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
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/mem_internal.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 static const vector unsigned char
147 perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
148 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
149 perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
150 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
151 perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
152 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
153 perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
154 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
155
156 #define vec_merge3(x2, x1, x0, y0, y1, y2) \
157 do { \
158 __typeof__(x0) o0, o2, o3; \
159 o0 = vec_mergeh(x0, x1); \
160 y0 = vec_perm(o0, x2, perm_rgb_0); \
161 o2 = vec_perm(o0, x2, perm_rgb_1); \
162 o3 = vec_mergel(x0, x1); \
163 y1 = vec_perm(o3, o2, perm_rgb_2); \
164 y2 = vec_perm(o3, o2, perm_rgb_3); \
165 } while (0)
166
167 #define vec_mstbgr24(x0, x1, x2, ptr) \
168 do { \
169 __typeof__(x0) _0, _1, _2; \
170 vec_merge3(x0, x1, x2, _0, _1, _2); \
171 vec_st(_0, 0, ptr++); \
172 vec_st(_1, 0, ptr++); \
173 vec_st(_2, 0, ptr++); \
174 } while (0)
175
176 #define vec_mstrgb24(x0, x1, x2, ptr) \
177 do { \
178 __typeof__(x0) _0, _1, _2; \
179 vec_merge3(x2, x1, x0, _0, _1, _2); \
180 vec_st(_0, 0, ptr++); \
181 vec_st(_1, 0, ptr++); \
182 vec_st(_2, 0, ptr++); \
183 } while (0)
184
185 /* pack the pixels in rgb0 format
186 * msb R
187 * lsb 0
188 */
189 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \
190 do { \
191 T _0, _1, _2, _3; \
192 _0 = vec_mergeh(x0, x1); \
193 _1 = vec_mergeh(x2, x3); \
194 _2 = (T) vec_mergeh((vector unsigned short) _0, \
195 (vector unsigned short) _1); \
196 _3 = (T) vec_mergel((vector unsigned short) _0, \
197 (vector unsigned short) _1); \
198 vec_st(_2, 0 * 16, (T *) ptr); \
199 vec_st(_3, 1 * 16, (T *) ptr); \
200 _0 = vec_mergel(x0, x1); \
201 _1 = vec_mergel(x2, x3); \
202 _2 = (T) vec_mergeh((vector unsigned short) _0, \
203 (vector unsigned short) _1); \
204 _3 = (T) vec_mergel((vector unsigned short) _0, \
205 (vector unsigned short) _1); \
206 vec_st(_2, 2 * 16, (T *) ptr); \
207 vec_st(_3, 3 * 16, (T *) ptr); \
208 ptr += 4; \
209 } while (0)
210
211 /*
212 * 1 0 1.4021 | | Y |
213 * 1 -0.3441 -0.7142 |x| Cb|
214 * 1 1.7718 0 | | Cr|
215 *
216 *
217 * Y: [-128 127]
218 * Cb/Cr : [-128 127]
219 *
220 * typical YUV conversion works on Y: 0-255 this version has been
221 * optimized for JPEG decoding.
222 */
223
224 #if HAVE_BIGENDIAN
225 #define vec_unh(x) \
226 (vector signed short) \
227 vec_perm(x, (__typeof__(x)) { 0 }, \
228 ((vector unsigned char) { \
229 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
230 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
231
232 #define vec_unl(x) \
233 (vector signed short) \
234 vec_perm(x, (__typeof__(x)) { 0 }, \
235 ((vector unsigned char) { \
236 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
237 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
238 #else
239 #define vec_unh(x)(vector signed short) vec_mergeh(x,(__typeof__(x)) { 0 })
240 #define vec_unl(x)(vector signed short) vec_mergel(x,(__typeof__(x)) { 0 })
241 #endif
242
243 #define vec_clip_s16(x) \
244 vec_max(vec_min(x, ((vector signed short) { \
245 235, 235, 235, 235, 235, 235, 235, 235 })), \
246 ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
247
248 #define vec_packclp(x, y) \
249 (vector unsigned char) \
250 vec_packs((vector unsigned short) \
251 vec_max(x, ((vector signed short) { 0 })), \
252 (vector unsigned short) \
253 vec_max(y, ((vector signed short) { 0 })))
254
cvtyuvtoRGB(SwsContext * c,vector signed short Y,vector signed short U,vector signed short V,vector signed short * R,vector signed short * G,vector signed short * B)255 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
256 vector signed short U, vector signed short V,
257 vector signed short *R, vector signed short *G,
258 vector signed short *B)
259 {
260 vector signed short vx, ux, uvx;
261
262 Y = vec_mradds(Y, c->CY, c->OY);
263 U = vec_sub(U, (vector signed short)
264 vec_splat((vector signed short) { 128 }, 0));
265 V = vec_sub(V, (vector signed short)
266 vec_splat((vector signed short) { 128 }, 0));
267
268 // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
269 ux = vec_sl(U, c->CSHIFT);
270 *B = vec_mradds(ux, c->CBU, Y);
271
272 // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
273 vx = vec_sl(V, c->CSHIFT);
274 *R = vec_mradds(vx, c->CRV, Y);
275
276 // uvx = ((CGU * u) + (CGV * v)) >> 15;
277 uvx = vec_mradds(U, c->CGU, Y);
