ## _vsx
83 #include "swscale_ppc_template.c"
84 #undef FUNC
85 
86 #undef vzero
87 
88 #endif /* !HAVE_BIGENDIAN */
89 
90 static void yuv2plane1_8_u(const int16_t *src, uint8_t *dest, int dstW,
91  const uint8_t *dither, int offset, int start)
92 {
93  int i;
94  for (i = start; i < dstW; i++) {
95  int val = (src[i] + dither[(i + offset) & 7]) >> 7;
96  dest[i] = av_clip_uint8(val);
97  }
98 }
99 
100 static void yuv2plane1_8_vsx(const int16_t *src, uint8_t *dest, int dstW,
101  const uint8_t *dither, int offset)
102 {
103  const int dst_u = -(uintptr_t)dest & 15;
104  int i, j;
105  LOCAL_ALIGNED(16, int16_t, val, [16]);
106  const vec_u16 shifts = (vec_u16) {7, 7, 7, 7, 7, 7, 7, 7};
107  vec_s16 vi, vileft, ditherleft, ditherright;
108  vec_u8 vd;
109 
110  for (j = 0; j < 16; j++) {
111  val[j] = dither[(dst_u + offset + j) & 7];
112  }
113 
114  ditherleft = vec_ld(0, val);
115  ditherright = vec_ld(0, &val[8]);
116 
117  yuv2plane1_8_u(src, dest, dst_u, dither, offset, 0);
118 
119  for (i = dst_u; i < dstW - 15; i += 16) {
120 
121  vi = vec_vsx_ld(0, &src[i]);
122  vi = vec_adds(ditherleft, vi);
123  vileft = vec_sra(vi, shifts);
124 
125  vi = vec_vsx_ld(0, &src[i + 8]);
126  vi = vec_adds(ditherright, vi);
127  vi = vec_sra(vi, shifts);
128 
129  vd = vec_packsu(vileft, vi);
130  vec_st(vd, 0, &dest[i]);
131  }
132 
133  yuv2plane1_8_u(src, dest, dstW, dither, offset, i);
134 }
135 
136 #if !HAVE_BIGENDIAN
137 
138 #define output_pixel(pos, val) \
139  if (big_endian) { \
140  AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \
141  } else { \
142  AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
143  }
144 
145 static void yuv2plane1_nbps_u(const int16_t *src, uint16_t *dest, int dstW,
146  int big_endian, int output_bits, int start)
147 {
148  int i;
149  int shift = 15 - output_bits;
150 
151  for (i = start; i < dstW; i++) {
152  int val = src[i] + (1 << (shift - 1));
153  output_pixel(&dest[i], val);
154  }
155 }
156 
157 static av_always_inline void yuv2plane1_nbps_vsx(const int16_t *src,
158  uint16_t *dest, int dstW,
159  const int big_endian,
160  const int output_bits)
161 {
162  const int dst_u = -(uintptr_t)dest & 7;
163  const int shift = 15 - output_bits;
164  const int add = (1 << (shift - 1));
165  const int clip = (1 << output_bits) - 1;
166  const vec_u16 vadd = (vec_u16) {add, add, add, add, add, add, add, add};
167  const vec_u16 vswap = (vec_u16) vec_splat_u16(big_endian ? 8 : 0);
168  const vec_u16 vshift = (vec_u16) vec_splat_u16(shift);
169  const vec_u16 vlargest = (vec_u16) {clip, clip, clip, clip, clip, clip, clip, clip};
170  vec_u16 v;
171  int i;
172 
173  yuv2plane1_nbps_u(src, dest, dst_u, big_endian, output_bits, 0);
174 
175  for (i = dst_u; i < dstW - 7; i += 8) {
176  v = vec_vsx_ld(0, (const uint16_t *) &src[i]);
177  v = vec_add(v, vadd);
178  v = vec_sr(v, vshift);
179  v = vec_min(v, vlargest);
180  v = vec_rl(v, vswap);
181  vec_st(v, 0, &dest[i]);
182  }
183 
184  yuv2plane1_nbps_u(src, dest, dstW, big_endian, output_bits, i);
185 }
186 
187 static void yuv2planeX_nbps_u(const int16_t *filter, int filterSize,
188  const int16_t **src, uint16_t *dest, int dstW,
189  int big_endian, int output_bits, int start)
190 {
191  int i;
192  int shift = 11 + 16 - output_bits;
193 
194  for (i = start; i < dstW; i++) {
195  int val = 1 << (shift - 1);
196  int j;
197 
198  for (j = 0; j < filterSize; j++)
199  val += src[j][i] * filter[j];
200 
201  output_pixel(&dest[i], val);
202  }
203 }
204 
205 static void yuv2planeX_nbps_vsx(const int16_t *filter, int filterSize,
206  const int16_t **src, uint16_t *dest, int dstW,
207  int big_endian, int output_bits)
208 {
209  const int dst_u = -(uintptr_t)dest & 7;
210  const int shift = 11 + 16 - output_bits;
211  const int add = (1 << (shift - 1));
212  const int clip = (1 << output_bits) - 1;
213  const uint16_t swap = big_endian ? 8 : 0;
214  const vec_u32 vadd = (vec_u32) {add, add, add, add};
215  const vec_u32 vshift = (vec_u32) {shift, shift, shift, shift};
216  const vec_u16 vswap = (vec_u16) {swap, swap, swap, swap, swap, swap, swap, swap};
217  const vec_u16 vlargest = (vec_u16) {clip, clip, clip, clip, clip, clip, clip, clip};
218  const vec_s16 vzero = vec_splat_s16(0);
219  const vec_u8 vperm = (vec_u8) {0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15};
220  vec_s16 vfilter[MAX_FILTER_SIZE], vin;
221  vec_u16 v;
222  vec_u32 vleft, vright, vtmp;
223  int i, j;
224 
225  for (i = 0; i < filterSize; i++) {
226  vfilter[i] = (vec_s16) {filter[i], filter[i], filter[i], filter[i],
227  filter[i], filter[i], filter[i], filter[i]};
228  }
229 
230  yuv2planeX_nbps_u(filter, filterSize, src, dest, dst_u, big_endian, output_bits, 0);
231 
232  for (i = dst_u; i < dstW - 7; i += 8) {
233  vleft = vright = vadd;
234 
235  for (j = 0; j < filterSize; j++) {
236  vin = vec_vsx_ld(0, &src[j][i]);
237  vtmp = (vec_u32) vec_mule(vin, vfilter[j]);
238  vleft = vec_add(vleft, vtmp);
239  vtmp = (vec_u32) vec_mulo(vin, vfilter[j]);
240  vright = vec_add(vright, vtmp);
241  }
242 
243  vleft = vec_sra(vleft, vshift);
244  vright = vec_sra(vright, vshift);
245  v = vec_packsu(vleft, vright);
246  v = (vec_u16) vec_max((vec_s16) v, vzero);
247  v = vec_min(v, vlargest);
248  v = vec_rl(v, vswap);
249  v = vec_perm(v, v, vperm);
250  vec_st(v, 0, &dest[i]);
251  }
252 
253  yuv2planeX_nbps_u(filter, filterSize, src, dest, dstW, big_endian, output_bits, i);
254 }
255 
256 
257 #undef output_pixel
258 
259 #define output_pixel(pos, val, bias, signedness) \
260  if (big_endian) { \
261  AV_WB16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
262  } else { \
263  AV_WL16(pos, bias + av_clip_ ## signedness ## 16(val >> shift)); \
264  }
265 
266 static void yuv2plane1_16_u(const int32_t *src, uint16_t *dest, int dstW,
267  int big_endian, int output_bits, int start)
268 {
269  int i;
270  const int shift = 3;
271 
272  for (i = start; i < dstW; i++) {
273  int val = src[i] + (1 << (shift - 1));
274  output_pixel(&dest[i], val, 0, uint);
275  }
276 }
277 
278 static av_always_inline void yuv2plane1_16_vsx(const int32_t *src,
279  uint16_t *dest, int dstW,
280  const int big_endian,
281  int output_bits)
282 {
283  const int dst_u = -(uintptr_t)dest & 7;
284  const int shift = 3;
285  const int add = (1 << (shift - 1));
286  const vec_u32 vadd = (vec_u32) {add, add, add, add};
287  const vec_u16 vswap = (vec_u16) vec_splat_u16(big_endian ? 8 : 0);
288  const vec_u32 vshift = (vec_u32) vec_splat_u32(shift);
289  vec_u32 v, v2;
290  vec_u16 vd;
291  int i;
292 
293  yuv2plane1_16_u(src, dest, dst_u, big_endian, output_bits, 0);
294 
295  for (i = dst_u; i < dstW - 7; i += 8) {
296  v = vec_vsx_ld(0, (const uint32_t *) &src[i]);
297  v = vec_add(v, vadd);
298  v = vec_sr(v, vshift);
299 
300  v2 = vec_vsx_ld(0, (const uint32_t *) &src[i + 4]);
301  v2 = vec_add(v2, vadd);
302  v2 = vec_sr(v2, vshift);
303 
304  vd = vec_packsu(v, v2);
305  vd = vec_rl(vd, vswap);
306 
307  vec_st(vd, 0, &dest[i]);
308  }
309 
310  yuv2plane1_16_u(src, dest, dstW, big_endian, output_bits, i);
311 }
312 
313 #if HAVE_POWER8
314 
315 static void yuv2planeX_16_u(const int16_t *filter, int filterSize,
316  const int32_t **src, uint16_t *dest, int dstW,
317  int big_endian, int output_bits, int start)
318 {
319  int i;
320  int shift = 15;
321 
322  for (i = start; i < dstW; i++) {
323  int val = 1 << (shift - 1);
324  int j;
325 
326  /* range of val is [0,0x7FFFFFFF], so 31 bits, but with lanczos/spline
327  * filters (or anything with negative coeffs, the range can be slightly
328  * wider in both directions. To account for this overflow, we subtract
329  * a constant so it always fits in the signed range (assuming a
330  * reasonable filterSize), and re-add that at the end. */
331  val -= 0x40000000;
332  for (j = 0; j < filterSize; j++)
333  val += src[j][i] * (unsigned)filter[j];
334 
335  output_pixel(&dest[i], val, 0x8000, int);
336  }
337 }
338 
339 static void yuv2planeX_16_vsx(const int16_t *filter, int filterSize,
340  const int32_t **src, uint16_t *dest, int dstW,
341  int big_endian, int output_bits)
342 {
343  const int dst_u = -(uintptr_t)dest & 7;
344  const int shift = 15;
345  const int bias = 0x8000;
346  const int add = (1 << (shift - 1)) - 0x40000000;
347  const uint16_t swap = big_endian ? 8 : 0;
348  const vec_u32 vadd = (vec_u32) {add, add, add, add};
349  const vec_u32 vshift = (vec_u32) {shift, shift, shift, shift};
350  const vec_u16 vswap = (vec_u16) {swap, swap, swap, swap, swap, swap, swap, swap};
351  const vec_u16 vbias = (vec_u16) {bias, bias, bias, bias, bias, bias, bias, bias};
352  vec_s32 vfilter[MAX_FILTER_SIZE];
353  vec_u16 v;
354  vec_u32 vleft, vright, vtmp;
355  vec_s32 vin32l, vin32r;
356  int i, j;
357 
358  for (i = 0; i < filterSize; i++) {
359  vfilter[i] = (vec_s32) {filter[i], filter[i], filter[i], filter[i]};
360  }
361 
362  yuv2planeX_16_u(filter, filterSize, src, dest, dst_u, big_endian, output_bits, 0);
363 
364  for (i = dst_u; i < dstW - 7; i += 8) {
365  vleft = vright = vadd;
366 
367  for (j = 0; j < filterSize; j++) {
368  vin32l = vec_vsx_ld(0, &src[j][i]);
369  vin32r = vec_vsx_ld(0, &src[j][i + 4]);
370 
371  vtmp = (vec_u32) vec_mul(vin32l, vfilter[j]);
372  vleft = vec_add(vleft, vtmp);
373  vtmp = (vec_u32) vec_mul(vin32r, vfilter[j]);
374  vright = vec_add(vright, vtmp);
375  }
376 
377  vleft = vec_sra(vleft, vshift);
378  vright = vec_sra(vright, vshift);
379  v = (vec_u16) vec_packs((vec_s32) vleft, (vec_s32) vright);
380  v = vec_add(v, vbias);
381  v = vec_rl(v, vswap);
382  vec_st(v, 0, &dest[i]);
383  }
384 
385  yuv2planeX_16_u(filter, filterSize, src, dest, dstW, big_endian, output_bits, i);
386 }
387 
388 #endif /* HAVE_POWER8 */
389 
390 #define yuv2NBPS(bits, BE_LE, is_be, template_size, typeX_t) \
391  yuv2NBPS1(bits, BE_LE, is_be, template_size, typeX_t) \
392  yuv2NBPSX(bits, BE_LE, is_be, template_size, typeX_t)
393 
394 #define yuv2NBPS1(bits, BE_LE, is_be, template_size, typeX_t) \
395 static void yuv2plane1_ ## bits ## BE_LE ## _vsx(const int16_t *src, \
396  uint8_t *dest, int dstW, \
397  const uint8_t *dither, int offset) \
398 { \
399  yuv2plane1_ ## template_size ## _vsx((const typeX_t *) src, \
400  (uint16_t *) dest, dstW, is_be, bits); \
401 }
402 
403 #define yuv2NBPSX(bits, BE_LE, is_be, template_size, typeX_t) \
404 static void yuv2planeX_ ## bits ## BE_LE ## _vsx(const int16_t *filter, int filterSize, \
405  const int16_t **src, uint8_t *dest, int dstW, \
406  const uint8_t *dither, int offset)\
407 { \
408  yuv2planeX_## template_size ## _vsx(filter, \
409  filterSize, (const typeX_t **) src, \
410  (uint16_t *) dest, dstW, is_be, bits); \
411 }
412 
