FFmpeg
Loading...
Searching...
No Matches
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 algorithm 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
93#include "config.h"
94#include "libswscale/rgb2rgb.h"
95#include "libswscale/swscale.h"
96#include "libswscale/swscale_internal.h"
97#include "libavutil/attributes.h"
98#include "libavutil/cpu.h"
99#include "libavutil/error.h"
100#include "libavutil/mem.h"
101#include "libavutil/mem_internal.h"
102#include "libavutil/pixdesc.h"
103#include "yuv2rgb_altivec.h"
104
105#if HAVE_ALTIVEC
106// util_altivec.h includes the system header altivec.h which redefines bool,
107// making it incompatible with stdbool.h. It therefore must not be included
108// before another one of our headers that may use bool.
109#include "libavutil/ppc/util_altivec.h"
110#include <assert.h>
111
112typedef struct SwsInternalAltivec {
113 SwsInternal c;
114
115 vector signed short CY;
116 vector signed short CRV;
117 vector signed short CBU;
118 vector signed short CGU;
119 vector signed short CGV;
120 vector signed short OY;
121 vector unsigned short CSHIFT;
122 vector signed short *vYCoeffsBank, *vCCoeffsBank;
123} SwsInternalAltivec;
124// Check that the sizes of the types match.
125// Note that given that SwsInternal is always allocated via av_malloc,
126// every pointer to an SwsInternal is always sufficiently aligned.
127static_assert(sizeof(SwsInternalAltivec) <= sizeof(SwsInternal) + SWSINTERNAL_ADDITIONAL_ASM_SIZE,
128 "SWSINTERNAL_ADDITIONAL_ASM_SIZE needs to be increased");
129
130static inline SwsInternalAltivec *sws_internal_altivec(SwsInternal *c)
131{
132 return (SwsInternalAltivec*)c;
133}
134
135#undef PROFILE_THE_BEAST
136#undef INC_SCALING
137
138typedef unsigned char ubyte;
139typedef signed char sbyte;
140
141/* RGB interleaver, 16 planar pels 8-bit samples per channel in
142 * homogeneous vector registers x0,x1,x2 are interleaved with the
143 * following technique:
144 *
145 * o0 = vec_mergeh(x0, x1);
146 * o1 = vec_perm(o0, x2, perm_rgb_0);
147 * o2 = vec_perm(o0, x2, perm_rgb_1);
148 * o3 = vec_mergel(x0, x1);
149 * o4 = vec_perm(o3, o2, perm_rgb_2);
150 * o5 = vec_perm(o3, o2, perm_rgb_3);
151 *
152 * perm_rgb_0: o0(RG).h v1(B) --> o1*
153 * 0 1 2 3 4
154 * rgbr|gbrg|brgb|rgbr
155 * 0010 0100 1001 0010
156 * 0102 3145 2673 894A
157 *
158 * perm_rgb_1: o0(RG).h v1(B) --> o2
159 * 0 1 2 3 4
160 * gbrg|brgb|bbbb|bbbb
161 * 0100 1001 1111 1111
162 * B5CD 6EF7 89AB CDEF
163 *
164 * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4*
165 * 0 1 2 3 4
166 * gbrg|brgb|rgbr|gbrg
167 * 1111 1111 0010 0100
168 * 89AB CDEF 0182 3945
169 *
170 * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5*
171 * 0 1 2 3 4
172 * brgb|rgbr|gbrg|brgb
173 * 1001 0010 0100 1001
174 * a67b 89cA BdCD eEFf
175 */
176static const vector unsigned char
177 perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
178 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
179 perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
180 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
181 perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
182 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
183 perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
184 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
185
186#define vec_merge3(x2, x1, x0, y0, y1, y2) \
187 do { \
188 __typeof__(x0) o0, o2, o3; \
189 o0 = vec_mergeh(x0, x1); \
190 y0 = vec_perm(o0, x2, perm_rgb_0); \
191 o2 = vec_perm(o0, x2, perm_rgb_1); \
192 o3 = vec_mergel(x0, x1); \
193 y1 = vec_perm(o3, o2, perm_rgb_2); \
194 y2 = vec_perm(o3, o2, perm_rgb_3); \
195 } while (0)
196
197#define vec_mstbgr24(x0, x1, x2, ptr) \
198 do { \
199 __typeof__(x0) _0, _1, _2; \
200 vec_merge3(x0, x1, x2, _0, _1, _2); \
201 vec_st(_0, 0, ptr++); \
202 vec_st(_1, 0, ptr++); \
203 vec_st(_2, 0, ptr++); \
204 } while (0)
205
