FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
yuv2rgb_altivec.c
Go to the documentation of this file.
1 /*
2  * AltiVec acceleration for colorspace conversion
3  *
4  * copyright (C) 2004 Marc Hoffman <marc.hoffman@analog.com>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 /*
24  * Convert I420 YV12 to RGB in various formats,
25  * it rejects images that are not in 420 formats,
26  * it rejects images that don't have widths of multiples of 16,
27  * it rejects images that don't have heights of multiples of 2.
28  * Reject defers to C simulation code.
29  *
30  * Lots of optimizations to be done here.
31  *
32  * 1. Need to fix saturation code. I just couldn't get it to fly with packs
33  * and adds, so we currently use max/min to clip.
34  *
35  * 2. The inefficient use of chroma loading needs a bit of brushing up.
36  *
37  * 3. Analysis of pipeline stalls needs to be done. Use shark to identify
38  * pipeline stalls.
39  *
40  *
41  * MODIFIED to calculate coeffs from currently selected color space.
42  * MODIFIED core to be a macro where you specify the output format.
43  * ADDED UYVY conversion which is never called due to some thing in swscale.
44  * CORRECTED algorithim selection to be strict on input formats.
45  * ADDED runtime detection of AltiVec.
46  *
47  * ADDED altivec_yuv2packedX vertical scl + RGB converter
48  *
49  * March 27,2004
50  * PERFORMANCE ANALYSIS
51  *
52  * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo
53  * used as test.
54  * The AltiVec version uses 10% of the processor or ~100Mips for D1 video
55  * same sequence.
56  *
57  * 720 * 480 * 30 ~10MPS
58  *
59  * so we have roughly 10 clocks per pixel. This is too high, something has
60  * to be wrong.
61  *
62  * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the
63  * need for vec_min.
64  *
65  * OPTIMIZED DST OUTPUT cache/DMA controls. We are pretty much guaranteed to
66  * have the input video frame, it was just decompressed so it probably resides
67  * in L1 caches. However, we are creating the output video stream. This needs
68  * to use the DSTST instruction to optimize for the cache. We couple this with
69  * the fact that we are not going to be visiting the input buffer again so we
70  * mark it Least Recently Used. This shaves 25% of the processor cycles off.
71  *
72  * Now memcpy is the largest mips consumer in the system, probably due
73  * to the inefficient X11 stuff.
74  *
75  * GL libraries seem to be very slow on this machine 1.33Ghz PB running
76  * Jaguar, this is not the case for my 1Ghz PB. I thought it might be
77  * a versioning issue, however I have libGL.1.2.dylib for both
78  * machines. (We need to figure this out now.)
79  *
80  * GL2 libraries work now with patch for RGB32.
81  *
82  * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor.
83  *
84  * Integrated luma prescaling adjustment for saturation/contrast/brightness
85  * adjustment.
86  */
87 
88 #include <stdio.h>
89 #include <stdlib.h>
90 #include <string.h>
91 #include <inttypes.h>
92 #include <assert.h>
93 
94 #include "config.h"
95 #include "libswscale/rgb2rgb.h"
96 #include "libswscale/swscale.h"
98 #include "libavutil/attributes.h"
99 #include "libavutil/cpu.h"
100 #include "libavutil/pixdesc.h"
101 #include "yuv2rgb_altivec.h"
102 
103 #if HAVE_ALTIVEC
104 
105 #undef PROFILE_THE_BEAST
106 #undef INC_SCALING
107 
108 typedef unsigned char ubyte;
109 typedef signed char sbyte;
110 
111 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