413 yuv2NBPS( 9, BE, 1, nbps, int16_t)
414 yuv2NBPS( 9, LE, 0, nbps, int16_t)
415 yuv2NBPS(10, BE, 1, nbps, int16_t)
416 yuv2NBPS(10, LE, 0, nbps, int16_t)
417 yuv2NBPS(12, BE, 1, nbps, int16_t)
418 yuv2NBPS(12, LE, 0, nbps, int16_t)
419 yuv2NBPS(14, BE, 1, nbps, int16_t)
420 yuv2NBPS(14, LE, 0, nbps, int16_t)
421 
422 yuv2NBPS1(16, BE, 1, 16, int32_t)
423 yuv2NBPS1(16, LE, 0, 16, int32_t)
424 #if HAVE_POWER8
425 yuv2NBPSX(16, BE, 1, 16, int32_t)
426 yuv2NBPSX(16, LE, 0, 16, int32_t)
427 #endif
428 
429 #define WRITERGB \
430  R_l = vec_max(R_l, zero32); \
431  R_r = vec_max(R_r, zero32); \
432  G_l = vec_max(G_l, zero32); \
433  G_r = vec_max(G_r, zero32); \
434  B_l = vec_max(B_l, zero32); \
435  B_r = vec_max(B_r, zero32); \
436 \
437  R_l = vec_min(R_l, rgbclip); \
438  R_r = vec_min(R_r, rgbclip); \
439  G_l = vec_min(G_l, rgbclip); \
440  G_r = vec_min(G_r, rgbclip); \
441  B_l = vec_min(B_l, rgbclip); \
442  B_r = vec_min(B_r, rgbclip); \
443 \
444  R_l = vec_sr(R_l, shift22); \
445  R_r = vec_sr(R_r, shift22); \
446  G_l = vec_sr(G_l, shift22); \
447  G_r = vec_sr(G_r, shift22); \
448  B_l = vec_sr(B_l, shift22); \
449  B_r = vec_sr(B_r, shift22); \
450 \
451  rd16 = vec_packsu(R_l, R_r); \
452  gd16 = vec_packsu(G_l, G_r); \
453  bd16 = vec_packsu(B_l, B_r); \
454  rd = vec_packsu(rd16, zero16); \
455  gd = vec_packsu(gd16, zero16); \
456  bd = vec_packsu(bd16, zero16); \
457 \
458  switch(target) { \
459  case AV_PIX_FMT_RGB24: \
460  out0 = vec_perm(rd, gd, perm3rg0); \
461  out0 = vec_perm(out0, bd, perm3tb0); \
462  out1 = vec_perm(rd, gd, perm3rg1); \
463  out1 = vec_perm(out1, bd, perm3tb1); \
464 \
465  vec_vsx_st(out0, 0, dest); \
466  vec_vsx_st(out1, 16, dest); \
467 \
468  dest += 24; \
469  break; \
470  case AV_PIX_FMT_BGR24: \
471  out0 = vec_perm(bd, gd, perm3rg0); \
472  out0 = vec_perm(out0, rd, perm3tb0); \
473  out1 = vec_perm(bd, gd, perm3rg1); \
474  out1 = vec_perm(out1, rd, perm3tb1); \
475 \
476  vec_vsx_st(out0, 0, dest); \
477  vec_vsx_st(out1, 16, dest); \
478 \
479  dest += 24; \
480  break; \
481  case AV_PIX_FMT_BGRA: \
482  out0 = vec_mergeh(bd, gd); \
483  out1 = vec_mergeh(rd, ad); \
484 \
485  tmp8 = (vec_u8) vec_mergeh((vec_u16) out0, (vec_u16) out1); \
486  vec_vsx_st(tmp8, 0, dest); \
487  tmp8 = (vec_u8) vec_mergel((vec_u16) out0, (vec_u16) out1); \
488  vec_vsx_st(tmp8, 16, dest); \
489 \
490  dest += 32; \
491  break; \
492  case AV_PIX_FMT_RGBA: \
493  out0 = vec_mergeh(rd, gd); \
494  out1 = vec_mergeh(bd, ad); \
495 \
496  tmp8 = (vec_u8) vec_mergeh((vec_u16) out0, (vec_u16) out1); \
497  vec_vsx_st(tmp8, 0, dest); \
498  tmp8 = (vec_u8) vec_mergel((vec_u16) out0, (vec_u16) out1); \
499  vec_vsx_st(tmp8, 16, dest); \
500 \
501  dest += 32; \
502  break; \
503  case AV_PIX_FMT_ARGB: \
504  out0 = vec_mergeh(ad, rd); \
505  out1 = vec_mergeh(gd, bd); \
506 \
507  tmp8 = (vec_u8) vec_mergeh((vec_u16) out0, (vec_u16) out1); \
508  vec_vsx_st(tmp8, 0, dest); \
509  tmp8 = (vec_u8) vec_mergel((vec_u16) out0, (vec_u16) out1); \
510  vec_vsx_st(tmp8, 16, dest); \
511 \
512  dest += 32; \
513  break; \
514  case AV_PIX_FMT_ABGR: \
515  out0 = vec_mergeh(ad, bd); \
516  out1 = vec_mergeh(gd, rd); \
517 \
518  tmp8 = (vec_u8) vec_mergeh((vec_u16) out0, (vec_u16) out1); \
519  vec_vsx_st(tmp8, 0, dest); \
520  tmp8 = (vec_u8) vec_mergel((vec_u16) out0, (vec_u16) out1); \
521  vec_vsx_st(tmp8, 16, dest); \
522 \
523  dest += 32; \
524  break; \
525  }
526 
527 static av_always_inline void
528 yuv2rgb_full_X_vsx_template(SwsContext *c, const int16_t *lumFilter,
529  const int16_t **lumSrc, int lumFilterSize,
530  const int16_t *chrFilter, const int16_t **chrUSrc,
531  const int16_t **chrVSrc, int chrFilterSize,
532  const int16_t **alpSrc, uint8_t *dest,
533  int dstW, int y, enum AVPixelFormat target, int hasAlpha)
534 {
535  vec_s16 vv;
536  vec_s32 vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
537  vec_s32 R_l, R_r, G_l, G_r, B_l, B_r;
538  vec_s32 tmp, tmp2, tmp3, tmp4;
539  vec_u16 rd16, gd16, bd16;
540  vec_u8 rd, bd, gd, ad, out0, out1, tmp8;
541  vec_s16 vlumFilter[MAX_FILTER_SIZE], vchrFilter[MAX_FILTER_SIZE];
542  const vec_s32 ystart = vec_splats(1 << 9);
543  const vec_s32 uvstart = vec_splats((1 << 9) - (128 << 19));
544  const vec_u16 zero16 = vec_splat_u16(0);
545  const vec_s32 y_offset = vec_splats(c->yuv2rgb_y_offset);
546  const vec_s32 y_coeff = vec_splats(c->yuv2rgb_y_coeff);
547  const vec_s32 y_add = vec_splats(1 << 21);
548  const vec_s32 v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff);
549  const vec_s32 v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff);
550  const vec_s32 u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff);
551  const vec_s32 u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff);
552  const vec_s32 rgbclip = vec_splats(1 << 30);
553  const vec_s32 zero32 = vec_splat_s32(0);
554  const vec_u32 shift22 = vec_splats(22U);
555  const vec_u32 shift10 = vec_splat_u32(10);
556  int i, j;
557 
558  // Various permutations
559  const vec_u8 perm3rg0 = (vec_u8) {0x0, 0x10, 0,
560  0x1, 0x11, 0,
561  0x2, 0x12, 0,
562  0x3, 0x13, 0,
563  0x4, 0x14, 0,
564  0x5 };
565  const vec_u8 perm3rg1 = (vec_u8) { 0x15, 0,
566  0x6, 0x16, 0,
567  0x7, 0x17, 0 };
568  const vec_u8 perm3tb0 = (vec_u8) {0x0, 0x1, 0x10,
569  0x3, 0x4, 0x11,
570  0x6, 0x7, 0x12,
571  0x9, 0xa, 0x13,
572  0xc, 0xd, 0x14,
573  0xf };
574  const vec_u8 perm3tb1 = (vec_u8) { 0x0, 0x15,
575  0x2, 0x3, 0x16,
576  0x5, 0x6, 0x17 };
577 
578  ad = vec_splats((uint8_t) 255);
579 
580  for (i = 0; i < lumFilterSize; i++)
581  vlumFilter[i] = vec_splats(lumFilter[i]);
582  for (i = 0; i < chrFilterSize; i++)
583  vchrFilter[i] = vec_splats(chrFilter[i]);
584 
585  for (i = 0; i < dstW; i += 8) {
586  vy32_l =
587  vy32_r = ystart;
588  vu32_l =
589  vu32_r =
590  vv32_l =
591  vv32_r = uvstart;
592 
593  for (j = 0; j < lumFilterSize; j++) {
594  vv = vec_ld(0, &lumSrc[j][i]);
595  tmp = vec_mule(vv, vlumFilter[j]);
596  tmp2 = vec_mulo(vv, vlumFilter[j]);
597  tmp3 = vec_mergeh(tmp, tmp2);
598  tmp4 = vec_mergel(tmp, tmp2);
599 
600  vy32_l = vec_adds(vy32_l, tmp3);
601  vy32_r = vec_adds(vy32_r, tmp4);
602  }
603 
604  for (j = 0; j < chrFilterSize; j++) {
605  vv = vec_ld(0, &chrUSrc[j][i]);
606  tmp = vec_mule(vv, vchrFilter[j]);
607  tmp2 = vec_mulo(vv, vchrFilter[j]);
608  tmp3 = vec_mergeh(tmp, tmp2);
609  tmp4 = vec_mergel(tmp, tmp2);
610 
611  vu32_l = vec_adds(vu32_l, tmp3);
612  vu32_r = vec_adds(vu32_r, tmp4);
613 
614  vv = vec_ld(0, &chrVSrc[j][i]);
615  tmp = vec_mule(vv, vchrFilter[j]);
616  tmp2 = vec_mulo(vv, vchrFilter[j]);
617  tmp3 = vec_mergeh(tmp, tmp2);
618  tmp4 = vec_mergel(tmp, tmp2);
619 
620  vv32_l = vec_adds(vv32_l, tmp3);
621  vv32_r = vec_adds(vv32_r, tmp4);
622  }
623 
624  vy32_l = vec_sra(vy32_l, shift10);
625  vy32_r = vec_sra(vy32_r, shift10);
626  vu32_l = vec_sra(vu32_l, shift10);
627  vu32_r = vec_sra(vu32_r, shift10);
628  vv32_l = vec_sra(vv32_l, shift10);
629  vv32_r = vec_sra(vv32_r, shift10);
630 
631  vy32_l = vec_sub(vy32_l, y_offset);
632  vy32_r = vec_sub(vy32_r, y_offset);
633  vy32_l = vec_mul(vy32_l, y_coeff);
634  vy32_r = vec_mul(vy32_r, y_coeff);
635  vy32_l = vec_add(vy32_l, y_add);
636  vy32_r = vec_add(vy32_r, y_add);
637 
638  R_l = vec_mul(vv32_l, v2r_coeff);
639  R_l = vec_add(R_l, vy32_l);
640  R_r = vec_mul(vv32_r, v2r_coeff);
641  R_r = vec_add(R_r, vy32_r);
642  G_l = vec_mul(vv32_l, v2g_coeff);
643  tmp32 = vec_mul(vu32_l, u2g_coeff);
644  G_l = vec_add(G_l, vy32_l);
645  G_l = vec_add(G_l, tmp32);
646  G_r = vec_mul(vv32_r, v2g_coeff);
647  tmp32 = vec_mul(vu32_r, u2g_coeff);
648  G_r = vec_add(G_r, vy32_r);
649  G_r = vec_add(G_r, tmp32);
650 
651  B_l = vec_mul(vu32_l, u2b_coeff);
652  B_l = vec_add(B_l, vy32_l);
653  B_r = vec_mul(vu32_r, u2b_coeff);
654  B_r = vec_add(B_r, vy32_r);
655 
656  WRITERGB
657  }
658 }
659 
660 #define SETUP(x, buf0, alpha1, buf1, alpha) { \
661  x = vec_ld(0, buf0); \
662  tmp = vec_mule(x, alpha1); \
663  tmp2 = vec_mulo(x, alpha1); \
664  tmp3 = vec_mergeh(tmp, tmp2); \
665  tmp4 = vec_mergel(tmp, tmp2); \
666 \
667  x = vec_ld(0, buf1); \
668  tmp = vec_mule(x, alpha); \
669  tmp2 = vec_mulo(x, alpha); \
670  tmp5 = vec_mergeh(tmp, tmp2); \
671  tmp6 = vec_mergel(tmp, tmp2); \
672 \
673  tmp3 = vec_add(tmp3, tmp5); \
674  tmp4 = vec_add(tmp4, tmp6); \
675 }
676 
677 
678 static av_always_inline void
679 yuv2rgb_full_2_vsx_template(SwsContext *c, const int16_t *buf[2],
680  const int16_t *ubuf[2], const int16_t *vbuf[2],
681  const int16_t *abuf[2], uint8_t *dest, int dstW,
682  int yalpha, int uvalpha, int y,
683  enum AVPixelFormat target, int hasAlpha)
684 {
685  const int16_t *buf0 = buf[0], *buf1 = buf[1],
686  *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
687  *vbuf0 = vbuf[0], *vbuf1 = vbuf[1],
688  *abuf0 = hasAlpha ? abuf[0] : NULL,
689  *abuf1 = hasAlpha ? abuf[1] : NULL;
690  const int16_t yalpha1 = 4096 - yalpha;
691  const int16_t uvalpha1 = 4096 - uvalpha;
692  vec_s16 vy, vu, vv, A = vec_splat_s16(0);
693  vec_s32 vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
694  vec_s32 R_l, R_r, G_l, G_r, B_l, B_r;
695  vec_s32 tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
696  vec_u16 rd16, gd16, bd16;
697  vec_u8 rd, bd, gd, ad, out0, out1, tmp8;
698  const vec_s16 vyalpha1 = vec_splats(yalpha1);
699  const vec_s16 vuvalpha1 = vec_splats(uvalpha1);
700  const vec_s16 vyalpha = vec_splats((int16_t) yalpha);
701  const vec_s16 vuvalpha = vec_splats((int16_t) uvalpha);
702  const vec_u16 zero16 = vec_splat_u16(0);
703  const vec_s32 y_offset = vec_splats(c->yuv2rgb_y_offset);
704  const vec_s32 y_coeff = vec_splats(c->yuv2rgb_y_coeff);
705  const vec_s32 y_add = vec_splats(1 << 21);
706  const vec_s32 v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff);
707  const vec_s32 v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff);
708  const vec_s32 u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff);
709  const vec_s32 u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff);
710  const vec_s32 rgbclip = vec_splats(1 << 30);
711  const vec_s32 zero32 = vec_splat_s32(0);
712  const vec_u32 shift19 = vec_splats(19U);
713  const vec_u32 shift22 = vec_splats(22U);
714  const vec_u32 shift10 = vec_splat_u32(10);
715  const vec_s32 dec128 = vec_splats(128 << 19);
716  const vec_s32 add18 = vec_splats(1 << 18);
717  int i;
718 
719  // Various permutations
720  const vec_u8 perm3rg0 = (vec_u8) {0x0, 0x10, 0,
721  0x1, 0x11, 0,
722  0x2, 0x12, 0,
723  0x3, 0x13, 0,
724  0x4, 0x14, 0,
725  0x5 };
726  const vec_u8 perm3rg1 = (vec_u8) { 0x15, 0,
727  0x6, 0x16, 0,
728  0x7, 0x17, 0 };
729  const vec_u8 perm3tb0 = (vec_u8) {0x0, 0x1, 0x10,