206#define vec_mstrgb24(x0, x1, x2, ptr) \
207 do { \
208 __typeof__(x0) _0, _1, _2; \
209 vec_merge3(x2, x1, x0, _0, _1, _2); \
210 vec_st(_0, 0, ptr++); \
211 vec_st(_1, 0, ptr++); \
212 vec_st(_2, 0, ptr++); \
213 } while (0)
214
215/* pack the pixels in rgb0 format
216 * msb R
217 * lsb 0
218 */
219#define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \
220 do { \
221 T _0, _1, _2, _3; \
222 _0 = vec_mergeh(x0, x1); \
223 _1 = vec_mergeh(x2, x3); \
224 _2 = (T) vec_mergeh((vector unsigned short) _0, \
225 (vector unsigned short) _1); \
226 _3 = (T) vec_mergel((vector unsigned short) _0, \
227 (vector unsigned short) _1); \
228 vec_st(_2, 0 * 16, (T *) ptr); \
229 vec_st(_3, 1 * 16, (T *) ptr); \
230 _0 = vec_mergel(x0, x1); \
231 _1 = vec_mergel(x2, x3); \
232 _2 = (T) vec_mergeh((vector unsigned short) _0, \
233 (vector unsigned short) _1); \
234 _3 = (T) vec_mergel((vector unsigned short) _0, \
235 (vector unsigned short) _1); \
236 vec_st(_2, 2 * 16, (T *) ptr); \
237 vec_st(_3, 3 * 16, (T *) ptr); \
238 ptr += 4; \
239 } while (0)
240
241/*
242 * 1 0 1.4021 | | Y |
243 * 1 -0.3441 -0.7142 |x| Cb|
244 * 1 1.7718 0 | | Cr|
245 *
246 *
247 * Y: [-128 127]
248 * Cb/Cr : [-128 127]
249 *
250 * typical YUV conversion works on Y: 0-255 this version has been
251 * optimized for JPEG decoding.
252 */
253
254#if HAVE_BIGENDIAN
255#define vec_unh(x) \
256 (vector signed short) \
257 vec_perm(x, (__typeof__(x)) { 0 }, \
258 ((vector unsigned char) { \
259 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
260 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
261
262#define vec_unl(x) \
263 (vector signed short) \
264 vec_perm(x, (__typeof__(x)) { 0 }, \
265 ((vector unsigned char) { \
266 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
267 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
268#else
269#define vec_unh(x)(vector signed short) vec_mergeh(x,(__typeof__(x)) { 0 })
270#define vec_unl(x)(vector signed short) vec_mergel(x,(__typeof__(x)) { 0 })
271#endif
272
273#define vec_clip_s16(x) \
274 vec_max(vec_min(x, ((vector signed short) { \
275 235, 235, 235, 235, 235, 235, 235, 235 })), \
276 ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
277
278#define vec_packclp(x, y) \
279 (vector unsigned char) \
280 vec_packs((vector unsigned short) \
281 vec_max(x, ((vector signed short) { 0 })), \
282 (vector unsigned short) \
283 vec_max(y, ((vector signed short) { 0 })))
284
285static inline void cvtyuvtoRGB(SwsInternal *c, vector signed short Y,
286 vector signed short U, vector signed short V,
287 vector signed short *R, vector signed short *G,
288 vector signed short *B)
289{
290 SwsInternalAltivec *const a = sws_internal_altivec(c);
291 vector signed short vx, ux, uvx;
292
293 Y = vec_mradds(Y, a->CY, a->OY);
294 U = vec_sub(U, (vector signed short)
295 vec_splat((vector signed short) { 128 }, 0));
296 V = vec_sub(V, (vector signed short)
297 vec_splat((vector signed short) { 128 }, 0));
298
299 // ux = (CBU * (u << a->CSHIFT) + 0x4000) >> 15;
300 ux = vec_sl(U, a->CSHIFT);
301 *B = vec_mradds(ux, a->CBU, Y);
302
303 // vx = (CRV * (v << a->CSHIFT) + 0x4000) >> 15;
304 vx = vec_sl(V, a->CSHIFT);
305 *R = vec_mradds(vx, a->CRV, Y);
306
307 // uvx = ((CGU * u) + (CGV * v)) >> 15;
308 uvx = vec_mradds(U, a->CGU, Y);
309 *G = vec_mradds(V, a->CGV, uvx);
310}
311
312/*
313 * ------------------------------------------------------------------------------
314 * CS converters
315 * ------------------------------------------------------------------------------
316 */
317
318#if !HAVE_VEC_XL
319static inline vector unsigned char vec_xl(signed long long offset, const ubyte *addr)
320{
321 const vector unsigned char *v_addr = (const vector unsigned char *) (addr + offset);
322 vector unsigned char align_perm = vec_lvsl(offset, addr);
323
324 return (vector unsigned char) vec_perm(v_addr[0], v_addr[1], align_perm);