112  * homogeneous vector registers x0,x1,x2 are interleaved with the
113  * following technique:
114  *
115  * o0 = vec_mergeh(x0, x1);
116  * o1 = vec_perm(o0, x2, perm_rgb_0);
117  * o2 = vec_perm(o0, x2, perm_rgb_1);
118  * o3 = vec_mergel(x0, x1);
119  * o4 = vec_perm(o3, o2, perm_rgb_2);
120  * o5 = vec_perm(o3, o2, perm_rgb_3);
121  *
122  * perm_rgb_0: o0(RG).h v1(B) --> o1*
123  * 0 1 2 3 4
124  * rgbr|gbrg|brgb|rgbr
125  * 0010 0100 1001 0010
126  * 0102 3145 2673 894A
127  *
128  * perm_rgb_1: o0(RG).h v1(B) --> o2
129  * 0 1 2 3 4
130  * gbrg|brgb|bbbb|bbbb
131  * 0100 1001 1111 1111
132  * B5CD 6EF7 89AB CDEF
133  *
134  * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4*
135  * 0 1 2 3 4
136  * gbrg|brgb|rgbr|gbrg
137  * 1111 1111 0010 0100
138  * 89AB CDEF 0182 3945
139  *
140  * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5*
141  * 0 1 2 3 4
142  * brgb|rgbr|gbrg|brgb
143  * 1001 0010 0100 1001
144  * a67b 89cA BdCD eEFf
145  *
146  */
147 static const vector unsigned char
148  perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
149  0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
150  perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
151  0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
152  perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
153  0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
154  perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
155  0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
156 
157 #define vec_merge3(x2, x1, x0, y0, y1, y2) \
158  do { \
159  __typeof__(x0) o0, o2, o3; \
160  o0 = vec_mergeh(x0, x1); \
161  y0 = vec_perm(o0, x2, perm_rgb_0); \
162  o2 = vec_perm(o0, x2, perm_rgb_1); \
163  o3 = vec_mergel(x0, x1); \
164  y1 = vec_perm(o3, o2, perm_rgb_2); \
165  y2 = vec_perm(o3, o2, perm_rgb_3); \
166  } while (0)
167 
168 #define vec_mstbgr24(x0, x1, x2, ptr) \
169  do { \
170  __typeof__(x0) _0, _1, _2; \
171  vec_merge3(x0, x1, x2, _0, _1, _2); \
172  vec_st(_0, 0, ptr++); \
173  vec_st(_1, 0, ptr++); \
174  vec_st(_2, 0, ptr++); \
175  } while (0)
176 
177 #define vec_mstrgb24(x0, x1, x2, ptr) \
178  do { \
179  __typeof__(x0) _0, _1, _2; \
180  vec_merge3(x2, x1, x0, _0, _1, _2); \
181  vec_st(_0, 0, ptr++); \
182  vec_st(_1, 0, ptr++); \
183  vec_st(_2, 0, ptr++); \
184  } while (0)
185 
186 /* pack the pixels in rgb0 format
187  * msb R
188  * lsb 0
189  */
190 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \
191  do { \
192  T _0, _1, _2, _3; \
193  _0 = vec_mergeh(x0, x1); \
194  _1 = vec_mergeh(x2, x3); \
195  _2 = (T) vec_mergeh((vector unsigned short) _0, \
196  (vector unsigned short) _1); \
197  _3 = (T) vec_mergel((vector unsigned short) _0, \
198  (vector unsigned short) _1); \
199  vec_st(_2, 0 * 16, (T *) ptr); \
200  vec_st(_3, 1 * 16, (T *) ptr); \
201  _0 = vec_mergel(x0, x1); \
202  _1 = vec_mergel(x2, x3); \
203  _2 = (T) vec_mergeh((vector unsigned short) _0, \
204  (vector unsigned short) _1); \
205  _3 = (T) vec_mergel((vector unsigned short) _0, \
206  (vector unsigned short) _1); \
207  vec_st(_2, 2 * 16, (T *) ptr); \
208  vec_st(_3, 3 * 16, (T *) ptr); \
209  ptr += 4; \
210  } while (0)
211 
212 /*
213  * 1 0 1.4021 | | Y |
214  * 1 -0.3441 -0.7142 |x| Cb|
215  * 1 1.7718 0 | | Cr|
216  *
217  *
218  * Y: [-128 127]
219  * Cb/Cr : [-128 127]
220  *
221  * typical YUV conversion works on Y: 0-255 this version has been
222  * optimized for JPEG decoding.