730  0x3, 0x4, 0x11,
731  0x6, 0x7, 0x12,
732  0x9, 0xa, 0x13,
733  0xc, 0xd, 0x14,
734  0xf };
735  const vec_u8 perm3tb1 = (vec_u8) { 0x0, 0x15,
736  0x2, 0x3, 0x16,
737  0x5, 0x6, 0x17 };
738 
739  av_assert2(yalpha <= 4096U);
740  av_assert2(uvalpha <= 4096U);
741 
742  for (i = 0; i < dstW; i += 8) {
743  SETUP(vy, &buf0[i], vyalpha1, &buf1[i], vyalpha);
744  vy32_l = vec_sra(tmp3, shift10);
745  vy32_r = vec_sra(tmp4, shift10);
746 
747  SETUP(vu, &ubuf0[i], vuvalpha1, &ubuf1[i], vuvalpha);
748  tmp3 = vec_sub(tmp3, dec128);
749  tmp4 = vec_sub(tmp4, dec128);
750  vu32_l = vec_sra(tmp3, shift10);
751  vu32_r = vec_sra(tmp4, shift10);
752 
753  SETUP(vv, &vbuf0[i], vuvalpha1, &vbuf1[i], vuvalpha);
754  tmp3 = vec_sub(tmp3, dec128);
755  tmp4 = vec_sub(tmp4, dec128);
756  vv32_l = vec_sra(tmp3, shift10);
757  vv32_r = vec_sra(tmp4, shift10);
758 
759  if (hasAlpha) {
760  SETUP(A, &abuf0[i], vyalpha1, &abuf1[i], vyalpha);
761  tmp3 = vec_add(tmp3, add18);
762  tmp4 = vec_add(tmp4, add18);
763  tmp3 = vec_sra(tmp3, shift19);
764  tmp4 = vec_sra(tmp4, shift19);
765  A = vec_packs(tmp3, tmp4);
766  ad = vec_packsu(A, (vec_s16) zero16);
767  } else {
768  ad = vec_splats((uint8_t) 255);
769  }
770 
771  vy32_l = vec_sub(vy32_l, y_offset);
772  vy32_r = vec_sub(vy32_r, y_offset);
773  vy32_l = vec_mul(vy32_l, y_coeff);
774  vy32_r = vec_mul(vy32_r, y_coeff);
775  vy32_l = vec_add(vy32_l, y_add);
776  vy32_r = vec_add(vy32_r, y_add);
777 
778  R_l = vec_mul(vv32_l, v2r_coeff);
779  R_l = vec_add(R_l, vy32_l);
780  R_r = vec_mul(vv32_r, v2r_coeff);
781  R_r = vec_add(R_r, vy32_r);
782  G_l = vec_mul(vv32_l, v2g_coeff);
783  tmp32 = vec_mul(vu32_l, u2g_coeff);
784  G_l = vec_add(G_l, vy32_l);
785  G_l = vec_add(G_l, tmp32);
786  G_r = vec_mul(vv32_r, v2g_coeff);
787  tmp32 = vec_mul(vu32_r, u2g_coeff);
788  G_r = vec_add(G_r, vy32_r);
789  G_r = vec_add(G_r, tmp32);
790 
791  B_l = vec_mul(vu32_l, u2b_coeff);
792  B_l = vec_add(B_l, vy32_l);
793  B_r = vec_mul(vu32_r, u2b_coeff);
794  B_r = vec_add(B_r, vy32_r);
795 
796  WRITERGB
797  }
798 }
799 
800 static av_always_inline void
801 yuv2rgb_2_vsx_template(SwsContext *c, const int16_t *buf[2],
802  const int16_t *ubuf[2], const int16_t *vbuf[2],
803  const int16_t *abuf[2], uint8_t *dest, int dstW,
804  int yalpha, int uvalpha, int y,
805  enum AVPixelFormat target, int hasAlpha)
806 {
807  const int16_t *buf0 = buf[0], *buf1 = buf[1],
808  *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
809  *vbuf0 = vbuf[0], *vbuf1 = vbuf[1],
810  *abuf0 = hasAlpha ? abuf[0] : NULL,
811  *abuf1 = hasAlpha ? abuf[1] : NULL;
812  const int16_t yalpha1 = 4096 - yalpha;
813  const int16_t uvalpha1 = 4096 - uvalpha;
814  vec_s16 vy, vu, vv, A = vec_splat_s16(0);
815  vec_s32 vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32;
816  vec_s32 R_l, R_r, G_l, G_r, B_l, B_r, vud32_l, vud32_r, vvd32_l, vvd32_r;
817  vec_s32 tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
818  vec_u16 rd16, gd16, bd16;
819  vec_u8 rd, bd, gd, ad, out0, out1, tmp8;
820  const vec_s16 vyalpha1 = vec_splats(yalpha1);
821  const vec_s16 vuvalpha1 = vec_splats(uvalpha1);
822  const vec_s16 vyalpha = vec_splats((int16_t) yalpha);
823  const vec_s16 vuvalpha = vec_splats((int16_t) uvalpha);
824  const vec_u16 zero16 = vec_splat_u16(0);
825  const vec_s32 y_offset = vec_splats(c->yuv2rgb_y_offset);
826  const vec_s32 y_coeff = vec_splats(c->yuv2rgb_y_coeff);
827  const vec_s32 y_add = vec_splats(1 << 21);
828  const vec_s32 v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff);
829  const vec_s32 v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff);
830  const vec_s32 u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff);
831  const vec_s32 u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff);
832  const vec_s32 rgbclip = vec_splats(1 << 30);
833  const vec_s32 zero32 = vec_splat_s32(0);
834  const vec_u32 shift19 = vec_splats(19U);
835  const vec_u32 shift22 = vec_splats(22U);
836  const vec_u32 shift10 = vec_splat_u32(10);
837  const vec_s32 dec128 = vec_splats(128 << 19);
838  const vec_s32 add18 = vec_splats(1 << 18);
839  int i;
840 
841  // Various permutations
842  const vec_u8 doubleleft = (vec_u8) {0, 1, 2, 3,
843  0, 1, 2, 3,
844  4, 5, 6, 7,
845  4, 5, 6, 7 };
846  const vec_u8 doubleright = (vec_u8) {8, 9, 10, 11,
847  8, 9, 10, 11,
848  12, 13, 14, 15,
849  12, 13, 14, 15 };
850  const vec_u8 perm3rg0 = (vec_u8) {0x0, 0x10, 0,
851  0x1, 0x11, 0,
852  0x2, 0x12, 0,
853  0x3, 0x13, 0,
854  0x4, 0x14, 0,
855  0x5 };
856  const vec_u8 perm3rg1 = (vec_u8) { 0x15, 0,
857  0x6, 0x16, 0,
858  0x7, 0x17, 0 };
859  const vec_u8 perm3tb0 = (vec_u8) {0x0, 0x1, 0x10,
860  0x3, 0x4, 0x11,
861  0x6, 0x7, 0x12,
862  0x9, 0xa, 0x13,
863  0xc, 0xd, 0x14,
864  0xf };
865  const vec_u8 perm3tb1 = (vec_u8) { 0x0, 0x15,
866  0x2, 0x3, 0x16,
867  0x5, 0x6, 0x17 };
868 
869  av_assert2(yalpha <= 4096U);
870  av_assert2(uvalpha <= 4096U);
871 
872  for (i = 0; i < (dstW + 1) >> 1; i += 8) {
873  SETUP(vy, &buf0[i * 2], vyalpha1, &buf1[i * 2], vyalpha);
874  vy32_l = vec_sra(tmp3, shift10);
875  vy32_r = vec_sra(tmp4, shift10);
876 
877  SETUP(vu, &ubuf0[i], vuvalpha1, &ubuf1[i], vuvalpha);
878  tmp3 = vec_sub(tmp3, dec128);
879  tmp4 = vec_sub(tmp4, dec128);
880  vu32_l = vec_sra(tmp3, shift10);
881  vu32_r = vec_sra(tmp4, shift10);
882 
883  SETUP(vv, &vbuf0[i], vuvalpha1, &vbuf1[i], vuvalpha);
884  tmp3 = vec_sub(tmp3, dec128);
885  tmp4 = vec_sub(tmp4, dec128);
886  vv32_l = vec_sra(tmp3, shift10);
887  vv32_r = vec_sra(tmp4, shift10);
888 
889  if (hasAlpha) {
890  SETUP(A, &abuf0[i], vyalpha1, &abuf1[i], vyalpha);
891  tmp3 = vec_add(tmp3, add18);
892  tmp4 = vec_add(tmp4, add18);
893  tmp3 = vec_sra(tmp3, shift19);
894  tmp4 = vec_sra(tmp4, shift19);
895  A = vec_packs(tmp3, tmp4);
896  ad = vec_packsu(A, (vec_s16) zero16);
897  } else {
898  ad = vec_splats((uint8_t) 255);
899  }
900 
901  vy32_l = vec_sub(vy32_l, y_offset);
902  vy32_r = vec_sub(vy32_r, y_offset);
903  vy32_l = vec_mul(vy32_l, y_coeff);
904  vy32_r = vec_mul(vy32_r, y_coeff);
905  vy32_l = vec_add(vy32_l, y_add);
906  vy32_r = vec_add(vy32_r, y_add);
907 
908  // Use the first UV half
909  vud32_l = vec_perm(vu32_l, vu32_l, doubleleft);
910  vud32_r = vec_perm(vu32_l, vu32_l, doubleright);
911  vvd32_l = vec_perm(vv32_l, vv32_l, doubleleft);
912  vvd32_r = vec_perm(vv32_l, vv32_l, doubleright);
913 
914  R_l = vec_mul(vvd32_l, v2r_coeff);
915  R_l = vec_add(R_l, vy32_l);
916  R_r = vec_mul(vvd32_r, v2r_coeff);
917  R_r = vec_add(R_r, vy32_r);
918  G_l = vec_mul(vvd32_l, v2g_coeff);
919  tmp32 = vec_mul(vud32_l, u2g_coeff);
920  G_l = vec_add(G_l, vy32_l);
921  G_l = vec_add(G_l, tmp32);
922  G_r = vec_mul(vvd32_r, v2g_coeff);
923  tmp32 = vec_mul(vud32_r, u2g_coeff);
924  G_r = vec_add(G_r, vy32_r);
925  G_r = vec_add(G_r, tmp32);
926 
927  B_l = vec_mul(vud32_l, u2b_coeff);
928  B_l = vec_add(B_l, vy32_l);
929  B_r = vec_mul(vud32_r, u2b_coeff);
930  B_r = vec_add(B_r, vy32_r);
931 
932  WRITERGB
933 
934  // New Y for the second half
935  SETUP(vy, &buf0[i * 2 + 8], vyalpha1, &buf1[i * 2 + 8], vyalpha);
936  vy32_l = vec_sra(tmp3, shift10);
937  vy32_r = vec_sra(tmp4, shift10);
938 
939  vy32_l = vec_sub(vy32_l, y_offset);
940  vy32_r = vec_sub(vy32_r, y_offset);
941  vy32_l = vec_mul(vy32_l, y_coeff);
942  vy32_r = vec_mul(vy32_r, y_coeff);
943  vy32_l = vec_add(vy32_l, y_add);
944  vy32_r = vec_add(vy32_r, y_add);
945 
946  // Second UV half
947  vud32_l = vec_perm(vu32_r, vu32_r, doubleleft);
948  vud32_r = vec_perm(vu32_r, vu32_r, doubleright);
949  vvd32_l = vec_perm(vv32_r, vv32_r, doubleleft);
950  vvd32_r = vec_perm(vv32_r, vv32_r, doubleright);
951 
952  R_l = vec_mul(vvd32_l, v2r_coeff);
953  R_l = vec_add(R_l, vy32_l);
954  R_r = vec_mul(vvd32_r, v2r_coeff);
955  R_r = vec_add(R_r, vy32_r);
956  G_l = vec_mul(vvd32_l, v2g_coeff);
957  tmp32 = vec_mul(vud32_l, u2g_coeff);
958  G_l = vec_add(G_l, vy32_l);
959  G_l = vec_add(G_l, tmp32);
960  G_r = vec_mul(vvd32_r, v2g_coeff);
961  tmp32 = vec_mul(vud32_r, u2g_coeff);
962  G_r = vec_add(G_r, vy32_r);
963  G_r = vec_add(G_r, tmp32);
964 
965  B_l = vec_mul(vud32_l, u2b_coeff);
966  B_l = vec_add(B_l, vy32_l);
967  B_r = vec_mul(vud32_r, u2b_coeff);
968  B_r = vec_add(B_r, vy32_r);
969 
970  WRITERGB
971  }
972 }
973 
974 #undef SETUP
975 
976 static av_always_inline void
977 yuv2rgb_full_1_vsx_template(SwsContext *c, const int16_t *buf0,
978  const int16_t *ubuf[2], const int16_t *vbuf[2],
979  const int16_t *abuf0, uint8_t *dest, int dstW,
980  int uvalpha, int y, enum AVPixelFormat target,
981  int hasAlpha)
982 {
983  const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
984  const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
985  vec_s16 vy, vu, vv, A = vec_splat_s16(0), tmp16;
986  vec_s32 vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32, tmp32_2;
987  vec_s32 R_l, R_r, G_l, G_r, B_l, B_r;
988  vec_u16 rd16, gd16, bd16;
989  vec_u8 rd, bd, gd, ad, out0, out1, tmp8;
990  const vec_u16 zero16 = vec_splat_u16(0);
991  const vec_s32 y_offset = vec_splats(c->yuv2rgb_y_offset);
992  const vec_s32 y_coeff = vec_splats(c->yuv2rgb_y_coeff);
993  const vec_s32 y_add = vec_splats(1 << 21);
994  const vec_s32 v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff);
995  const vec_s32 v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff);
996  const vec_s32 u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff);
997  const vec_s32 u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff);
998  const vec_s32 rgbclip = vec_splats(1 << 30);
999  const vec_s32 zero32 = vec_splat_s32(0);
1000  const vec_u32 shift2 = vec_splat_u32(2);
1001  const vec_u32 shift22 = vec_splats(22U);
1002  const vec_u16 sub7 = vec_splats((uint16_t) (128 << 7));
1003  const vec_u16 sub8 = vec_splats((uint16_t) (128 << 8));
1004  const vec_s16 mul4 = vec_splat_s16(4);
1005  const vec_s16 mul8 = vec_splat_s16(8);
1006  const vec_s16 add64 = vec_splat_s16(64);
1007  const vec_u16 shift7 = vec_splat_u16(7);
1008  const vec_s16 max255 = vec_splat_s16(255);
1009  int i;
1010 
1011  // Various permutations
1012  const vec_u8 perm3rg0 = (vec_u8) {0x0, 0x10, 0,
1013  0x1, 0x11, 0,
1014  0x2, 0x12, 0,
1015  0x3, 0x13, 0,
1016  0x4, 0x14, 0,
1017  0x5 };
1018  const vec_u8 perm3rg1 = (vec_u8) { 0x15, 0,
1019  0x6, 0x16, 0,
1020  0x7, 0x17, 0 };
1021  const vec_u8 perm3tb0 = (vec_u8) {0x0, 0x1, 0x10,
1022  0x3, 0x4, 0x11,
1023  0x6, 0x7, 0x12,
1024  0x9, 0xa, 0x13,
1025  0xc, 0xd, 0x14,
1026  0xf };
1027  const vec_u8 perm3tb1 = (vec_u8) { 0x0, 0x15,
1028  0x2, 0x3, 0x16,
1029  0x5, 0x6, 0x17 };
1030 
1031  for (i = 0; i < dstW; i += 8) { // The x86 asm also overwrites padding bytes.