325}
326#endif /* !HAVE_VEC_XL */
327
328#define DEFCSP420_CVT(name, out_pixels) \
329static int altivec_ ## name(SwsInternal *c, const unsigned char *const *in, \
330 const int *instrides, int srcSliceY, int srcSliceH, \
331 unsigned char *const *oplanes, const int *outstrides) \
332{ \
333 SwsInternalAltivec *const a = sws_internal_altivec(c); \
334 int w = c->opts.src_w; \
335 int h = srcSliceH; \
336 int i, j; \
337 int instrides_scl[3]; \
338 vector unsigned char y0, y1; \
339 \
340 vector signed char u, v; \
341 \
342 vector signed short Y0, Y1, Y2, Y3; \
343 vector signed short U, V; \
344 vector signed short vx, ux, uvx; \
345 vector signed short vx0, ux0, uvx0; \
346 vector signed short vx1, ux1, uvx1; \
347 vector signed short R0, G0, B0; \
348 vector signed short R1, G1, B1; \
349 vector unsigned char R, G, B; \
350 \
351 vector signed short lCY = a->CY; \
352 vector signed short lOY = a->OY; \
353 vector signed short lCRV = a->CRV; \
354 vector signed short lCBU = a->CBU; \
355 vector signed short lCGU = a->CGU; \
356 vector signed short lCGV = a->CGV; \
357 vector unsigned short lCSHIFT = a->CSHIFT; \
358 \
359 const ubyte *y1i = in[0]; \
360 const ubyte *y2i = in[0] + instrides[0]; \
361 const ubyte *ui = in[1]; \
362 const ubyte *vi = in[2]; \
363 \
364 vector unsigned char *oute, *outo; \
365 \
366 /* loop moves y{1, 2}i by w */ \
367 instrides_scl[0] = instrides[0] * 2 - w; \
368 /* loop moves ui by w / 2 */ \
369 instrides_scl[1] = instrides[1] - w / 2; \
370 /* loop moves vi by w / 2 */ \
371 instrides_scl[2] = instrides[2] - w / 2; \
372 \
373 for (i = 0; i < h / 2; i++) { \
374 oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
375 (srcSliceY + i * 2)); \
376 outo = oute + (outstrides[0] >> 4); \
377 vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
378 vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
379 \
380 for (j = 0; j < w / 16; j++) { \
381 y0 = vec_xl(0, y1i); \
382 \
383 y1 = vec_xl(0, y2i); \
384 \
385 u = (vector signed char) vec_xl(0, ui); \
386 \
387 v = (vector signed char) vec_xl(0, vi); \
388 \
389 u = (vector signed char) \
390 vec_sub(u, \
391 (vector signed char) \
392 vec_splat((vector signed char) { 128 }, 0)); \
393 v = (vector signed char) \
394 vec_sub(v, \
395 (vector signed char) \
396 vec_splat((vector signed char) { 128 }, 0)); \
397 \
398 U = vec_unpackh(u); \
399 V = vec_unpackh(v); \
400 \
401 Y0 = vec_unh(y0); \
402 Y1 = vec_unl(y0); \
403 Y2 = vec_unh(y1); \
404 Y3 = vec_unl(y1); \
405 \
406 Y0 = vec_mradds(Y0, lCY, lOY); \
407 Y1 = vec_mradds(Y1, lCY, lOY); \
408 Y2 = vec_mradds(Y2, lCY, lOY); \
409 Y3 = vec_mradds(Y3, lCY, lOY); \
410 \
411 /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \
412 ux = vec_sl(U, lCSHIFT); \
413 ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
414 ux0 = vec_mergeh(ux, ux); \
415 ux1 = vec_mergel(ux, ux); \
416 \
417 /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \
418 vx = vec_sl(V, lCSHIFT); \
419 vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
420 vx0 = vec_mergeh(vx, vx); \
421 vx1 = vec_mergel(vx, vx); \
422 \
423 /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \
424 uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
425 uvx = vec_mradds(V, lCGV, uvx); \
426 uvx0 = vec_mergeh(uvx, uvx); \
427 uvx1 = vec_mergel(uvx, uvx); \
428 \
429 R0 = vec_add(Y0, vx0); \
430 G0 = vec_add(Y0, uvx0); \
431 B0 = vec_add(Y0, ux0); \
432 R1 = vec_add(Y1, vx1); \
433 G1 = vec_add(Y1, uvx1); \
434 B1 = vec_add(Y1, ux1); \
435 \
436 R = vec_packclp(R0, R1); \
437 G = vec_packclp(G0, G1); \
438 B = vec_packclp(B0, B1); \
439 \
440 out_pixels(R, G, B, oute); \
441 \
442 R0 = vec_add(Y2, vx0); \
443 G0 = vec_add(Y2, uvx0); \
444 B0 = vec_add(Y2, ux0); \
445 R1 = vec_add(Y3, vx1); \
446 G1 = vec_add(Y3, uvx1); \
447 B1 = vec_add(Y3, ux1); \
448 R = vec_packclp(R0, R1); \