223  */
224 
225 #if HAVE_BIGENDIAN
226 #define vec_unh(x) \
227  (vector signed short) \
228  vec_perm(x, (__typeof__(x)) { 0 }, \
229  ((vector unsigned char) { \
230  0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
231  0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
232 
233 #define vec_unl(x) \
234  (vector signed short) \
235  vec_perm(x, (__typeof__(x)) { 0 }, \
236  ((vector unsigned char) { \
237  0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
238  0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
239 #else
240 #define vec_unh(x)(vector signed short) vec_mergeh(x,(__typeof__(x)) { 0 })
241 #define vec_unl(x)(vector signed short) vec_mergel(x,(__typeof__(x)) { 0 })
242 #endif
243 
244 #define vec_clip_s16(x) \
245  vec_max(vec_min(x, ((vector signed short) { \
246  235, 235, 235, 235, 235, 235, 235, 235 })), \
247  ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
248 
249 #define vec_packclp(x, y) \
250  (vector unsigned char) \
251  vec_packs((vector unsigned short) \
252  vec_max(x, ((vector signed short) { 0 })), \
253  (vector unsigned short) \
254  vec_max(y, ((vector signed short) { 0 })))
255 
256 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
257  vector signed short U, vector signed short V,
258  vector signed short *R, vector signed short *G,
259  vector signed short *B)
260 {
261  vector signed short vx, ux, uvx;
262 
263  Y = vec_mradds(Y, c->CY, c->OY);
264  U = vec_sub(U, (vector signed short)
265  vec_splat((vector signed short) { 128 }, 0));
266  V = vec_sub(V, (vector signed short)
267  vec_splat((vector signed short) { 128 }, 0));
268 
269  // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
270  ux = vec_sl(U, c->CSHIFT);
271  *B = vec_mradds(ux, c->CBU, Y);
272 
273  // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
274  vx = vec_sl(V, c->CSHIFT);
275  *R = vec_mradds(vx, c->CRV, Y);
276 
277  // uvx = ((CGU * u) + (CGV * v)) >> 15;
278  uvx = vec_mradds(U, c->CGU, Y);
279  *G = vec_mradds(V, c->CGV, uvx);
280 }
281 
282 /*
283  * ------------------------------------------------------------------------------
284  * CS converters
285  * ------------------------------------------------------------------------------
286  */
287 
288 #define DEFCSP420_CVT(name, out_pixels) \
289 static int altivec_ ## name(SwsContext *c, const unsigned char **in, \
290  int *instrides, int srcSliceY, int srcSliceH, \
291  unsigned char **oplanes, int *outstrides) \
292 { \
293  int w = c->srcW; \
294  int h = srcSliceH; \
295  int i, j; \
296  int instrides_scl[3]; \
297  vector unsigned char y0, y1; \
298  \
299  vector signed char u, v; \
300  \
301  vector signed short Y0, Y1, Y2, Y3; \
302  vector signed short U, V; \
303  vector signed short vx, ux, uvx; \
304  vector signed short vx0, ux0, uvx0; \
305  vector signed short vx1, ux1, uvx1; \
306  vector signed short R0, G0, B0; \
307  vector signed short R1, G1, B1; \
308  vector unsigned char R, G, B; \
309  \
310  const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \
311  vector unsigned char align_perm; \
312  \
313  vector signed short lCY = c->CY; \
314  vector signed short lOY = c->OY; \
315  vector signed short lCRV = c->CRV; \
316  vector signed short lCBU = c->CBU; \
317  vector signed short lCGU = c->CGU; \
318  vector signed short lCGV = c->CGV; \
319  vector unsigned short lCSHIFT = c->CSHIFT; \
320  \
321  const ubyte *y1i = in[0]; \
322  const ubyte *y2i = in[0] + instrides[0]; \
323  const ubyte *ui = in[1]; \
324  const ubyte *vi = in[2]; \
325  \
326  vector unsigned char *oute, *outo; \
327  \
328  /* loop moves y{1, 2}i by w */ \
329  instrides_scl[0] = instrides[0] * 2 - w; \
330  /* loop moves ui by w / 2 */ \
331  instrides_scl[1] = instrides[1] - w / 2; \
332  /* loop moves vi by w / 2 */ \
333  instrides_scl[2] = instrides[2] - w / 2; \
334  \
335  for (i = 0; i < h / 2; i++) { \
336  oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
337  (srcSliceY + i * 2)); \