1032  vy = vec_ld(0, &buf0[i]);
1033  vy32_l = vec_unpackh(vy);
1034  vy32_r = vec_unpackl(vy);
1035  vy32_l = vec_sl(vy32_l, shift2);
1036  vy32_r = vec_sl(vy32_r, shift2);
1037 
1038  vu = vec_ld(0, &ubuf0[i]);
1039  vv = vec_ld(0, &vbuf0[i]);
1040  if (uvalpha < 2048) {
1041  vu = (vec_s16) vec_sub((vec_u16) vu, sub7);
1042  vv = (vec_s16) vec_sub((vec_u16) vv, sub7);
1043 
1044  tmp32 = vec_mule(vu, mul4);
1045  tmp32_2 = vec_mulo(vu, mul4);
1046  vu32_l = vec_mergeh(tmp32, tmp32_2);
1047  vu32_r = vec_mergel(tmp32, tmp32_2);
1048  tmp32 = vec_mule(vv, mul4);
1049  tmp32_2 = vec_mulo(vv, mul4);
1050  vv32_l = vec_mergeh(tmp32, tmp32_2);
1051  vv32_r = vec_mergel(tmp32, tmp32_2);
1052  } else {
1053  tmp16 = vec_ld(0, &ubuf1[i]);
1054  vu = vec_add(vu, tmp16);
1055  vu = (vec_s16) vec_sub((vec_u16) vu, sub8);
1056  tmp16 = vec_ld(0, &vbuf1[i]);
1057  vv = vec_add(vv, tmp16);
1058  vv = (vec_s16) vec_sub((vec_u16) vv, sub8);
1059 
1060  vu32_l = vec_mule(vu, mul8);
1061  vu32_r = vec_mulo(vu, mul8);
1062  vv32_l = vec_mule(vv, mul8);
1063  vv32_r = vec_mulo(vv, mul8);
1064  }
1065 
1066  if (hasAlpha) {
1067  A = vec_ld(0, &abuf0[i]);
1068  A = vec_add(A, add64);
1069  A = vec_sr(A, shift7);
1070  A = vec_max(A, max255);
1071  ad = vec_packsu(A, (vec_s16) zero16);
1072  } else {
1073  ad = vec_splats((uint8_t) 255);
1074  }
1075 
1076  vy32_l = vec_sub(vy32_l, y_offset);
1077  vy32_r = vec_sub(vy32_r, y_offset);
1078  vy32_l = vec_mul(vy32_l, y_coeff);
1079  vy32_r = vec_mul(vy32_r, y_coeff);
1080  vy32_l = vec_add(vy32_l, y_add);
1081  vy32_r = vec_add(vy32_r, y_add);
1082 
1083  R_l = vec_mul(vv32_l, v2r_coeff);
1084  R_l = vec_add(R_l, vy32_l);
1085  R_r = vec_mul(vv32_r, v2r_coeff);
1086  R_r = vec_add(R_r, vy32_r);
1087  G_l = vec_mul(vv32_l, v2g_coeff);
1088  tmp32 = vec_mul(vu32_l, u2g_coeff);
1089  G_l = vec_add(G_l, vy32_l);
1090  G_l = vec_add(G_l, tmp32);
1091  G_r = vec_mul(vv32_r, v2g_coeff);
1092  tmp32 = vec_mul(vu32_r, u2g_coeff);
1093  G_r = vec_add(G_r, vy32_r);
1094  G_r = vec_add(G_r, tmp32);
1095 
1096  B_l = vec_mul(vu32_l, u2b_coeff);
1097  B_l = vec_add(B_l, vy32_l);
1098  B_r = vec_mul(vu32_r, u2b_coeff);
1099  B_r = vec_add(B_r, vy32_r);
1100 
1101  WRITERGB
1102  }
1103 }
1104 
1105 static av_always_inline void
1106 yuv2rgb_1_vsx_template(SwsContext *c, const int16_t *buf0,
1107  const int16_t *ubuf[2], const int16_t *vbuf[2],
1108  const int16_t *abuf0, uint8_t *dest, int dstW,
1109  int uvalpha, int y, enum AVPixelFormat target,
1110  int hasAlpha)
1111 {
1112  const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
1113  const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
1114  vec_s16 vy, vu, vv, A = vec_splat_s16(0), tmp16;
1115  vec_s32 vy32_l, vy32_r, vu32_l, vu32_r, vv32_l, vv32_r, tmp32, tmp32_2;
1116  vec_s32 vud32_l, vud32_r, vvd32_l, vvd32_r;
1117  vec_s32 R_l, R_r, G_l, G_r, B_l, B_r;
1118  vec_u16 rd16, gd16, bd16;
1119  vec_u8 rd, bd, gd, ad, out0, out1, tmp8;
1120  const vec_u16 zero16 = vec_splat_u16(0);
1121  const vec_s32 y_offset = vec_splats(c->yuv2rgb_y_offset);
1122  const vec_s32 y_coeff = vec_splats(c->yuv2rgb_y_coeff);
1123  const vec_s32 y_add = vec_splats(1 << 21);
1124  const vec_s32 v2r_coeff = vec_splats(c->yuv2rgb_v2r_coeff);
1125  const vec_s32 v2g_coeff = vec_splats(c->yuv2rgb_v2g_coeff);
1126  const vec_s32 u2g_coeff = vec_splats(c->yuv2rgb_u2g_coeff);
1127  const vec_s32 u2b_coeff = vec_splats(c->yuv2rgb_u2b_coeff);
1128  const vec_s32 rgbclip = vec_splats(1 << 30);
1129  const vec_s32 zero32 = vec_splat_s32(0);
1130  const vec_u32 shift2 = vec_splat_u32(2);
1131  const vec_u32 shift22 = vec_splats(22U);
1132  const vec_u16 sub7 = vec_splats((uint16_t) (128 << 7));
1133  const vec_u16 sub8 = vec_splats((uint16_t) (128 << 8));
1134  const vec_s16 mul4 = vec_splat_s16(4);
1135  const vec_s16 mul8 = vec_splat_s16(8);
1136  const vec_s16 add64 = vec_splat_s16(64);
1137  const vec_u16 shift7 = vec_splat_u16(7);
1138  const vec_s16 max255 = vec_splat_s16(255);
1139  int i;
1140 
1141  // Various permutations
1142  const vec_u8 doubleleft = (vec_u8) {0, 1, 2, 3,
1143  0, 1, 2, 3,
1144  4, 5, 6, 7,
1145  4, 5, 6, 7 };
1146  const vec_u8 doubleright = (vec_u8) {8, 9, 10, 11,
1147  8, 9, 10, 11,
1148  12, 13, 14, 15,
1149  12, 13, 14, 15 };
1150  const vec_u8 perm3rg0 = (vec_u8) {0x0, 0x10, 0,
1151  0x1, 0x11, 0,
1152  0x2, 0x12, 0,
1153  0x3, 0x13, 0,
1154  0x4, 0x14, 0,
1155  0x5 };
1156  const vec_u8 perm3rg1 = (vec_u8) { 0x15, 0,
1157  0x6, 0x16, 0,
1158  0x7, 0x17, 0 };
1159  const vec_u8 perm3tb0 = (vec_u8) {0x0, 0x1, 0x10,
1160  0x3, 0x4, 0x11,
1161  0x6, 0x7, 0x12,
1162  0x9, 0xa, 0x13,
1163  0xc, 0xd, 0x14,
1164  0xf };
1165  const vec_u8 perm3tb1 = (vec_u8) { 0x0, 0x15,
1166  0x2, 0x3, 0x16,
1167  0x5, 0x6, 0x17 };
1168 
1169  for (i = 0; i < (dstW + 1) >> 1; i += 8) { // The x86 asm also overwrites padding bytes.