449 G = vec_packclp(G0, G1); \
450 B = vec_packclp(B0, B1); \
451 \
452 \
453 out_pixels(R, G, B, outo); \
454 \
455 y1i += 16; \
456 y2i += 16; \
457 ui += 8; \
458 vi += 8; \
459 } \
460 \
461 ui += instrides_scl[1]; \
462 vi += instrides_scl[2]; \
463 y1i += instrides_scl[0]; \
464 y2i += instrides_scl[0]; \
465 } \
466 return srcSliceH; \
467}
468
469#define out_abgr(a, b, c, ptr) \
470 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), c, b, a, ptr)
471#define out_bgra(a, b, c, ptr) \
472 vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), ptr)
473#define out_rgba(a, b, c, ptr) \
474 vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), ptr)
475#define out_argb(a, b, c, ptr) \
476 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), a, b, c, ptr)
477#define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
478#define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
479
480DEFCSP420_CVT(yuv2_abgr, out_abgr)
481DEFCSP420_CVT(yuv2_bgra, out_bgra)
482DEFCSP420_CVT(yuv2_rgba, out_rgba)
483DEFCSP420_CVT(yuv2_argb, out_argb)
484DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
485DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
486
487// uyvy|uyvy|uyvy|uyvy
488// 0123 4567 89ab cdef
489static const vector unsigned char
490 demux_u = { 0x10, 0x00, 0x10, 0x00,
491 0x10, 0x04, 0x10, 0x04,
492 0x10, 0x08, 0x10, 0x08,
493 0x10, 0x0c, 0x10, 0x0c },
494 demux_v = { 0x10, 0x02, 0x10, 0x02,
495 0x10, 0x06, 0x10, 0x06,
496 0x10, 0x0A, 0x10, 0x0A,
497 0x10, 0x0E, 0x10, 0x0E },
498 demux_y = { 0x10, 0x01, 0x10, 0x03,
499 0x10, 0x05, 0x10, 0x07,
500 0x10, 0x09, 0x10, 0x0B,
501 0x10, 0x0D, 0x10, 0x0F };
502
503/*
504 * this is so I can play live CCIR raw video
505 */
506static int altivec_uyvy_rgb32(SwsInternal *c, const unsigned char *const *in,
507 const int *instrides, int srcSliceY, int srcSliceH,
508 unsigned char *const *oplanes, const int *outstrides)
509{
510 int w = c->opts.src_w;
511 int h = srcSliceH;
512 int i, j;
513 vector unsigned char uyvy;
514 vector signed short Y, U, V;
515 vector signed short R0, G0, B0, R1, G1, B1;
516 vector unsigned char R, G, B;
517 vector unsigned char *out;
518 const ubyte *img;
519
520 img = in[0];
521 out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
522
523 for (i = 0; i < h; i++)
524 for (j = 0; j < w / 16; j++) {
525 uyvy = vec_ld(0, img);
526
527 U = (vector signed short)
528 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
529 V = (vector signed short)
530 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
531 Y = (vector signed short)
532 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
533
534 cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
535
536 uyvy = vec_ld(16, img);
537
538 U = (vector signed short)
539 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
540 V = (vector signed short)
541 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
542 Y = (vector signed short)
543 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
544
545 cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
546
547 R = vec_packclp(R0, R1);
548 G = vec_packclp(G0, G1);
549 B = vec_packclp(B0, B1);
550
551 // vec_mstbgr24 (R,G,B, out);
552 out_rgba(R, G, B, out);
553
554 img += 32;
555 }
556 return srcSliceH;
557}
558
559#endif /* HAVE_ALTIVEC */
560
561/* Ok currently the acceleration routine only supports
562 * inputs of widths a multiple of 16
563 * and heights a multiple 2
564 *
565 * So we just fall back to the C codes for this.
566 */
567av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsInternal *c)
568{
569#if HAVE_ALTIVEC
570 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
571 return NULL;
572
573 /*
574 * and this seems not to matter too much I tried a bunch of
575 * videos with abnormal widths and MPlayer crashes elsewhere.
576 * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
577 * boom with X11 bad match.