338  outo = oute + (outstrides[0] >> 4); \
339  vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
340  vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
341  \
342  for (j = 0; j < w / 16; j++) { \
343  y1ivP = (const vector unsigned char *) y1i; \
344  y2ivP = (const vector unsigned char *) y2i; \
345  uivP = (const vector unsigned char *) ui; \
346  vivP = (const vector unsigned char *) vi; \
347  \
348  align_perm = vec_lvsl(0, y1i); \
349  y0 = (vector unsigned char) \
350  vec_perm(y1ivP[0], y1ivP[1], align_perm); \
351  \
352  align_perm = vec_lvsl(0, y2i); \
353  y1 = (vector unsigned char) \
354  vec_perm(y2ivP[0], y2ivP[1], align_perm); \
355  \
356  align_perm = vec_lvsl(0, ui); \
357  u = (vector signed char) \
358  vec_perm(uivP[0], uivP[1], align_perm); \
359  \
360  align_perm = vec_lvsl(0, vi); \
361  v = (vector signed char) \
362  vec_perm(vivP[0], vivP[1], align_perm); \
363  \
364  u = (vector signed char) \
365  vec_sub(u, \
366  (vector signed char) \
367  vec_splat((vector signed char) { 128 }, 0)); \
368  v = (vector signed char) \
369  vec_sub(v, \
370  (vector signed char) \
371  vec_splat((vector signed char) { 128 }, 0)); \
372  \
373  U = vec_unpackh(u); \
374  V = vec_unpackh(v); \
375  \
376  Y0 = vec_unh(y0); \
377  Y1 = vec_unl(y0); \
378  Y2 = vec_unh(y1); \
379  Y3 = vec_unl(y1); \
380  \
381  Y0 = vec_mradds(Y0, lCY, lOY); \
382  Y1 = vec_mradds(Y1, lCY, lOY); \
383  Y2 = vec_mradds(Y2, lCY, lOY); \
384  Y3 = vec_mradds(Y3, lCY, lOY); \
385  \
386  /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \
387  ux = vec_sl(U, lCSHIFT); \
388  ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
389  ux0 = vec_mergeh(ux, ux); \
390  ux1 = vec_mergel(ux, ux); \
391  \
392  /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \
393  vx = vec_sl(V, lCSHIFT); \
394  vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
395  vx0 = vec_mergeh(vx, vx); \
396  vx1 = vec_mergel(vx, vx); \
397  \
398  /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \
399  uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
400  uvx = vec_mradds(V, lCGV, uvx); \
401  uvx0 = vec_mergeh(uvx, uvx); \
402  uvx1 = vec_mergel(uvx, uvx); \
403  \
404  R0 = vec_add(Y0, vx0); \
405  G0 = vec_add(Y0, uvx0); \
406  B0 = vec_add(Y0, ux0); \
407  R1 = vec_add(Y1, vx1); \
408  G1 = vec_add(Y1, uvx1); \
409  B1 = vec_add(Y1, ux1); \
410  \
411  R = vec_packclp(R0, R1); \
412  G = vec_packclp(G0, G1); \
413  B = vec_packclp(B0, B1); \
414  \
415  out_pixels(R, G, B, oute); \
416  \
417  R0 = vec_add(Y2, vx0); \
418  G0 = vec_add(Y2, uvx0); \
419  B0 = vec_add(Y2, ux0); \
420  R1 = vec_add(Y3, vx1); \
421  G1 = vec_add(Y3, uvx1); \
422  B1 = vec_add(Y3, ux1); \
423  R = vec_packclp(R0, R1); \
424  G = vec_packclp(G0, G1); \
425  B = vec_packclp(B0, B1); \
426  \
427  \
428  out_pixels(R, G, B, outo); \
429  \
430  y1i += 16; \
431  y2i += 16; \
432  ui += 8; \
433  vi += 8; \
434  } \
435  \
436  ui += instrides_scl[1]; \
437  vi += instrides_scl[2]; \
438  y1i += instrides_scl[0]; \
439  y2i += instrides_scl[0]; \
440  } \
441  return srcSliceH; \
442 }
443 
444 #define out_abgr(a, b, c, ptr) \
445  vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
446 #define out_bgra(a, b, c, ptr) \
447  vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
448 #define out_rgba(a, b, c, ptr) \
449  vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
450 #define out_argb(a, b, c, ptr) \
451  vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
452 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
453 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
454 
455 DEFCSP420_CVT(yuv2_abgr, out_abgr)
456 DEFCSP420_CVT(yuv2_bgra, out_bgra)
457 DEFCSP420_CVT(yuv2_rgba, out_rgba)
458 DEFCSP420_CVT(yuv2_argb, out_argb)
459 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
460 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
461 
462 // uyvy|uyvy|uyvy|uyvy
463 // 0123 4567 89ab cdef
464 static const vector unsigned char