1170  vy = vec_ld(0, &buf0[i * 2]);
1171  vy32_l = vec_unpackh(vy);
1172  vy32_r = vec_unpackl(vy);
1173  vy32_l = vec_sl(vy32_l, shift2);
1174  vy32_r = vec_sl(vy32_r, shift2);
1175 
1176  vu = vec_ld(0, &ubuf0[i]);
1177  vv = vec_ld(0, &vbuf0[i]);
1178  if (uvalpha < 2048) {
1179  vu = (vec_s16) vec_sub((vec_u16) vu, sub7);
1180  vv = (vec_s16) vec_sub((vec_u16) vv, sub7);
1181 
1182  tmp32 = vec_mule(vu, mul4);
1183  tmp32_2 = vec_mulo(vu, mul4);
1184  vu32_l = vec_mergeh(tmp32, tmp32_2);
1185  vu32_r = vec_mergel(tmp32, tmp32_2);
1186  tmp32 = vec_mule(vv, mul4);
1187  tmp32_2 = vec_mulo(vv, mul4);
1188  vv32_l = vec_mergeh(tmp32, tmp32_2);
1189  vv32_r = vec_mergel(tmp32, tmp32_2);
1190  } else {
1191  tmp16 = vec_ld(0, &ubuf1[i]);
1192  vu = vec_add(vu, tmp16);
1193  vu = (vec_s16) vec_sub((vec_u16) vu, sub8);
1194  tmp16 = vec_ld(0, &vbuf1[i]);
1195  vv = vec_add(vv, tmp16);
1196  vv = (vec_s16) vec_sub((vec_u16) vv, sub8);
1197 
1198  vu32_l = vec_mule(vu, mul8);
1199  vu32_r = vec_mulo(vu, mul8);
1200  vv32_l = vec_mule(vv, mul8);
1201  vv32_r = vec_mulo(vv, mul8);
1202  }
1203 
1204  if (hasAlpha) {
1205  A = vec_ld(0, &abuf0[i]);
1206  A = vec_add(A, add64);
1207  A = vec_sr(A, shift7);
1208  A = vec_max(A, max255);
1209  ad = vec_packsu(A, (vec_s16) zero16);
1210  } else {
1211  ad = vec_splats((uint8_t) 255);
1212  }
1213 
1214  vy32_l = vec_sub(vy32_l, y_offset);
1215  vy32_r = vec_sub(vy32_r, y_offset);
1216  vy32_l = vec_mul(vy32_l, y_coeff);
1217  vy32_r = vec_mul(vy32_r, y_coeff);
1218  vy32_l = vec_add(vy32_l, y_add);
1219  vy32_r = vec_add(vy32_r, y_add);
1220 
1221  // Use the first UV half
1222  vud32_l = vec_perm(vu32_l, vu32_l, doubleleft);
1223  vud32_r = vec_perm(vu32_l, vu32_l, doubleright);
1224  vvd32_l = vec_perm(vv32_l, vv32_l, doubleleft);
1225  vvd32_r = vec_perm(vv32_l, vv32_l, doubleright);
1226 
1227  R_l = vec_mul(vvd32_l, v2r_coeff);
1228  R_l = vec_add(R_l, vy32_l);
1229  R_r = vec_mul(vvd32_r, v2r_coeff);
1230  R_r = vec_add(R_r, vy32_r);
1231  G_l = vec_mul(vvd32_l, v2g_coeff);
1232  tmp32 = vec_mul(vud32_l, u2g_coeff);
1233  G_l = vec_add(G_l, vy32_l);
1234  G_l = vec_add(G_l, tmp32);
1235  G_r = vec_mul(vvd32_r, v2g_coeff);
1236  tmp32 = vec_mul(vud32_r, u2g_coeff);
1237  G_r = vec_add(G_r, vy32_r);
1238  G_r = vec_add(G_r, tmp32);
1239 
1240  B_l = vec_mul(vud32_l, u2b_coeff);
1241  B_l = vec_add(B_l, vy32_l);
1242  B_r = vec_mul(vud32_r, u2b_coeff);
1243  B_r = vec_add(B_r, vy32_r);
1244 
1245  WRITERGB
1246 
1247  // New Y for the second half
1248  vy = vec_ld(16, &buf0[i * 2]);
1249  vy32_l = vec_unpackh(vy);
1250  vy32_r = vec_unpackl(vy);
1251  vy32_l = vec_sl(vy32_l, shift2);
1252  vy32_r = vec_sl(vy32_r, shift2);
1253 
1254  vy32_l = vec_sub(vy32_l, y_offset);
1255  vy32_r = vec_sub(vy32_r, y_offset);
1256  vy32_l = vec_mul(vy32_l, y_coeff);
1257  vy32_r = vec_mul(vy32_r, y_coeff);
1258  vy32_l = vec_add(vy32_l, y_add);
1259  vy32_r = vec_add(vy32_r, y_add);
1260 
1261  // Second UV half
1262  vud32_l = vec_perm(vu32_r, vu32_r, doubleleft);
1263  vud32_r = vec_perm(vu32_r, vu32_r, doubleright);
1264  vvd32_l = vec_perm(vv32_r, vv32_r, doubleleft);
1265  vvd32_r = vec_perm(vv32_r, vv32_r, doubleright);
1266 
1267  R_l = vec_mul(vvd32_l, v2r_coeff);
1268  R_l = vec_add(R_l, vy32_l);
1269  R_r = vec_mul(vvd32_r, v2r_coeff);
1270  R_r = vec_add(R_r, vy32_r);
1271  G_l = vec_mul(vvd32_l, v2g_coeff);
1272  tmp32 = vec_mul(vud32_l, u2g_coeff);
1273  G_l = vec_add(G_l, vy32_l);
1274  G_l = vec_add(G_l, tmp32);
1275  G_r = vec_mul(vvd32_r, v2g_coeff);
1276  tmp32 = vec_mul(vud32_r, u2g_coeff);
1277  G_r = vec_add(G_r, vy32_r);
1278  G_r = vec_add(G_r, tmp32);
1279 
1280  B_l = vec_mul(vud32_l, u2b_coeff);
1281  B_l = vec_add(B_l, vy32_l);
1282  B_r = vec_mul(vud32_r, u2b_coeff);
1283  B_r = vec_add(B_r, vy32_r);
1284 
1285  WRITERGB
1286  }
1287 }
1288 
1289 #undef WRITERGB
1290 
1291 #define YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha) \
1292 static void name ## ext ## _X_vsx(SwsContext *c, const int16_t *lumFilter, \
1293  const int16_t **lumSrc, int lumFilterSize, \
1294  const int16_t *chrFilter, const int16_t **chrUSrc, \
1295  const int16_t **chrVSrc, int chrFilterSize, \
1296  const int16_t **alpSrc, uint8_t *dest, int dstW, \
1297  int y) \
1298 { \
1299  name ## base ## _X_vsx_template(c, lumFilter, lumSrc, lumFilterSize, \
1300  chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
1301  alpSrc, dest, dstW, y, fmt, hasAlpha); \
1302 }
1303 
1304 #define YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha) \
1305 static void name ## ext ## _2_vsx(SwsContext *c, const int16_t *buf[2], \
1306  const int16_t *ubuf[2], const int16_t *vbuf[2], \
1307  const int16_t *abuf[2], uint8_t *dest, int dstW, \
1308  int yalpha, int uvalpha, int y) \
1309 { \
1310  name ## base ## _2_vsx_template(c, buf, ubuf, vbuf, abuf, \
1311  dest, dstW, yalpha, uvalpha, y, fmt, hasAlpha); \
1312 }
1313 
1314 #define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha) \
1315 static void name ## ext ## _1_vsx(SwsContext *c, const int16_t *buf0, \
1316  const int16_t *ubuf[2], const int16_t *vbuf[2], \
1317  const int16_t *abuf0, uint8_t *dest, int dstW, \
1318  int uvalpha, int y) \
1319 { \
1320  name ## base ## _1_vsx_template(c, buf0, ubuf, vbuf, abuf0, dest, \
1321  dstW, uvalpha, y, fmt, hasAlpha); \
1322 }
1323 
1324 YUV2RGBWRAPPER(yuv2, rgb, bgrx32, AV_PIX_FMT_BGRA, 0)
1325 YUV2RGBWRAPPER(yuv2, rgb, rgbx32, AV_PIX_FMT_RGBA, 0)
1326 YUV2RGBWRAPPER(yuv2, rgb, xrgb32, AV_PIX_FMT_ARGB, 0)
1327 YUV2RGBWRAPPER(yuv2, rgb, xbgr32, AV_PIX_FMT_ABGR, 0)
1328 
1329 YUV2RGBWRAPPER(yuv2, rgb, rgb24, AV_PIX_FMT_RGB24, 0)
1330 YUV2RGBWRAPPER(yuv2, rgb, bgr24, AV_PIX_FMT_BGR24, 0)
1331 
1332 YUV2RGBWRAPPERX2(yuv2, rgb, bgrx32, AV_PIX_FMT_BGRA, 0)
1333 YUV2RGBWRAPPERX2(yuv2, rgb, rgbx32, AV_PIX_FMT_RGBA, 0)
1334 YUV2RGBWRAPPERX2(yuv2, rgb, xrgb32, AV_PIX_FMT_ARGB, 0)
1335 YUV2RGBWRAPPERX2(yuv2, rgb, xbgr32, AV_PIX_FMT_ABGR, 0)
1336 
1337 YUV2RGBWRAPPERX2(yuv2, rgb, rgb24, AV_PIX_FMT_RGB24, 0)
1338 YUV2RGBWRAPPERX2(yuv2, rgb, bgr24, AV_PIX_FMT_BGR24, 0)
1339 
1340 YUV2RGBWRAPPER(yuv2, rgb_full, bgrx32_full, AV_PIX_FMT_BGRA, 0)
1341 YUV2RGBWRAPPER(yuv2, rgb_full, rgbx32_full, AV_PIX_FMT_RGBA, 0)
1342 YUV2RGBWRAPPER(yuv2, rgb_full, xrgb32_full, AV_PIX_FMT_ARGB, 0)
1343 YUV2RGBWRAPPER(yuv2, rgb_full, xbgr32_full, AV_PIX_FMT_ABGR, 0)
1344 
1345 YUV2RGBWRAPPER(yuv2, rgb_full, rgb24_full, AV_PIX_FMT_RGB24, 0)
1346 YUV2RGBWRAPPER(yuv2, rgb_full, bgr24_full, AV_PIX_FMT_BGR24, 0)
1347 
1348 YUV2RGBWRAPPERX2(yuv2, rgb_full, bgrx32_full, AV_PIX_FMT_BGRA, 0)
1349 YUV2RGBWRAPPERX2(yuv2, rgb_full, rgbx32_full, AV_PIX_FMT_RGBA, 0)
1350 YUV2RGBWRAPPERX2(yuv2, rgb_full, xrgb32_full, AV_PIX_FMT_ARGB, 0)
1351 YUV2RGBWRAPPERX2(yuv2, rgb_full, xbgr32_full, AV_PIX_FMT_ABGR, 0)
1352 
1353 YUV2RGBWRAPPERX2(yuv2, rgb_full, rgb24_full, AV_PIX_FMT_RGB24, 0)
1354 YUV2RGBWRAPPERX2(yuv2, rgb_full, bgr24_full, AV_PIX_FMT_BGR24, 0)
1355 
1356 YUV2RGBWRAPPERX(yuv2, rgb_full, bgrx32_full, AV_PIX_FMT_BGRA, 0)
1357 YUV2RGBWRAPPERX(yuv2, rgb_full, rgbx32_full, AV_PIX_FMT_RGBA, 0)
1358 YUV2RGBWRAPPERX(yuv2, rgb_full, xrgb32_full, AV_PIX_FMT_ARGB, 0)
1359 YUV2RGBWRAPPERX(yuv2, rgb_full, xbgr32_full, AV_PIX_FMT_ABGR, 0)
1360 
1361 YUV2RGBWRAPPERX(yuv2, rgb_full, rgb24_full, AV_PIX_FMT_RGB24, 0)
1362 YUV2RGBWRAPPERX(yuv2, rgb_full, bgr24_full, AV_PIX_FMT_BGR24, 0)
1363 
1364 static av_always_inline void
1365 write422(const vec_s16 vy1, const vec_s16 vy2,
1366  const vec_s16 vu, const vec_s16 vv,
1367  uint8_t *dest, const enum AVPixelFormat target)
1368 {
1369  vec_u8 vd1, vd2, tmp;
1370  const vec_u8 yuyv1 = (vec_u8) {
1371  0x0, 0x10, 0x1, 0x18,
1372  0x2, 0x11, 0x3, 0x19,
1373  0x4, 0x12, 0x5, 0x1a,
1374  0x6, 0x13, 0x7, 0x1b };
1375  const vec_u8 yuyv2 = (vec_u8) {
1376  0x8, 0x14, 0x9, 0x1c,
1377  0xa, 0x15, 0xb, 0x1d,
1378  0xc, 0x16, 0xd, 0x1e,
1379  0xe, 0x17, 0xf, 0x1f };
1380  const vec_u8 yvyu1 = (vec_u8) {
1381  0x0, 0x18, 0x1, 0x10,
1382  0x2, 0x19, 0x3, 0x11,
1383  0x4, 0x1a, 0x5, 0x12,
1384  0x6, 0x1b, 0x7, 0x13 };
1385  const vec_u8 yvyu2 = (vec_u8) {
1386  0x8, 0x1c, 0x9, 0x14,
1387  0xa, 0x1d, 0xb, 0x15,
1388  0xc, 0x1e, 0xd, 0x16,
1389  0xe, 0x1f, 0xf, 0x17 };
1390  const vec_u8 uyvy1 = (vec_u8) {
1391  0x10, 0x0, 0x18, 0x1,
1392  0x11, 0x2, 0x19, 0x3,
1393  0x12, 0x4, 0x1a, 0x5,
1394  0x13, 0x6, 0x1b, 0x7 };
1395  const vec_u8 uyvy2 = (vec_u8) {
1396  0x14, 0x8, 0x1c, 0x9,
1397  0x15, 0xa, 0x1d, 0xb,
1398  0x16, 0xc, 0x1e, 0xd,
1399  0x17, 0xe, 0x1f, 0xf };
1400 
1401  vd1 = vec_packsu(vy1, vy2);
1402  vd2 = vec_packsu(vu, vv);
1403 
1404  switch (target) {
1405  case AV_PIX_FMT_YUYV422:
1406  tmp = vec_perm(vd1, vd2, yuyv1);
1407  vec_st(tmp, 0, dest);
1408  tmp = vec_perm(vd1, vd2, yuyv2);
1409  vec_st(tmp, 16, dest);
1410  break;
1411  case AV_PIX_FMT_YVYU422:
1412  tmp = vec_perm(vd1, vd2, yvyu1);
1413  vec_st(tmp, 0, dest);
1414  tmp = vec_perm(vd1, vd2, yvyu2);
1415  vec_st(tmp, 16, dest);
1416  break;
1417  case AV_PIX_FMT_UYVY422:
1418  tmp = vec_perm(vd1, vd2, uyvy1);
1419  vec_st(tmp, 0, dest);
1420  tmp = vec_perm(vd1, vd2, uyvy2);
1421  vec_st(tmp, 16, dest);
1422  break;
1423  }
1424 }
1425 
1426 static av_always_inline void
1427 yuv2422_X_vsx_template(SwsContext *c, const int16_t *lumFilter,
1428  const int16_t **lumSrc, int lumFilterSize,
1429  const int16_t *chrFilter, const int16_t **chrUSrc,
1430  const int16_t **chrVSrc, int chrFilterSize,
1431  const int16_t **alpSrc, uint8_t *dest, int dstW,
1432  int y, enum AVPixelFormat target)
1433 {
1434  int i, j;
1435  vec_s16 vy1, vy2, vu, vv;
1436  vec_s32 vy32[4], vu32[2], vv32[2], tmp, tmp2, tmp3, tmp4;
1437  vec_s16 vlumFilter[MAX_FILTER_SIZE], vchrFilter[MAX_FILTER_SIZE];
1438  const vec_s32 start = vec_splats(1 << 18);
1439  const vec_u32 shift19 = vec_splats(19U);
1440 
1441  for (i = 0; i < lumFilterSize; i++)
1442  vlumFilter[i] = vec_splats(lumFilter[i]);
1443  for (i = 0; i < chrFilterSize; i++)
1444  vchrFilter[i] = vec_splats(chrFilter[i]);
1445 
1446  for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1447  vy32[0] =
1448  vy32[1] =
1449  vy32[2] =
1450  vy32[3] =
1451  vu32[0] =
1452  vu32[1] =
1453  vv32[0] =
1454  vv32[1] = start;
1455 
1456  for (j = 0; j < lumFilterSize; j++) {
1457  vv = vec_ld(0, &lumSrc[j][i * 2]);
1458  tmp = vec_mule(vv, vlumFilter[j]);
1459  tmp2 = vec_mulo(vv, vlumFilter[j]);
1460  tmp3 = vec_mergeh(tmp, tmp2);
1461  tmp4 = vec_mergel(tmp, tmp2);
1462 
1463  vy32[0] = vec_adds(vy32[0], tmp3);
1464  vy32[1] = vec_adds(vy32[1], tmp4);
1465 
1466  vv = vec_ld(0, &lumSrc[j][(i + 4) * 2]);
1467  tmp = vec_mule(vv, vlumFilter[j]);
1468  tmp2 = vec_mulo(vv, vlumFilter[j]);
1469  tmp3 = vec_mergeh(tmp, tmp2);
1470  tmp4 = vec_mergel(tmp, tmp2);
1471 
1472  vy32[2] = vec_adds(vy32[2], tmp3);
1473  vy32[3] = vec_adds(vy32[3], tmp4);