578 *
579 */
580 if ((c->opts.src_w & 0xf) != 0)
581 return NULL;
582
583 switch (c->opts.src_format) {
584 case AV_PIX_FMT_YUV410P:
585 case AV_PIX_FMT_YUV420P:
586 /*case IMGFMT_CLPL: ??? */
587 case AV_PIX_FMT_GRAY8:
588 case AV_PIX_FMT_NV12:
589 case AV_PIX_FMT_NV21:
590 if ((c->opts.src_h & 0x1) != 0)
591 return NULL;
592
593/*
594 * The below accelerations for YUV2RGB are known broken.
595 * See: 'fate-checkasm-sw_yuv2rgb' with --enable-altivec
596 * They are disabled for the moment, until such time as
597 * they can be repaired.
598 */
599#if 0
600 switch (c->opts.dst_format) {
601 case AV_PIX_FMT_RGB24:
602 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
603 return altivec_yuv2_rgb24;
604 case AV_PIX_FMT_BGR24:
605 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
606 return altivec_yuv2_bgr24;
607 case AV_PIX_FMT_ARGB:
608 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
609 return altivec_yuv2_argb;
610 case AV_PIX_FMT_ABGR:
611 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
612 return altivec_yuv2_abgr;
613 case AV_PIX_FMT_RGBA:
614 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
615 return altivec_yuv2_rgba;
616 case AV_PIX_FMT_BGRA:
617 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
618 return altivec_yuv2_bgra;
619 default: return NULL;
620 }
621#endif /* disabled YUV2RGB acceleration */
622 break;
623
624 case AV_PIX_FMT_UYVY422:
625 switch (c->opts.dst_format) {
626 case AV_PIX_FMT_BGR32:
627 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
628 return altivec_uyvy_rgb32;
629 default: return NULL;
630 }
631 break;
632 }
633#endif /* HAVE_ALTIVEC */
634
635 return NULL;
636}
637
638av_cold void ff_yuv2rgb_init_tables_ppc(SwsInternal *c,
639 const int inv_table[4],
640 int brightness,
641 int contrast,
642 int saturation)
643{
644#if HAVE_ALTIVEC
645 SwsInternalAltivec *const a = sws_internal_altivec(c);
646
647 union {
648 DECLARE_ALIGNED(16, signed short, tmp)[8];
649 vector signed short vec;
650 } buf;
651
652 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
653 return;
654
655 buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy
656 buf.tmp[1] = -256 * brightness; // oy
657 buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv
658 buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu
659 buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu
660 buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv
661
662 a->CSHIFT = (vector unsigned short) vec_splat_u16(2);
663 a->CY = vec_splat((vector signed short) buf.vec, 0);
664 a->OY = vec_splat((vector signed short) buf.vec, 1);
665 a->CRV = vec_splat((vector signed short) buf.vec, 2);
666 a->CBU = vec_splat((vector signed short) buf.vec, 3);
667 a->CGU = vec_splat((vector signed short) buf.vec, 4);
668 a->CGV = vec_splat((vector signed short) buf.vec, 5);
669 return;
670#endif /* HAVE_ALTIVEC */
671}
672
673#if HAVE_ALTIVEC
674
675static av_always_inline void yuv2packedX_altivec(SwsInternal *c,
676 const int16_t *lumFilter,
677 const int16_t **lumSrc,
678 int lumFilterSize,
679 const int16_t *chrFilter,
680 const int16_t **chrUSrc,
681 const int16_t **chrVSrc,
682 int chrFilterSize,
683 const int16_t **alpSrc,
684 uint8_t *dest,
685 int dstW, int dstY,
686 enum AVPixelFormat target)
687{
688 SwsInternalAltivec *const a = sws_internal_altivec(c);
689
690 int i, j;
691 vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
692 vector signed short R0, G0, B0, R1, G1, B1;
693
694 vector unsigned char R, G, B;
695 vector unsigned char *out, *nout;
696
697 vector signed short RND = vec_splat_s16(1 << 3);
698 vector unsigned short SCL = vec_splat_u16(4);
699 DECLARE_ALIGNED(16, unsigned int, scratch)[16];
700
701 vector signed short *YCoeffs, *CCoeffs;
702
703 YCoeffs = a->vYCoeffsBank + dstY * lumFilterSize;
704 CCoeffs = a->vCCoeffsBank + dstY * chrFilterSize;
705
706 out = (vector unsigned char *) dest;
707
708 for (i = 0; i < dstW; i += 16) {
709 Y0 = RND;
710 Y1 = RND;