465  demux_u = { 0x10, 0x00, 0x10, 0x00,
466  0x10, 0x04, 0x10, 0x04,
467  0x10, 0x08, 0x10, 0x08,
468  0x10, 0x0c, 0x10, 0x0c },
469  demux_v = { 0x10, 0x02, 0x10, 0x02,
470  0x10, 0x06, 0x10, 0x06,
471  0x10, 0x0A, 0x10, 0x0A,
472  0x10, 0x0E, 0x10, 0x0E },
473  demux_y = { 0x10, 0x01, 0x10, 0x03,
474  0x10, 0x05, 0x10, 0x07,
475  0x10, 0x09, 0x10, 0x0B,
476  0x10, 0x0D, 0x10, 0x0F };
477 
478 /*
479  * this is so I can play live CCIR raw video
480  */
481 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
482  int *instrides, int srcSliceY, int srcSliceH,
483  unsigned char **oplanes, int *outstrides)
484 {
485  int w = c->srcW;
486  int h = srcSliceH;
487  int i, j;
488  vector unsigned char uyvy;
489  vector signed short Y, U, V;
490  vector signed short R0, G0, B0, R1, G1, B1;
491  vector unsigned char R, G, B;
492  vector unsigned char *out;
493  const ubyte *img;
494 
495  img = in[0];
496  out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
497 
498  for (i = 0; i < h; i++)
499  for (j = 0; j < w / 16; j++) {
500  uyvy = vec_ld(0, img);
501 
502  U = (vector signed short)
503  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
504  V = (vector signed short)
505  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
506  Y = (vector signed short)
507  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
508 
509  cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
510 
511  uyvy = vec_ld(16, img);
512 
513  U = (vector signed short)
514  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
515  V = (vector signed short)
516  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
517  Y = (vector signed short)
518  vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
519 
520  cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
521 
522  R = vec_packclp(R0, R1);
523  G = vec_packclp(G0, G1);
524  B = vec_packclp(B0, B1);
525 
526  // vec_mstbgr24 (R,G,B, out);
527  out_rgba(R, G, B, out);
528 
529  img += 32;
530  }
531  return srcSliceH;
532 }
533 
534 #endif /* HAVE_ALTIVEC */
535 
536 /* Ok currently the acceleration routine only supports
537  * inputs of widths a multiple of 16
538  * and heights a multiple 2
539  *
540  * So we just fall back to the C codes for this.
541  */
543 {
544 #if HAVE_ALTIVEC
546  return NULL;
547 
548  /*
549  * and this seems not to matter too much I tried a bunch of
550  * videos with abnormal widths and MPlayer crashes elsewhere.
551  * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
552  * boom with X11 bad match.
553  *
554  */
555  if ((c->srcW & 0xf) != 0)
556  return NULL;
557 
558  switch (c->srcFormat) {
559  case AV_PIX_FMT_YUV410P:
560  case AV_PIX_FMT_YUV420P:
561  /*case IMGFMT_CLPL: ??? */
562  case AV_PIX_FMT_GRAY8:
563  case AV_PIX_FMT_NV12:
564  case AV_PIX_FMT_NV21:
565  if ((c->srcH & 0x1) != 0)
566  return NULL;
567 
568  switch (c->dstFormat) {
569  case AV_PIX_FMT_RGB24:
570  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
571  return altivec_yuv2_rgb24;
572  case AV_PIX_FMT_BGR24:
573  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
574  return altivec_yuv2_bgr24;
575  case AV_PIX_FMT_ARGB:
576  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
577  return altivec_yuv2_argb;
578  case AV_PIX_FMT_ABGR:
579  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
580  return altivec_yuv2_abgr;
581  case AV_PIX_FMT_RGBA:
582  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
583  return altivec_yuv2_rgba;
584  case AV_PIX_FMT_BGRA:
585  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
586  return altivec_yuv2_bgra;
587  default: return NULL;
588  }
589  break;
590 
591  case AV_PIX_FMT_UYVY422:
592  switch (c->dstFormat) {
593  case AV_PIX_FMT_BGR32:
594  av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
595  return altivec_uyvy_rgb32;
596  default: return NULL;
597  }
598  break;
599  }
600 #endif /* HAVE_ALTIVEC */
601 
602  return NULL;
603 }
604 
606  const int inv_table[4],