1474  }
1475 
1476  for (j = 0; j < chrFilterSize; j++) {
1477  vv = vec_ld(0, &chrUSrc[j][i]);
1478  tmp = vec_mule(vv, vchrFilter[j]);
1479  tmp2 = vec_mulo(vv, vchrFilter[j]);
1480  tmp3 = vec_mergeh(tmp, tmp2);
1481  tmp4 = vec_mergel(tmp, tmp2);
1482 
1483  vu32[0] = vec_adds(vu32[0], tmp3);
1484  vu32[1] = vec_adds(vu32[1], tmp4);
1485 
1486  vv = vec_ld(0, &chrVSrc[j][i]);
1487  tmp = vec_mule(vv, vchrFilter[j]);
1488  tmp2 = vec_mulo(vv, vchrFilter[j]);
1489  tmp3 = vec_mergeh(tmp, tmp2);
1490  tmp4 = vec_mergel(tmp, tmp2);
1491 
1492  vv32[0] = vec_adds(vv32[0], tmp3);
1493  vv32[1] = vec_adds(vv32[1], tmp4);
1494  }
1495 
1496  for (j = 0; j < 4; j++) {
1497  vy32[j] = vec_sra(vy32[j], shift19);
1498  }
1499  for (j = 0; j < 2; j++) {
1500  vu32[j] = vec_sra(vu32[j], shift19);
1501  vv32[j] = vec_sra(vv32[j], shift19);
1502  }
1503 
1504  vy1 = vec_packs(vy32[0], vy32[1]);
1505  vy2 = vec_packs(vy32[2], vy32[3]);
1506  vu = vec_packs(vu32[0], vu32[1]);
1507  vv = vec_packs(vv32[0], vv32[1]);
1508 
1509  write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1510  }
1511 }
1512 
1513 #define SETUP(x, buf0, buf1, alpha) { \
1514  x = vec_ld(0, buf0); \
1515  tmp = vec_mule(x, alpha); \
1516  tmp2 = vec_mulo(x, alpha); \
1517  tmp3 = vec_mergeh(tmp, tmp2); \
1518  tmp4 = vec_mergel(tmp, tmp2); \
1519 \
1520  x = vec_ld(0, buf1); \
1521  tmp = vec_mule(x, alpha); \
1522  tmp2 = vec_mulo(x, alpha); \
1523  tmp5 = vec_mergeh(tmp, tmp2); \
1524  tmp6 = vec_mergel(tmp, tmp2); \
1525 \
1526  tmp3 = vec_add(tmp3, tmp5); \
1527  tmp4 = vec_add(tmp4, tmp6); \
1528 \
1529  tmp3 = vec_sra(tmp3, shift19); \
1530  tmp4 = vec_sra(tmp4, shift19); \
1531  x = vec_packs(tmp3, tmp4); \
1532 }
1533 
1534 static av_always_inline void
1535 yuv2422_2_vsx_template(SwsContext *c, const int16_t *buf[2],
1536  const int16_t *ubuf[2], const int16_t *vbuf[2],
1537  const int16_t *abuf[2], uint8_t *dest, int dstW,
1538  int yalpha, int uvalpha, int y,
1539  enum AVPixelFormat target)
1540 {
1541  const int16_t *buf0 = buf[0], *buf1 = buf[1],
1542  *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
1543  *vbuf0 = vbuf[0], *vbuf1 = vbuf[1];
1544  const int16_t yalpha1 = 4096 - yalpha;
1545  const int16_t uvalpha1 = 4096 - uvalpha;
1546  vec_s16 vy1, vy2, vu, vv;
1547  vec_s32 tmp, tmp2, tmp3, tmp4, tmp5, tmp6;
1548  const vec_s16 vyalpha1 = vec_splats(yalpha1);
1549  const vec_s16 vuvalpha1 = vec_splats(uvalpha1);
1550  const vec_u32 shift19 = vec_splats(19U);
1551  int i;
1552  av_assert2(yalpha <= 4096U);
1553  av_assert2(uvalpha <= 4096U);
1554 
1555  for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1556 
1557  SETUP(vy1, &buf0[i * 2], &buf1[i * 2], vyalpha1)
1558  SETUP(vy2, &buf0[(i + 4) * 2], &buf1[(i + 4) * 2], vyalpha1)
1559  SETUP(vu, &ubuf0[i], &ubuf1[i], vuvalpha1)
1560  SETUP(vv, &vbuf0[i], &vbuf1[i], vuvalpha1)
1561 
1562  write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1563  }
1564 }
1565 
1566 #undef SETUP
1567 
1568 static av_always_inline void
1569 yuv2422_1_vsx_template(SwsContext *c, const int16_t *buf0,
1570  const int16_t *ubuf[2], const int16_t *vbuf[2],
1571  const int16_t *abuf0, uint8_t *dest, int dstW,
1572  int uvalpha, int y, enum AVPixelFormat target)
1573 {
1574  const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
1575  vec_s16 vy1, vy2, vu, vv, tmp;
1576  const vec_s16 add64 = vec_splats((int16_t) 64);
1577  const vec_s16 add128 = vec_splats((int16_t) 128);
1578  const vec_u16 shift7 = vec_splat_u16(7);
1579  const vec_u16 shift8 = vec_splat_u16(8);
1580  int i;
1581 
1582  if (uvalpha < 2048) {
1583  for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1584  vy1 = vec_ld(0, &buf0[i * 2]);
1585  vy2 = vec_ld(0, &buf0[(i + 4) * 2]);
1586  vu = vec_ld(0, &ubuf0[i]);
1587  vv = vec_ld(0, &vbuf0[i]);
1588 
1589  vy1 = vec_add(vy1, add64);
1590  vy2 = vec_add(vy2, add64);
1591  vu = vec_add(vu, add64);
1592  vv = vec_add(vv, add64);
1593 
1594  vy1 = vec_sra(vy1, shift7);
1595  vy2 = vec_sra(vy2, shift7);
1596  vu = vec_sra(vu, shift7);
1597  vv = vec_sra(vv, shift7);
1598 
1599  write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1600  }
1601  } else {
1602  const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
1603  for (i = 0; i < ((dstW + 1) >> 1); i += 8) {
1604  vy1 = vec_ld(0, &buf0[i * 2]);
1605  vy2 = vec_ld(0, &buf0[(i + 4) * 2]);
1606  vu = vec_ld(0, &ubuf0[i]);
1607  tmp = vec_ld(0, &ubuf1[i]);
1608  vu = vec_adds(vu, tmp);
1609  vv = vec_ld(0, &vbuf0[i]);
1610  tmp = vec_ld(0, &vbuf1[i]);
1611  vv = vec_adds(vv, tmp);
1612 
1613  vy1 = vec_add(vy1, add64);
1614  vy2 = vec_add(vy2, add64);
1615  vu = vec_adds(vu, add128);
1616  vv = vec_adds(vv, add128);
1617 
1618  vy1 = vec_sra(vy1, shift7);
1619  vy2 = vec_sra(vy2, shift7);
1620  vu = vec_sra(vu, shift8);
1621  vv = vec_sra(vv, shift8);
1622 
1623  write422(vy1, vy2, vu, vv, &dest[i * 4], target);
1624  }
1625  }
1626 }
1627 
1628 #define YUV2PACKEDWRAPPERX(name, base, ext, fmt) \
1629 static void name ## ext ## _X_vsx(SwsContext *c, const int16_t *lumFilter, \
1630  const int16_t **lumSrc, int lumFilterSize, \
1631  const int16_t *chrFilter, const int16_t **chrUSrc, \
1632  const int16_t **chrVSrc, int chrFilterSize, \
1633  const int16_t **alpSrc, uint8_t *dest, int dstW, \
1634  int y) \
1635 { \
1636  name ## base ## _X_vsx_template(c, lumFilter, lumSrc, lumFilterSize, \
1637  chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
1638  alpSrc, dest, dstW, y, fmt); \
1639 }
1640 
1641 #define YUV2PACKEDWRAPPER2(name, base, ext, fmt) \
1642 YUV2PACKEDWRAPPERX(name, base, ext, fmt) \
1643 static void name ## ext ## _2_vsx(SwsContext *c, const int16_t *buf[2], \
1644  const int16_t *ubuf[2], const int16_t *vbuf[2], \
1645  const int16_t *abuf[2], uint8_t *dest, int dstW, \
1646  int yalpha, int uvalpha, int y) \
1647 { \
1648  name ## base ## _2_vsx_template(c, buf, ubuf, vbuf, abuf, \
1649  dest, dstW, yalpha, uvalpha, y, fmt); \
1650 }
1651 
1652 #define YUV2PACKEDWRAPPER(name, base, ext, fmt) \
1653 YUV2PACKEDWRAPPER2(name, base, ext, fmt) \
1654 static void name ## ext ## _1_vsx(SwsContext *c, const int16_t *buf0, \
1655  const int16_t *ubuf[2], const int16_t *vbuf[2], \
1656  const int16_t *abuf0, uint8_t *dest, int dstW, \
1657  int uvalpha, int y) \
1658 { \
1659  name ## base ## _1_vsx_template(c, buf0, ubuf, vbuf, \
1660  abuf0, dest, dstW, uvalpha, \
1661  y, fmt); \
1662 }
1663 
1664 YUV2PACKEDWRAPPER(yuv2, 422, yuyv422, AV_PIX_FMT_YUYV422)
1665 YUV2PACKEDWRAPPER(yuv2, 422, yvyu422, AV_PIX_FMT_YVYU422)
1666 YUV2PACKEDWRAPPER(yuv2, 422, uyvy422, AV_PIX_FMT_UYVY422)
1667 
1668 static void hyscale_fast_vsx(SwsContext *c, int16_t *dst, int dstWidth,
1669  const uint8_t *src, int srcW, int xInc)
1670 {
1671  int i;
1672  unsigned int xpos = 0, xx;
1673  vec_u8 vin, vin2, vperm;
1674  vec_s8 vmul, valpha;
1675  vec_s16 vtmp, vtmp2, vtmp3, vtmp4;
1676  vec_u16 vd_l, vd_r, vcoord16[2];
1677  vec_u32 vcoord[4];
1678  const vec_u32 vadd = (vec_u32) {
1679  0,
1680  xInc * 1,
1681  xInc * 2,
1682  xInc * 3,
1683  };
1684  const vec_u16 vadd16 = (vec_u16) { // Modulo math
1685  0,
1686  xInc * 1,
1687  xInc * 2,
1688  xInc * 3,
1689  xInc * 4,
1690  xInc * 5,
1691  xInc * 6,
1692  xInc * 7,
1693  };
1694  const vec_u32 vshift16 = vec_splats((uint32_t) 16);
1695  const vec_u16 vshift9 = vec_splat_u16(9);
1696  const vec_u8 vzero = vec_splat_u8(0);
1697  const vec_u16 vshift = vec_splat_u16(7);
1698 
1699  for (i = 0; i < dstWidth; i += 16) {
1700  vcoord16[0] = vec_splats((uint16_t) xpos);
1701  vcoord16[1] = vec_splats((uint16_t) (xpos + xInc * 8));
1702 
1703  vcoord16[0] = vec_add(vcoord16[0], vadd16);
1704  vcoord16[1] = vec_add(vcoord16[1], vadd16);
1705 
1706  vcoord16[0] = vec_sr(vcoord16[0], vshift9);
1707  vcoord16[1] = vec_sr(vcoord16[1], vshift9);
1708  valpha = (vec_s8) vec_pack(vcoord16[0], vcoord16[1]);
1709 
1710  xx = xpos >> 16;
1711  vin = vec_vsx_ld(0, &src[xx]);
1712 
1713  vcoord[0] = vec_splats(xpos & 0xffff);
1714  vcoord[1] = vec_splats((xpos & 0xffff) + xInc * 4);
1715  vcoord[2] = vec_splats((xpos & 0xffff) + xInc * 8);
1716  vcoord[3] = vec_splats((xpos & 0xffff) + xInc * 12);
1717 
1718  vcoord[0] = vec_add(vcoord[0], vadd);
1719  vcoord[1] = vec_add(vcoord[1], vadd);
1720  vcoord[2] = vec_add(vcoord[2], vadd);
1721  vcoord[3] = vec_add(vcoord[3], vadd);
1722 
1723  vcoord[0] = vec_sr(vcoord[0], vshift16);
1724  vcoord[1] = vec_sr(vcoord[1], vshift16);
1725  vcoord[2] = vec_sr(vcoord[2], vshift16);
1726  vcoord[3] = vec_sr(vcoord[3], vshift16);
1727 
1728  vcoord16[0] = vec_pack(vcoord[0], vcoord[1]);
1729  vcoord16[1] = vec_pack(vcoord[2], vcoord[3]);
1730  vperm = vec_pack(vcoord16[0], vcoord16[1]);
1731 
1732  vin = vec_perm(vin, vin, vperm);
1733 
1734  vin2 = vec_vsx_ld(1, &src[xx]);
1735  vin2 = vec_perm(vin2, vin2, vperm);
1736 
1737  vmul = (vec_s8) vec_sub(vin2, vin);
1738  vtmp = vec_mule(vmul, valpha);
1739  vtmp2 = vec_mulo(vmul, valpha);
1740  vtmp3 = vec_mergeh(vtmp, vtmp2);
1741  vtmp4 = vec_mergel(vtmp, vtmp2);
1742 
1743  vd_l = (vec_u16) vec_mergeh(vin, vzero);
1744  vd_r = (vec_u16) vec_mergel(vin, vzero);
1745  vd_l = vec_sl(vd_l, vshift);
1746  vd_r = vec_sl(vd_r, vshift);
1747 
1748  vd_l = vec_add(vd_l, (vec_u16) vtmp3);
1749  vd_r = vec_add(vd_r, (vec_u16) vtmp4);
1750 
1751  vec_st((vec_s16) vd_l, 0, &dst[i]);
1752  vec_st((vec_s16) vd_r, 0, &dst[i + 8]);
1753 
1754  xpos += xInc * 16;
1755  }
1756  for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)
1757  dst[i] = src[srcW-1]*128;
1758 }
1759 
1760 #define HCSCALE(in, out) \
1761  vin = vec_vsx_ld(0, &in[xx]); \
1762  vin = vec_perm(vin, vin, vperm); \
1763 \
1764  vin2 = vec_vsx_ld(1, &in[xx]); \
1765  vin2 = vec_perm(vin2, vin2, vperm); \
1766 \
1767  vtmp = vec_mule(vin, valphaxor); \
1768  vtmp2 = vec_mulo(vin, valphaxor); \
1769  vtmp3 = vec_mergeh(vtmp, vtmp2); \
1770  vtmp4 = vec_mergel(vtmp, vtmp2); \
1771 \
1772  vtmp = vec_mule(vin2, valpha); \
1773  vtmp2 = vec_mulo(vin2, valpha); \
1774  vd_l = vec_mergeh(vtmp, vtmp2); \
1775  vd_r = vec_mergel(vtmp, vtmp2); \
1776 \
1777  vd_l = vec_add(vd_l, vtmp3); \
1778  vd_r = vec_add(vd_r, vtmp4); \
1779 \
1780  vec_st((vec_s16) vd_l, 0, &out[i]); \
1781  vec_st((vec_s16) vd_r, 0, &out[i + 8])
1782 
1783 static void hcscale_fast_vsx(SwsContext *c, int16_t *dst1, int16_t *dst2,
1784  int dstWidth, const uint8_t *src1,
1785  const uint8_t *src2, int srcW, int xInc)
1786 {
1787  int i;
1788  unsigned int xpos = 0, xx;
1789  vec_u8 vin, vin2, vperm;
1790  vec_u8 valpha, valphaxor;
1791  vec_u16 vtmp, vtmp2, vtmp3, vtmp4;
1792  vec_u16 vd_l, vd_r, vcoord16[2];
1793  vec_u32 vcoord[4];
1794  const vec_u8 vxor = vec_splats((uint8_t) 127);
1795  const vec_u32 vadd = (vec_u32) {
1796  0,
1797  xInc * 1,
1798  xInc * 2,
1799  xInc * 3,
1800  };