711 /* extract 16 coeffs from lumSrc */
712 for (j = 0; j < lumFilterSize; j++) {
713 X0 = vec_ld(0, &lumSrc[j][i]);
714 X1 = vec_ld(16, &lumSrc[j][i]);
715 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
716 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
717 }
718
719 U = RND;
720 V = RND;
721 /* extract 8 coeffs from U,V */
722 for (j = 0; j < chrFilterSize; j++) {
723 X = vec_ld(0, &chrUSrc[j][i / 2]);
724 U = vec_mradds(X, CCoeffs[j], U);
725 X = vec_ld(0, &chrVSrc[j][i / 2]);
726 V = vec_mradds(X, CCoeffs[j], V);
727 }
728
729 /* scale and clip signals */
730 Y0 = vec_sra(Y0, SCL);
731 Y1 = vec_sra(Y1, SCL);
732 U = vec_sra(U, SCL);
733 V = vec_sra(V, SCL);
734
735 Y0 = vec_clip_s16(Y0);
736 Y1 = vec_clip_s16(Y1);
737 U = vec_clip_s16(U);
738 V = vec_clip_s16(V);
739
740 /* now we have
741 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
742 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
743 *
744 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
745 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
746 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
747 */
748
749 U0 = vec_mergeh(U, U);
750 V0 = vec_mergeh(V, V);
751
752 U1 = vec_mergel(U, U);
753 V1 = vec_mergel(V, V);
754
755 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
756 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
757
758 R = vec_packclp(R0, R1);
759 G = vec_packclp(G0, G1);
760 B = vec_packclp(B0, B1);
761
762 switch (target) {
763 case AV_PIX_FMT_ABGR:
764 out_abgr(R, G, B, out);
765 break;
766 case AV_PIX_FMT_BGRA:
767 out_bgra(R, G, B, out);
768 break;
769 case AV_PIX_FMT_RGBA:
770 out_rgba(R, G, B, out);
771 break;
772 case AV_PIX_FMT_ARGB:
773 out_argb(R, G, B, out);
774 break;
775 case AV_PIX_FMT_RGB24:
776 out_rgb24(R, G, B, out);
777 break;
778 case AV_PIX_FMT_BGR24:
779 out_bgr24(R, G, B, out);
780 break;
781 default:
782 {
783 /* If this is reached, the caller should have called yuv2packedXinC
784 * instead. */
785 static int printed_error_message;
786 if (!printed_error_message) {
787 av_log(c, AV_LOG_ERROR,
788 "altivec_yuv2packedX doesn't support %s output\n",
789 av_get_pix_fmt_name(c->opts.dst_format));
790 printed_error_message = 1;
791 }
792 return;
793 }
794 }
795 }
796
797 if (i < dstW) {
798 i -= 16;
799
800 Y0 = RND;
801 Y1 = RND;
802 /* extract 16 coeffs from lumSrc */
803 for (j = 0; j < lumFilterSize; j++) {
804 X0 = vec_ld(0, &lumSrc[j][i]);
805 X1 = vec_ld(16, &lumSrc[j][i]);
806 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
807 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
808 }
809
810 U = RND;
811 V = RND;
812 /* extract 8 coeffs from U,V */
813 for (j = 0; j < chrFilterSize; j++) {
814 X = vec_ld(0, &chrUSrc[j][i / 2]);
815 U = vec_mradds(X, CCoeffs[j], U);
816 X = vec_ld(0, &chrVSrc[j][i / 2]);
817 V = vec_mradds(X, CCoeffs[j], V);
818 }
819
820 /* scale and clip signals */
821 Y0 = vec_sra(Y0, SCL);
822 Y1 = vec_sra(Y1, SCL);
823 U = vec_sra(U, SCL);
824 V = vec_sra(V, SCL);
825
826 Y0 = vec_clip_s16(Y0);
827 Y1 = vec_clip_s16(Y1);
828 U = vec_clip_s16(U);
829 V = vec_clip_s16(V);
830
831 /* now we have
832 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
833 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
834 *
835 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
836 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
837 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
838 */
839
840 U0 = vec_mergeh(U, U);
841 V0 = vec_mergeh(V, V);
842
843 U1 = vec_mergel(U, U);
844 V1 = vec_mergel(V, V);
845
846 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
847 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
848
849 R = vec_packclp(R0, R1);
850 G = vec_packclp(G0, G1);
851 B = vec_packclp(B0, B1);
852
853 nout = (vector unsigned char *) scratch;
854 switch (target) {
855 case AV_PIX_FMT_ABGR:
856 out_abgr(R, G, B, nout);
857 break;
858 case AV_PIX_FMT_BGRA:
859 out_bgra(R, G, B, nout);
860 break;