607  int brightness,
608  int contrast,
609  int saturation)
610 {
611 #if HAVE_ALTIVEC
612  union {
613  DECLARE_ALIGNED(16, signed short, tmp)[8];
614  vector signed short vec;
615  } buf;
616 
618  return;
619 
620  buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy
621  buf.tmp[1] = -256 * brightness; // oy
622  buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv
623  buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu
624  buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu
625  buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv
626 
627  c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
628  c->CY = vec_splat((vector signed short) buf.vec, 0);
629  c->OY = vec_splat((vector signed short) buf.vec, 1);
630  c->CRV = vec_splat((vector signed short) buf.vec, 2);
631  c->CBU = vec_splat((vector signed short) buf.vec, 3);
632  c->CGU = vec_splat((vector signed short) buf.vec, 4);
633  c->CGV = vec_splat((vector signed short) buf.vec, 5);
634  return;
635 #endif /* HAVE_ALTIVEC */
636 }
637 
638 #if HAVE_ALTIVEC
639 
640 static av_always_inline void yuv2packedX_altivec(SwsContext *c,
641  const int16_t *lumFilter,
642  const int16_t **lumSrc,
643  int lumFilterSize,
644  const int16_t *chrFilter,
645  const int16_t **chrUSrc,
646  const int16_t **chrVSrc,
647  int chrFilterSize,
648  const int16_t **alpSrc,
649  uint8_t *dest,
650  int dstW, int dstY,
651  enum AVPixelFormat target)
652 {
653  int i, j;
654  vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
655  vector signed short R0, G0, B0, R1, G1, B1;
656 
657  vector unsigned char R, G, B;
658  vector unsigned char *out, *nout;
659 
660  vector signed short RND = vec_splat_s16(1 << 3);
661  vector unsigned short SCL = vec_splat_u16(4);
662  DECLARE_ALIGNED(16, unsigned int, scratch)[16];
663 
664  vector signed short *YCoeffs, *CCoeffs;
665 
666  YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
667  CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
668 
669  out = (vector unsigned char *) dest;
670 
671  for (i = 0; i < dstW; i += 16) {
672  Y0 = RND;
673  Y1 = RND;
674  /* extract 16 coeffs from lumSrc */
675  for (j = 0; j < lumFilterSize; j++) {
676  X0 = vec_ld(0, &lumSrc[j][i]);
677  X1 = vec_ld(16, &lumSrc[j][i]);
678  Y0 = vec_mradds(X0, YCoeffs[j], Y0);
679  Y1 = vec_mradds(X1, YCoeffs[j], Y1);
680  }
681 
682  U = RND;
683  V = RND;
684  /* extract 8 coeffs from U,V */
685  for (j = 0; j < chrFilterSize; j++) {
686  X = vec_ld(0, &chrUSrc[j][i / 2]);
687  U = vec_mradds(X, CCoeffs[j], U);
688  X = vec_ld(0, &chrVSrc[j][i / 2]);
689  V = vec_mradds(X, CCoeffs[j], V);
690  }
691 
692  /* scale and clip signals */
693  Y0 = vec_sra(Y0, SCL);
694  Y1 = vec_sra(Y1, SCL);
695  U = vec_sra(U, SCL);
696  V = vec_sra(V, SCL);
697 
698  Y0 = vec_clip_s16(Y0);
699  Y1 = vec_clip_s16(Y1);
700  U = vec_clip_s16(U);
701  V = vec_clip_s16(V);
702 
703  /* now we have
704  * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
705  * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
706  *
707  * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
708  * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
709  * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
710  */
711 
712  U0 = vec_mergeh(U, U);
713  V0 = vec_mergeh(V, V);
714 
715  U1 = vec_mergel(U, U);
716  V1 = vec_mergel(V, V);
717 
718  cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
719  cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
720 
721  R = vec_packclp(R0, R1);
722  G = vec_packclp(G0, G1);
723  B = vec_packclp(B0, B1);
724 
725  switch (target) {
726  case AV_PIX_FMT_ABGR:
727  out_abgr(R, G, B, out);
728  break;
729  case AV_PIX_FMT_BGRA:
730  out_bgra(R, G, B, out);
731  break;
732  case AV_PIX_FMT_RGBA:
733  out_rgba(R, G, B, out);
734  break;
735  case AV_PIX_FMT_ARGB:
736  out_argb(R, G, B, out);
737  break;
738  case AV_PIX_FMT_RGB24:
739  out_rgb24(R, G, B, out);
740  break;
741  case AV_PIX_FMT_BGR24:
742  out_bgr24(R, G, B, out);
743  break;
744  default:
745  {
746  /* If this is reached, the caller should have called yuv2packedXinC
747  * instead. */
748  static int printed_error_message;
749  if (!printed_error_message) {
750  av_log(c, AV_LOG_ERROR,
751  "altivec_yuv2packedX doesn't support %s output\n",
753  printed_error_message = 1;
754  }
755  return;
756  }
757  }
758  }
759 
760  if (i < dstW) {
761  i -= 16;
762 
763  Y0 = RND;
764  Y1 = RND;
765  /* extract 16 coeffs from lumSrc */
766  for (j = 0; j < lumFilterSize; j++) {
767  X0 = vec_ld(0, &lumSrc[j][i]);
768  X1 = vec_ld(16, &lumSrc[j][i]);
769  Y0 = vec_mradds(X0, YCoeffs[j], Y0);
770  Y1 = vec_mradds(X1, YCoeffs[j], Y1);
771  }
772 
773  U = RND;
774  V = RND;
775  /* extract 8 coeffs from U,V */
776  for (j = 0; j < chrFilterSize; j++) {
777  X = vec_ld(0, &chrUSrc[j][i / 2]);
778  U = vec_mradds(X, CCoeffs[j], U);
779  X = vec_ld(0, &chrVSrc[j][i / 2]);
780  V = vec_mradds(X, CCoeffs[j], V);
781  }
782 
783  /* scale and clip signals */
784  Y0 = vec_sra(Y0, SCL);
785  Y1 = vec_sra(Y1, SCL);
786  U = vec_sra(U, SCL);
787  V = vec_sra(V, SCL);
788 
789  Y0 = vec_clip_s16(Y0);
790  Y1 = vec_clip_s16(Y1);
791  U = vec_clip_s16(U);
792  V = vec_clip_s16(V);
793 
794  /* now we have
795  * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
796  * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
797  *
798  * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
799  * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
800  * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
801  */
802 
803  U0 = vec_mergeh(U, U);
804  V0 = vec_mergeh(V, V);
805 
806  U1 = vec_mergel(U, U);
807  V1 = vec_mergel(V, V);
808 
809  cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
810  cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
811 
812  R = vec_packclp(R0, R1);
813  G = vec_packclp(G0, G1);
814  B = vec_packclp(B0, B1);
815 
816  nout = (vector unsigned char *) scratch;
817  switch (target) {
818  case AV_PIX_FMT_ABGR:
819  out_abgr(R, G, B, nout);
820  break;
821  case AV_PIX_FMT_BGRA:
822  out_bgra(R, G, B, nout);
823  break;
824  case AV_PIX_FMT_RGBA:
825  out_rgba(R, G, B, nout);
826  break;
827  case AV_PIX_FMT_ARGB:
828  out_argb(R, G, B, nout);
829  break;
830  case AV_PIX_FMT_RGB24:
831  out_rgb24(R, G, B, nout);
832  break;
833  case AV_PIX_FMT_BGR24:
834  out_bgr24(R, G, B, nout);
835  break;
836  default:
837  /* Unreachable, I think. */
838  av_log(c, AV_LOG_ERROR,
839  "altivec_yuv2packedX doesn't support %s output\n",
841  return;
842  }
843 
844  memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
845  }
846 }
847 
848 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
849 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \
850  const int16_t *lumFilter, \
851  const int16_t **lumSrc, \
852  int lumFilterSize, \
853  const int16_t *chrFilter, \
854  const int16_t **chrUSrc, \
855  const int16_t **chrVSrc, \
856  int chrFilterSize, \
857  const int16_t **alpSrc, \
858  uint8_t *dest, int dstW, int dstY) \
859 { \
860  yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
861  chrFilter, chrUSrc, chrVSrc, \
862  chrFilterSize, alpSrc, \
863  dest, dstW, dstY, pixfmt); \
864 }
865 
866 YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR);
867 YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA);
868 YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB);
869 YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA);
870 YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
871 YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);
872 
873 #endif /* HAVE_ALTIVEC */
packed YUV 4:2:2, 16bpp, Cb Y0 Cr Y1
Definition: pixfmt.h:83
#define NULL
Definition: coverity.c:32
#define AV_CPU_FLAG_ALTIVEC
standard
Definition: cpu.h:56
av_cold void ff_yuv2rgb_init_tables_ppc(SwsContext *c, const int inv_table[4], int brightness, int contrast, int saturation)
#define AV_LOG_WARNING
Something somehow does not look correct.