1801  const vec_u16 vadd16 = (vec_u16) { // Modulo math
1802  0,
1803  xInc * 1,
1804  xInc * 2,
1805  xInc * 3,
1806  xInc * 4,
1807  xInc * 5,
1808  xInc * 6,
1809  xInc * 7,
1810  };
1811  const vec_u32 vshift16 = vec_splats((uint32_t) 16);
1812  const vec_u16 vshift9 = vec_splat_u16(9);
1813 
1814  for (i = 0; i < dstWidth; i += 16) {
1815  vcoord16[0] = vec_splats((uint16_t) xpos);
1816  vcoord16[1] = vec_splats((uint16_t) (xpos + xInc * 8));
1817 
1818  vcoord16[0] = vec_add(vcoord16[0], vadd16);
1819  vcoord16[1] = vec_add(vcoord16[1], vadd16);
1820 
1821  vcoord16[0] = vec_sr(vcoord16[0], vshift9);
1822  vcoord16[1] = vec_sr(vcoord16[1], vshift9);
1823  valpha = vec_pack(vcoord16[0], vcoord16[1]);
1824  valphaxor = vec_xor(valpha, vxor);
1825 
1826  xx = xpos >> 16;
1827 
1828  vcoord[0] = vec_splats(xpos & 0xffff);
1829  vcoord[1] = vec_splats((xpos & 0xffff) + xInc * 4);
1830  vcoord[2] = vec_splats((xpos & 0xffff) + xInc * 8);
1831  vcoord[3] = vec_splats((xpos & 0xffff) + xInc * 12);
1832 
1833  vcoord[0] = vec_add(vcoord[0], vadd);
1834  vcoord[1] = vec_add(vcoord[1], vadd);
1835  vcoord[2] = vec_add(vcoord[2], vadd);
1836  vcoord[3] = vec_add(vcoord[3], vadd);
1837 
1838  vcoord[0] = vec_sr(vcoord[0], vshift16);
1839  vcoord[1] = vec_sr(vcoord[1], vshift16);
1840  vcoord[2] = vec_sr(vcoord[2], vshift16);
1841  vcoord[3] = vec_sr(vcoord[3], vshift16);
1842 
1843  vcoord16[0] = vec_pack(vcoord[0], vcoord[1]);
1844  vcoord16[1] = vec_pack(vcoord[2], vcoord[3]);
1845  vperm = vec_pack(vcoord16[0], vcoord16[1]);
1846 
1847  HCSCALE(src1, dst1);
1848  HCSCALE(src2, dst2);
1849 
1850  xpos += xInc * 16;
1851  }
1852  for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) {
1853  dst1[i] = src1[srcW-1]*128;
1854  dst2[i] = src2[srcW-1]*128;
1855  }
1856 }
1857 
1858 #undef HCSCALE
1859 
1860 static void hScale8To19_vsx(SwsContext *c, int16_t *_dst, int dstW,
1861  const uint8_t *src, const int16_t *filter,
1862  const int32_t *filterPos, int filterSize)
1863 {
1864  int i, j;
1865  int32_t *dst = (int32_t *) _dst;
1866  vec_s16 vfilter, vin;
1867  vec_u8 vin8;
1868  vec_s32 vout;
1869  const vec_u8 vzero = vec_splat_u8(0);
1870  const vec_u8 vunusedtab[8] = {
1871  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1872  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
1873  (vec_u8) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
1874  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1875  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
1876  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1877  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
1878  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1879  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1880  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1881  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1882  0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1883  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1884  0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
1885  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1886  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
1887  };
1888  const vec_u8 vunused = vunusedtab[filterSize % 8];
1889 
1890  if (filterSize == 1) {
1891  for (i = 0; i < dstW; i++) {
1892  int srcPos = filterPos[i];
1893  int val = 0;
1894  for (j = 0; j < filterSize; j++) {
1895  val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
1896  }
1897  dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
1898  }
1899  } else {
1900  for (i = 0; i < dstW; i++) {
1901  const int srcPos = filterPos[i];
1902  vout = vec_splat_s32(0);
1903  for (j = 0; j < filterSize; j += 8) {
1904  vin8 = vec_vsx_ld(0, &src[srcPos + j]);
1905  vin = (vec_s16) vec_mergeh(vin8, vzero);
1906  if (j + 8 > filterSize) // Remove the unused elements on the last round
1907  vin = vec_perm(vin, (vec_s16) vzero, vunused);
1908 
1909  vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
1910  vout = vec_msums(vin, vfilter, vout);
1911  }
1912  vout = vec_sums(vout, (vec_s32) vzero);
1913  dst[i] = FFMIN(vout[3] >> 3, (1 << 19) - 1);
1914  }
1915  }
1916 }
1917 
1918 static void hScale16To19_vsx(SwsContext *c, int16_t *_dst, int dstW,
1919  const uint8_t *_src, const int16_t *filter,
1920  const int32_t *filterPos, int filterSize)
1921 {
1923  int i, j;
1924  int32_t *dst = (int32_t *) _dst;
1925  const uint16_t *src = (const uint16_t *) _src;
1926  int bits = desc->comp[0].depth - 1;
1927  int sh = bits - 4;
1928  vec_s16 vfilter, vin;
1929  vec_s32 vout, vtmp, vtmp2, vfilter32_l, vfilter32_r;
1930  const vec_u8 vzero = vec_splat_u8(0);
1931  const vec_u8 vunusedtab[8] = {
1932  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1933  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
1934  (vec_u8) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
1935  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1936  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
1937  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1938  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
1939  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1940  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1941  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1942  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1943  0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
1944  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1945  0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
1946  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
1947  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
1948  };
1949  const vec_u8 vunused = vunusedtab[filterSize % 8];
1950 
1951  if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
1952  sh = 9;
1953  } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
1954  sh = 16 - 1 - 4;
1955  }
1956 
1957  if (filterSize == 1) {
1958  for (i = 0; i < dstW; i++) {
1959  int srcPos = filterPos[i];
1960  int val = 0;
1961 
1962  for (j = 0; j < filterSize; j++) {
1963  val += src[srcPos + j] * filter[filterSize * i + j];
1964  }
1965  // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
1966  dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
1967  }
1968  } else {
1969  for (i = 0; i < dstW; i++) {
1970  const int srcPos = filterPos[i];
1971  vout = vec_splat_s32(0);
1972  for (j = 0; j < filterSize; j += 8) {
1973  vin = (vec_s16) vec_vsx_ld(0, &src[srcPos + j]);
1974  if (j + 8 > filterSize) // Remove the unused elements on the last round
1975  vin = vec_perm(vin, (vec_s16) vzero, vunused);
1976 
1977  vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
1978  vfilter32_l = vec_unpackh(vfilter);
1979  vfilter32_r = vec_unpackl(vfilter);
1980 
1981  vtmp = (vec_s32) vec_mergeh(vin, (vec_s16) vzero);
1982  vtmp2 = (vec_s32) vec_mergel(vin, (vec_s16) vzero);
1983 
1984  vtmp = vec_mul(vtmp, vfilter32_l);
1985  vtmp2 = vec_mul(vtmp2, vfilter32_r);
1986 
1987  vout = vec_adds(vout, vtmp);
1988  vout = vec_adds(vout, vtmp2);
1989  }
1990  vout = vec_sums(vout, (vec_s32) vzero);
1991  dst[i] = FFMIN(vout[3] >> sh, (1 << 19) - 1);
1992  }
1993  }
1994 }
1995 
1996 static void hScale16To15_vsx(SwsContext *c, int16_t *dst, int dstW,
1997  const uint8_t *_src, const int16_t *filter,
1998  const int32_t *filterPos, int filterSize)
1999 {
2001  int i, j;
2002  const uint16_t *src = (const uint16_t *) _src;
2003  int sh = desc->comp[0].depth - 1;
2004  vec_s16 vfilter, vin;
2005  vec_s32 vout, vtmp, vtmp2, vfilter32_l, vfilter32_r;
2006  const vec_u8 vzero = vec_splat_u8(0);
2007  const vec_u8 vunusedtab[8] = {
2008  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2009  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf},
2010  (vec_u8) {0x0, 0x1, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
2011  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2012  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x10, 0x10, 0x10, 0x10,
2013  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2014  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x10, 0x10,
2015  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2016  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2017  0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2018  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2019  0x8, 0x9, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10},
2020  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2021  0x8, 0x9, 0xa, 0xb, 0x10, 0x10, 0x10, 0x10},
2022  (vec_u8) {0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7,
2023  0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0x10, 0x10},
2024  };
2025  const vec_u8 vunused = vunusedtab[filterSize % 8];
2026 
2027  if (sh<15) {
2028  sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
2029  } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
2030  sh = 16 - 1;
2031  }
2032 
2033  if (filterSize == 1) {
2034  for (i = 0; i < dstW; i++) {
2035  int srcPos = filterPos[i];
2036  int val = 0;
2037 
2038  for (j = 0; j < filterSize; j++) {
2039  val += src[srcPos + j] * filter[filterSize * i + j];
2040  }
2041  // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
2042  dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
2043  }
2044  } else {
2045  for (i = 0; i < dstW; i++) {
2046  const int srcPos = filterPos[i];
2047  vout = vec_splat_s32(0);
2048  for (j = 0; j < filterSize; j += 8) {
2049  vin = (vec_s16) vec_vsx_ld(0, &src[srcPos + j]);
2050  if (j + 8 > filterSize) // Remove the unused elements on the last round
2051  vin = vec_perm(vin, (vec_s16) vzero, vunused);
2052 
2053  vfilter = vec_vsx_ld(0, &filter[filterSize * i + j]);
2054  vfilter32_l = vec_unpackh(vfilter);
2055  vfilter32_r = vec_unpackl(vfilter);
2056 
2057  vtmp = (vec_s32) vec_mergeh(vin, (vec_s16) vzero);
2058  vtmp2 = (vec_s32) vec_mergel(vin, (vec_s16) vzero);
2059 
2060  vtmp = vec_mul(vtmp, vfilter32_l);
2061  vtmp2 = vec_mul(vtmp2, vfilter32_r);
2062 
2063  vout = vec_adds(vout, vtmp);
2064  vout = vec_adds(vout, vtmp2);
2065  }
2066  vout = vec_sums(vout, (vec_s32) vzero);
2067  dst[i] = FFMIN(vout[3] >> sh, (1 << 15) - 1);
2068  }
2069  }
2070 }
2071 
2072 #endif /* !HAVE_BIGENDIAN */
2073 
2074 #endif /* HAVE_VSX */
2075 
2077 {
2078 #if HAVE_VSX
2079  enum AVPixelFormat dstFormat = c->dstFormat;
2080  const int cpu_flags = av_get_cpu_flags();
2081  const unsigned char power8 = HAVE_POWER8 && cpu_flags & AV_CPU_FLAG_POWER8;
2082 
2083  if (!(cpu_flags & AV_CPU_FLAG_VSX))
2084  return;
2085 
2086 #if !HAVE_BIGENDIAN
2087  if (c->srcBpc == 8) {
2088  if (c->dstBpc <= 14) {
2089  c->hyScale = c->hcScale = hScale_real_vsx;
2090  if (c->flags & SWS_FAST_BILINEAR && c->dstW >= c->srcW && c->chrDstW >= c->chrSrcW) {
2091  c->hyscale_fast = hyscale_fast_vsx;
2092  c->hcscale_fast = hcscale_fast_vsx;
2093  }