861 case AV_PIX_FMT_RGBA:
862 out_rgba(R, G, B, nout);
863 break;
864 case AV_PIX_FMT_ARGB:
865 out_argb(R, G, B, nout);
866 break;
867 case AV_PIX_FMT_RGB24:
868 out_rgb24(R, G, B, nout);
869 break;
870 case AV_PIX_FMT_BGR24:
871 out_bgr24(R, G, B, nout);
872 break;
873 default:
874 /* Unreachable, I think. */
875 av_log(c, AV_LOG_ERROR,
876 "altivec_yuv2packedX doesn't support %s output\n",
877 av_get_pix_fmt_name(c->opts.dst_format));
878 return;
879 }
880
881 memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
882 }
883}
884
885#define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
886void ff_yuv2 ## suffix ## _X_altivec(SwsInternal *c, \
887 const int16_t *lumFilter, \
888 const int16_t **lumSrc, \
889 int lumFilterSize, \
890 const int16_t *chrFilter, \
891 const int16_t **chrUSrc, \
892 const int16_t **chrVSrc, \
893 int chrFilterSize, \
894 const int16_t **alpSrc, \
895 uint8_t *dest, int dstW, int dstY) \
896{ \
897 yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
898 chrFilter, chrUSrc, chrVSrc, \
899 chrFilterSize, alpSrc, \
900 dest, dstW, dstY, pixfmt); \
901}
902
903YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR);
904YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA);
905YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB);
906YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA);
907YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
908YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);
909
910av_cold int ff_sws_init_altivec_bufs(SwsInternal *c)
911{
912 const SwsContext *const sws = &c->opts;
913 SwsInternalAltivec *const a = sws_internal_altivec(c);
914
915 a->vYCoeffsBank = av_malloc_array(sws->dst_h, c->vLumFilterSize * sizeof(*a->vYCoeffsBank));
916 a->vCCoeffsBank = av_malloc_array(c->chrDstH, c->vChrFilterSize * sizeof(*a->vCCoeffsBank));
917 if (!a->vYCoeffsBank || !a->vCCoeffsBank)
918 return AVERROR(ENOMEM);
919
920 for (int i = 0; i < c->vLumFilterSize * sws->dst_h; ++i) {
921 short *p = (short *)&a->vYCoeffsBank[i];
922 for (int j = 0; j < 8; ++j)
923 p[j] = c->vLumFilter[i];
924 }
925
926 for (int i = 0; i < c->vChrFilterSize * c->chrDstH; ++i) {
927 short *p = (short *)&a->vCCoeffsBank[i];
928 for (int j = 0; j < 8; ++j)
929 p[j] = c->vChrFilter[i];
930 }
931
932 return 0;
933}
934
935av_cold void ff_sws_free_altivec_bufs(SwsInternal *c)
936{
937 SwsInternalAltivec *const a = sws_internal_altivec(c);
938 av_freep(&a->vYCoeffsBank);
939 av_freep(&a->vCCoeffsBank);
940}
941#endif /* HAVE_ALTIVEC */
U
#define U(x)
Definition vpx_arith.h:37
CSHIFT
#define CSHIFT
Definition audiogen.c:72
V
#define V
Definition avdct.c:32
Y
#define Y
Definition boxblur.h:37
i
#define i(width, name, range_min, range_max)
Definition cbs_h264.c:63
saturation
static IPT saturation(const CmsCtx *ctx, IPT ipt)
Definition cms.c:559
NULL
#define NULL
Definition coverity.c:32
error.h
error code definitions
X
#define X
Definition f_ebur128.c:157
B1
#define B1
Definition faandct.c:42
B0
#define B0
Definition faandct.c:41
AVERROR
#define AVERROR(e)
Definition error.h:45
AV_LOG_WARNING
#define AV_LOG_WARNING
Something somehow does not look correct.
Definition log.h:216
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition log.h:210
a
int a
Definition h264pred_template.c:416
B
#define B
Definition huffyuv.h:42
R
#define R
Definition huffyuv.h:44
G
#define G
Definition huffyuv.h:43
offset
unsigned offset
Definition libaomenc.c:763
attributes.h
Macro definitions for various function/variable attributes.
av_always_inline
#define av_always_inline
Definition attributes.h:72
av_cold
#define av_cold
Definition attributes.h:117
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition cpu.c:109
AV_CPU_FLAG_ALTIVEC
#define AV_CPU_FLAG_ALTIVEC
standard
Definition cpu.h:64
w
uint8_t w
Definition llvidencdsp.c:39
mem.h
Memory handling functions.
mem_internal.h
DECLARE_ALIGNED
#define DECLARE_ALIGNED(n, t, v)
Declare a variable that is aligned in memory.