Definition: log.h:182
packed RGB 8:8:8, 24bpp, RGBRGB...
Definition: pixfmt.h:65
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:53
Macro definitions for various function/variable attributes.
int srcH
Height of source luma/alpha planes.
#define B1
Definition: faandct.c:41
#define img
Definition: vf_geq.c:46
uint8_t
#define av_cold
Definition: attributes.h:82
#define Y
Definition: vf_boxblur.c:76
packed ABGR 8:8:8:8, 32bpp, ABGRABGR...
Definition: pixfmt.h:96
external API header
enum AVPixelFormat dstFormat
Destination pixel format.
#define av_log(a,...)
#define U(x)
Definition: vp56_arith.h:37
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:176
#define R0(v, w, x, y, z, i)
Definition: sha.c:56
packed BGRA 8:8:8:8, 32bpp, BGRABGRA...
Definition: pixfmt.h:97
planar YUV 4:2:0, 12bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (firs...
Definition: pixfmt.h:91
packed ARGB 8:8:8:8, 32bpp, ARGBARGB...
Definition: pixfmt.h:94
packed RGBA 8:8:8:8, 32bpp, RGBARGBA...
Definition: pixfmt.h:95
as above, but U and V bytes are swapped
Definition: pixfmt.h:92
packed RGB 8:8:8, 24bpp, BGRBGR...
Definition: pixfmt.h:66
FILE * out
Definition: movenc-test.c:54
Definition: vf_geq.c:46
#define AV_PIX_FMT_BGR32
Definition: pixfmt.h:309
#define R1
Definition: simple_idct.c:163
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(constint16_t *) pi >>8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(constint32_t *) pi >>24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(constfloat *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(constfloat *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(constfloat *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(constdouble *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(constdouble *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(constdouble *) pi *(1U<< 31))))#defineSET_CONV_FUNC_GROUP(ofmt, ifmt) staticvoidset_generic_function(AudioConvert *ac){}voidff_audio_convert_free(AudioConvert **ac){if(!*ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);}AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enumAVSampleFormatout_fmt, enumAVSampleFormatin_fmt, intchannels, intsample_rate, intapply_map){AudioConvert *ac;intin_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) returnNULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method!=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt)>2){ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc){av_free(ac);returnNULL;}returnac;}in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar){ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar?ac->channels:1;}elseif(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;elseac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);returnac;}intff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in){intuse_generic=1;intlen=in->nb_samples;intp;if(ac->dc){av_log(ac->avr, AV_LOG_TRACE,"%dsamples-audio_convert:%sto%s(dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));returnff_convert_dither(ac-> in
void * buf
Definition: avisynth_c.h:553
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
Definition: pixfmt.h:69
int(* SwsFunc)(struct SwsContext *context, const uint8_t *src[], int srcStride[], int srcSliceY, int srcSliceH, uint8_t *dst[], int dstStride[])
Definition: vf_geq.c:46
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:76
av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c)
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:63
Y , 8bpp.
Definition: pixfmt.h:71
static double c[64]
enum AVPixelFormat srcFormat
Source pixel format.
#define av_always_inline
Definition: attributes.h:39
const char * av_get_pix_fmt_name(enum AVPixelFormat pix_fmt)
Return the short name for a pixel format, NULL in case pix_fmt is unknown.
Definition: pixdesc.c:2078
int srcW
Width of source luma/alpha planes.
AVPixelFormat
Pixel format.
Definition: pixfmt.h:61
#define B0
Definition: faandct.c:40
#define V
Definition: avdct.c:30