2094  } else {
2095  c->hyScale = c->hcScale = hScale8To19_vsx;
2096  }
2097  } else {
2098  if (power8) {
2099  c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_vsx
2100  : hScale16To15_vsx;
2101  }
2102  }
2103  if (!is16BPS(dstFormat) && !isNBPS(dstFormat) && !isSemiPlanarYUV(dstFormat) &&
2104  dstFormat != AV_PIX_FMT_GRAYF32BE && dstFormat != AV_PIX_FMT_GRAYF32LE &&
2105  !c->needAlpha) {
2106  c->yuv2planeX = yuv2planeX_vsx;
2107  }
2108 #endif
2109 
2110  if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->needAlpha) {
2111  switch (c->dstBpc) {
2112  case 8:
2113  c->yuv2plane1 = yuv2plane1_8_vsx;
2114  break;
2115 #if !HAVE_BIGENDIAN
2116  case 9:
2117  c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_9BE_vsx : yuv2plane1_9LE_vsx;
2118  c->yuv2planeX = isBE(dstFormat) ? yuv2planeX_9BE_vsx : yuv2planeX_9LE_vsx;
2119  break;
2120  case 10:
2121  c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_10BE_vsx : yuv2plane1_10LE_vsx;
2122  c->yuv2planeX = isBE(dstFormat) ? yuv2planeX_10BE_vsx : yuv2planeX_10LE_vsx;
2123  break;
2124  case 12:
2125  c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_12BE_vsx : yuv2plane1_12LE_vsx;
2126  c->yuv2planeX = isBE(dstFormat) ? yuv2planeX_12BE_vsx : yuv2planeX_12LE_vsx;
2127  break;
2128  case 14:
2129  c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_14BE_vsx : yuv2plane1_14LE_vsx;
2130  c->yuv2planeX = isBE(dstFormat) ? yuv2planeX_14BE_vsx : yuv2planeX_14LE_vsx;
2131  break;
2132  case 16:
2133  c->yuv2plane1 = isBE(dstFormat) ? yuv2plane1_16BE_vsx : yuv2plane1_16LE_vsx;
2134 #if HAVE_POWER8
2135  if (cpu_flags & AV_CPU_FLAG_POWER8) {
2136  c->yuv2planeX = isBE(dstFormat) ? yuv2planeX_16BE_vsx : yuv2planeX_16LE_vsx;
2137  }
2138 #endif /* HAVE_POWER8 */
2139  break;
2140 #endif /* !HAVE_BIGENDIAN */
2141  }
2142  }
2143 
2144  if (c->flags & SWS_BITEXACT)
2145  return;
2146 
2147 #if !HAVE_BIGENDIAN
2148  if (c->flags & SWS_FULL_CHR_H_INT) {
2149  switch (dstFormat) {
2150  case AV_PIX_FMT_RGB24:
2151  if (power8) {
2152  c->yuv2packed1 = yuv2rgb24_full_1_vsx;
2153  c->yuv2packed2 = yuv2rgb24_full_2_vsx;
2154  c->yuv2packedX = yuv2rgb24_full_X_vsx;
2155  }
2156  break;
2157  case AV_PIX_FMT_BGR24:
2158  if (power8) {
2159  c->yuv2packed1 = yuv2bgr24_full_1_vsx;
2160  c->yuv2packed2 = yuv2bgr24_full_2_vsx;
2161  c->yuv2packedX = yuv2bgr24_full_X_vsx;
2162  }
2163  break;
2164  case AV_PIX_FMT_BGRA:
2165  if (power8) {
2166  if (!c->needAlpha) {
2167  c->yuv2packed1 = yuv2bgrx32_full_1_vsx;
2168  c->yuv2packed2 = yuv2bgrx32_full_2_vsx;
2169  c->yuv2packedX = yuv2bgrx32_full_X_vsx;
2170  }
2171  }
2172  break;
2173  case AV_PIX_FMT_RGBA:
2174  if (power8) {
2175  if (!c->needAlpha) {
2176  c->yuv2packed1 = yuv2rgbx32_full_1_vsx;
2177  c->yuv2packed2 = yuv2rgbx32_full_2_vsx;
2178  c->yuv2packedX = yuv2rgbx32_full_X_vsx;
2179  }
2180  }
2181  break;
2182  case AV_PIX_FMT_ARGB:
2183  if (power8) {
2184  if (!c->needAlpha) {
2185  c->yuv2packed1 = yuv2xrgb32_full_1_vsx;
2186  c->yuv2packed2 = yuv2xrgb32_full_2_vsx;
2187  c->yuv2packedX = yuv2xrgb32_full_X_vsx;
2188  }
2189  }
2190  break;
2191  case AV_PIX_FMT_ABGR:
2192  if (power8) {
2193  if (!c->needAlpha) {
2194  c->yuv2packed1 = yuv2xbgr32_full_1_vsx;
2195  c->yuv2packed2 = yuv2xbgr32_full_2_vsx;
2196  c->yuv2packedX = yuv2xbgr32_full_X_vsx;
2197  }
2198  }
2199  break;
2200  }
2201  } else { /* !SWS_FULL_CHR_H_INT */
2202  switch (dstFormat) {
2203  case AV_PIX_FMT_YUYV422:
2204  c->yuv2packed1 = yuv2yuyv422_1_vsx;
2205  c->yuv2packed2 = yuv2yuyv422_2_vsx;
2206  c->yuv2packedX = yuv2yuyv422_X_vsx;
2207  break;
2208  case AV_PIX_FMT_YVYU422:
2209  c->yuv2packed1 = yuv2yvyu422_1_vsx;
2210  c->yuv2packed2 = yuv2yvyu422_2_vsx;
2211  c->yuv2packedX = yuv2yvyu422_X_vsx;
2212  break;
2213  case AV_PIX_FMT_UYVY422:
2214  c->yuv2packed1 = yuv2uyvy422_1_vsx;
2215  c->yuv2packed2 = yuv2uyvy422_2_vsx;
2216  c->yuv2packedX = yuv2uyvy422_X_vsx;
2217  break;
2218  case AV_PIX_FMT_BGRA:
2219  if (power8) {
2220  if (!c->needAlpha) {
2221  c->yuv2packed1 = yuv2bgrx32_1_vsx;
2222  c->yuv2packed2 = yuv2bgrx32_2_vsx;
2223  }
2224  }
2225  break;
2226  case AV_PIX_FMT_RGBA:
2227  if (power8) {
2228  if (!c->needAlpha) {
2229  c->yuv2packed1 = yuv2rgbx32_1_vsx;
2230  c->yuv2packed2 = yuv2rgbx32_2_vsx;
2231  }
2232  }
2233  break;
2234  case AV_PIX_FMT_ARGB:
2235  if (power8) {
2236  if (!c->needAlpha) {
2237  c->yuv2packed1 = yuv2xrgb32_1_vsx;
2238  c->yuv2packed2 = yuv2xrgb32_2_vsx;
2239  }
2240  }
2241  break;
2242  case AV_PIX_FMT_ABGR:
2243  if (power8) {
2244  if (!c->needAlpha) {
2245  c->yuv2packed1 = yuv2xbgr32_1_vsx;
2246  c->yuv2packed2 = yuv2xbgr32_2_vsx;
2247  }
2248  }
2249  break;
2250  case AV_PIX_FMT_RGB24:
2251  if (power8) {
2252  c->yuv2packed1 = yuv2rgb24_1_vsx;
2253  c->yuv2packed2 = yuv2rgb24_2_vsx;
2254  }
2255  break;
2256  case AV_PIX_FMT_BGR24:
2257  if (power8) {
2258  c->yuv2packed1 = yuv2bgr24_1_vsx;
2259  c->yuv2packed2 = yuv2bgr24_2_vsx;
2260  }
2261  break;
2262  }
2263  }
2264 #endif /* !HAVE_BIGENDIAN */
2265 
2266 #endif /* HAVE_VSX */
2267 }
packed YUV 4:2:2, 16bpp, Cb Y0 Cr Y1
Definition: pixfmt.h:81
#define vec_s8
Definition: util_altivec.h:35
IEEE-754 single precision Y, 32bpp, big-endian.
Definition: pixfmt.h:340
#define NULL
Definition: coverity.c:32
void(* hcScale)(struct SwsContext *c, int16_t *dst, int dstW, const uint8_t *src, const int16_t *filter, const int32_t *filterPos, int filterSize)
#define YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha)
Definition: output.c:1802
static av_always_inline int isAnyRGB(enum AVPixelFormat pix_fmt)
static int shift(int a, int b)
Definition: sonic.c:82
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
Definition: pixdesc.c:2549
#define HAVE_POWER8
Definition: config.h:52
packed RGB 8:8:8, 24bpp, RGBRGB...
Definition: pixfmt.h:68
const char * desc
Definition: nvenc.c:79
#define mul8(a, b)
#define vec_s32
Definition: util_altivec.h:39
#define YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha)
Definition: output.c:1789
static av_always_inline int is16BPS(enum AVPixelFormat pix_fmt)
void(* hyScale)(struct SwsContext *c, int16_t *dst, int dstW, const uint8_t *src, const int16_t *filter, const int32_t *filterPos, int filterSize)
Scale one horizontal line of input data using a filter over the input lines, to produce one (differen...
static atomic_int cpu_flags
Definition: cpu.c:50
void(* hyscale_fast)(struct SwsContext *c, int16_t *dst, int dstWidth, const uint8_t *src, int srcW, int xInc)
Scale one horizontal line of input data using a bilinear filter to produce one line of output data...
Macro definitions for various function/variable attributes.
static void filter(int16_t *output, ptrdiff_t out_stride, int16_t *low, ptrdiff_t low_stride, int16_t *high, ptrdiff_t high_stride, int len, int clip)
Definition: cfhd.c:196
AVComponentDescriptor comp[4]
Parameters that describe how pixels are packed.
Definition: pixdesc.h:117
uint8_t
#define av_cold
Definition: attributes.h:88
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition: avassert.h:64
8 bits with AV_PIX_FMT_RGB32 palette
Definition: pixfmt.h:77
#define AV_PIX_FMT_FLAG_FLOAT
The pixel format contains IEEE-754 floating point values.
Definition: pixdesc.h:188
#define SWS_FULL_CHR_H_INT
Definition: swscale.h:79
packed ABGR 8:8:8:8, 32bpp, ABGRABGR...
Definition: pixfmt.h:94
#define SWS_FAST_BILINEAR
Definition: swscale.h:58
external API header
enum AVPixelFormat dstFormat
Destination pixel format.
#define A(x)
Definition: vp56_arith.h:28
yuv2packedX_fn yuv2packedX
#define vec_s16
Definition: util_altivec.h:37
#define U(x)
Definition: vp56_arith.h:37
#define src
Definition: vp8dsp.c:254
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
packed BGRA 8:8:8:8, 32bpp, BGRABGRA...
Definition: pixfmt.h:95
static const uint8_t dither[8][8]
Definition: vf_fspp.c:57
yuv2packed1_fn yuv2packed1
uint8_t bits
Definition: vp3data.h:202
static const uint8_t offset[127][2]
Definition: vf_spp.c:93
static av_always_inline int isSemiPlanarYUV(enum AVPixelFormat pix_fmt)
void(* hcscale_fast)(struct SwsContext *c, int16_t *dst1, int16_t *dst2, int dstWidth, const uint8_t *src1, const uint8_t *src2, int srcW, int xInc)
packed ARGB 8:8:8:8, 32bpp, ARGBARGB...
Definition: pixfmt.h:92
int chrDstW
Width of destination chroma planes.
packed RGBA 8:8:8:8, 32bpp, RGBARGBA...
Definition: pixfmt.h:93
uint64_t flags
Combination of AV_PIX_FMT_FLAG_...
Definition: pixdesc.h:106
#define FFMIN(a, b)
Definition: common.h:96
yuv2planar1_fn yuv2plane1
int32_t
packed YUV 4:2:2, 16bpp, Y0 Cr Y1 Cb
Definition: pixfmt.h:210
packed RGB 8:8:8, 24bpp, BGRBGR...
Definition: pixfmt.h:69
#define vec_u8
Definition: util_altivec.h:34
int dstW
Width of destination luma/alpha planes.
#define MAX_FILTER_SIZE
Definition: af_dynaudnorm.c:33
static av_always_inline int isBE(enum AVPixelFormat pix_fmt)
#define src1
Definition: h264pred.c:139
#define vec_u32
Definition: util_altivec.h:38
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
Definition: pixdesc.h:81
yuv2planarX_fn yuv2planeX
packed YUV 4:2:2, 16bpp, Y0 Cb Y1 Cr
Definition: pixfmt.h:67
#define AV_CPU_FLAG_VSX
ISA 2.06.
Definition: cpu.h:61
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:93
#define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha)
Definition: output.c:1813
Contains misc utility macros and inline functions.
#define SWS_BITEXACT
Definition: swscale.h:84
#define AV_CPU_FLAG_POWER8
ISA 2.07.
Definition: cpu.h:62
yuv2NBPS(yuv2NBPS(9, yuv2NBPS(BE, yuv2NBPS(1, yuv2NBPS(10, int16_t)
Definition: output.c:369
int
yuv2packed2_fn yuv2packed2
#define LOCAL_ALIGNED(a, t, v,...)
Definition: internal.h:114
static double clip(void *opaque, double val)
Clip value val in the minval - maxval range.
Definition: vf_lut.c:162
static const int shift2[6]
Definition: dxa.c:51
static double c[64]
static const uint8_t shifts[2][12]
Definition: camellia.c:174
enum AVPixelFormat srcFormat
Source pixel format.
#define output_pixel(pos, val, bias, signedness)
Definition: output.c:888
av_cold void ff_sws_init_swscale_vsx(SwsContext *c)
Definition: swscale_vsx.c:2076
#define vec_u16
Definition: util_altivec.h:36
#define av_always_inline
Definition: attributes.h:45
int chrSrcW
Width of source chroma planes.
int depth
Number of bits in the component.
Definition: pixdesc.h:58
IEEE-754 single precision Y, 32bpp, little-endian.
Definition: pixfmt.h:341
#define YUV2PACKEDWRAPPER(name, base, ext, fmt)
Definition: output.c:710
int srcW
Width of source luma/alpha planes.
int flags
Flags passed by the user to select scaler algorithm, optimizations, subsampling, etc...
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
static double val(void *priv, double ch)
Definition: aeval.c:76
static av_always_inline int isNBPS(enum AVPixelFormat pix_fmt)
static uint8_t tmp[11]
Definition: aes_ctr.c:26