Definition mem_internal.h:104
Windows::Graphics::DirectX::Direct3D11::p
IDirect3DDxgiInterfaceAccess _COM_Outptr_ void ** p
Definition vsrc_gfxcapture_winrt.hpp:53
av_get_pix_fmt_name
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:3380
AV_PIX_FMT_BGR32
#define AV_PIX_FMT_BGR32
Definition pixfmt.h:519
AVPixelFormat
AVPixelFormat
Pixel format.
Definition pixfmt.h:71
AV_PIX_FMT_NV12
@ AV_PIX_FMT_NV12
planar YUV 4:2:0, 12bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (firs...
Definition pixfmt.h:96
AV_PIX_FMT_RGB24
@ AV_PIX_FMT_RGB24
packed RGB 8:8:8, 24bpp, RGBRGB...
Definition pixfmt.h:75
AV_PIX_FMT_YUV420P
@ AV_PIX_FMT_YUV420P
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition pixfmt.h:73
AV_PIX_FMT_NV21
@ AV_PIX_FMT_NV21
as above, but U and V bytes are swapped
Definition pixfmt.h:97
AV_PIX_FMT_ARGB
@ AV_PIX_FMT_ARGB
packed ARGB 8:8:8:8, 32bpp, ARGBARGB...
Definition pixfmt.h:99
AV_PIX_FMT_BGRA
@ AV_PIX_FMT_BGRA
packed BGRA 8:8:8:8, 32bpp, BGRABGRA...
Definition pixfmt.h:102
AV_PIX_FMT_GRAY8
@ AV_PIX_FMT_GRAY8
Y , 8bpp.
Definition pixfmt.h:81
AV_PIX_FMT_UYVY422
@ AV_PIX_FMT_UYVY422
packed YUV 4:2:2, 16bpp, Cb Y0 Cr Y1
Definition pixfmt.h:88
AV_PIX_FMT_ABGR
@ AV_PIX_FMT_ABGR
packed ABGR 8:8:8:8, 32bpp, ABGRABGR...
Definition pixfmt.h:101
AV_PIX_FMT_YUV410P
@ AV_PIX_FMT_YUV410P
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
Definition pixfmt.h:79
AV_PIX_FMT_RGBA
@ AV_PIX_FMT_RGBA
packed RGBA 8:8:8:8, 32bpp, RGBARGBA...
Definition pixfmt.h:100
AV_PIX_FMT_BGR24
@ AV_PIX_FMT_BGR24
packed RGB 8:8:8, 24bpp, BGRBGR...
Definition pixfmt.h:76
R0
#define R0(v, w, x, y, z, i)
Definition sha.c:57
R1
#define R1
Definition simple_idct.c:166
SwsContext
Main external API structure.
Definition swscale.h:227
SwsContext::dst_h
int dst_h
Width and height of the destination frame.
Definition swscale.h:273
swscale.h
external API header
swscale_internal.h
SWSINTERNAL_ADDITIONAL_ASM_SIZE
#define SWSINTERNAL_ADDITIONAL_ASM_SIZE
Definition swscale_internal.h:47
ff_sws_init_altivec_bufs
int ff_sws_init_altivec_bufs(SwsInternal *c)
SwsFunc
int(* SwsFunc)(SwsInternal *c, const uint8_t *const src[], const int srcStride[], int srcSliceY, int srcSliceH, uint8_t *const dst[], const int dstStride[])
Definition swscale_internal.h:99
ff_sws_free_altivec_bufs
void ff_sws_free_altivec_bufs(SwsInternal *c)
av_malloc_array
#define av_malloc_array(a, b)
Definition tableprint_vlc.h:32
av_freep
#define av_freep(p)
Definition tableprint_vlc.h:35
av_log
#define av_log(a,...)
Definition tableprint_vlc.h:27
tmp
static uint8_t tmp[40]
Definition aes_ctr.c:52
out
static FILE * out
Definition movenc.c:55
util_altivec.h
Contains misc utility macros and inline functions.
img
#define img
Definition vf_colormatrix.c:114
h
h
Definition vp9dsp_template.c:2070
c
static double c[64]
Definition vsrc_mptestsrc.c:89
ff_yuv2rgb_init_tables_ppc
av_cold void ff_yuv2rgb_init_tables_ppc(SwsInternal *c, const int inv_table[4], int brightness, int contrast, int saturation)
Definition yuv2rgb_altivec.c:638
ff_yuv2rgb_init_ppc
av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsInternal *c)
Definition yuv2rgb_altivec.c:567
yuv2rgb_altivec.h
Generated on Sun Aug 16 2026 20:21:04 for FFmpeg by doxygen 1.13.2