278 *G = vec_mradds(V, c->CGV, uvx);
279 }
280
281 /*
282 * ------------------------------------------------------------------------------
283 * CS converters
284 * ------------------------------------------------------------------------------
285 */
286
287 #if !HAVE_VSX
vec_xl(signed long long offset,const ubyte * addr)288 static inline vector unsigned char vec_xl(signed long long offset, const ubyte *addr)
289 {
290 const vector unsigned char *v_addr = (const vector unsigned char *) (addr + offset);
291 vector unsigned char align_perm = vec_lvsl(offset, addr);
292
293 return (vector unsigned char) vec_perm(v_addr[0], v_addr[1], align_perm);
294 }
295 #endif /* !HAVE_VSX */
296
297 #define DEFCSP420_CVT(name, out_pixels) \
298 static int altivec_ ## name(SwsContext *c, const unsigned char **in, \
299 int *instrides, int srcSliceY, int srcSliceH, \
300 unsigned char **oplanes, int *outstrides) \
301 { \
302 int w = c->srcW; \
303 int h = srcSliceH; \
304 int i, j; \
305 int instrides_scl[3]; \
306 vector unsigned char y0, y1; \
307 \
308 vector signed char u, v; \
309 \
310 vector signed short Y0, Y1, Y2, Y3; \
311 vector signed short U, V; \
312 vector signed short vx, ux, uvx; \
313 vector signed short vx0, ux0, uvx0; \
314 vector signed short vx1, ux1, uvx1; \
315 vector signed short R0, G0, B0; \
316 vector signed short R1, G1, B1; \
317 vector unsigned char R, G, B; \
318 \
319 vector signed short lCY = c->CY; \
320 vector signed short lOY = c->OY; \
321 vector signed short lCRV = c->CRV; \
322 vector signed short lCBU = c->CBU; \
323 vector signed short lCGU = c->CGU; \
324 vector signed short lCGV = c->CGV; \
325 vector unsigned short lCSHIFT = c->CSHIFT; \
326 \
327 const ubyte *y1i = in[0]; \
328 const ubyte *y2i = in[0] + instrides[0]; \
329 const ubyte *ui = in[1]; \
330 const ubyte *vi = in[2]; \
331 \
332 vector unsigned char *oute, *outo; \
333 \
334 /* loop moves y{1, 2}i by w */ \
335 instrides_scl[0] = instrides[0] * 2 - w; \
336 /* loop moves ui by w / 2 */ \
337 instrides_scl[1] = instrides[1] - w / 2; \
338 /* loop moves vi by w / 2 */ \
339 instrides_scl[2] = instrides[2] - w / 2; \
340 \
341 for (i = 0; i < h / 2; i++) { \
342 oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
343 (srcSliceY + i * 2)); \
344 outo = oute + (outstrides[0] >> 4); \
345 vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
346 vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
347 \
348 for (j = 0; j < w / 16; j++) { \
349 y0 = vec_xl(0, y1i); \
350 \
351 y1 = vec_xl(0, y2i); \
352 \
353 u = (vector signed char) vec_xl(0, ui); \
354 \
355 v = (vector signed char) vec_xl(0, vi); \
356 \
357 u = (vector signed char) \
358 vec_sub(u, \
359 (vector signed char) \
360 vec_splat((vector signed char) { 128 }, 0)); \
361 v = (vector signed char) \
362 vec_sub(v, \
363 (vector signed char) \
364 vec_splat((vector signed char) { 128 }, 0)); \
365 \
366 U = vec_unpackh(u); \
367 V = vec_unpackh(v); \
368 \
369 Y0 = vec_unh(y0); \
370 Y1 = vec_unl(y0); \
371 Y2 = vec_unh(y1); \
372 Y3 = vec_unl(y1); \
373 \
374 Y0 = vec_mradds(Y0, lCY, lOY); \
375 Y1 = vec_mradds(Y1, lCY, lOY); \
376 Y2 = vec_mradds(Y2, lCY, lOY); \
377 Y3 = vec_mradds(Y3, lCY, lOY); \
378 \
379 /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \
380 ux = vec_sl(U, lCSHIFT); \
381 ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
382 ux0 = vec_mergeh(ux, ux); \
383 ux1 = vec_mergel(ux, ux); \
384 \
385 /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \
386 vx = vec_sl(V, lCSHIFT); \
387 vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
388 vx0 = vec_mergeh(vx, vx); \
389 vx1 = vec_mergel(vx, vx); \
390 \
391 /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \
392 uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
393 uvx = vec_mradds(V, lCGV, uvx); \
394 uvx0 = vec_mergeh(uvx, uvx); \
395 uvx1 = vec_mergel(uvx, uvx); \
396 \
397 R0 = vec_add(Y0, vx0); \
398 G0 = vec_add(Y0, uvx0); \
399 B0 = vec_add(Y0, ux0); \
400 R1 = vec_add(Y1, vx1); \
401 G1 = vec_add(Y1, uvx1); \
402 B1 = vec_add(Y1, ux1); \
403 \
404 R = vec_packclp(R0, R1); \
405 G = vec_packclp(G0, G1); \
406 B = vec_packclp(B0, B1); \
407 \
408 out_pixels(R, G, B, oute); \
409 \
410 R0 = vec_add(Y2, vx0); \
411 G0 = vec_add(Y2, uvx0); \
412 B0 = vec_add(Y2, ux0); \
413 R1 = vec_add(Y3, vx1); \
414 G1 = vec_add(Y3, uvx1); \
415 B1 = vec_add(Y3, ux1); \
416 R = vec_packclp(R0, R1); \
417 G = vec_packclp(G0, G1); \
418 B = vec_packclp(B0, B1); \
419 \
420 \
421 out_pixels(R, G, B, outo); \
422 \
423 y1i += 16; \
424 y2i += 16; \
425 ui += 8; \
426 vi += 8; \
427 } \
428 \
429 ui += instrides_scl[1]; \
430 vi += instrides_scl[2]; \
431 y1i += instrides_scl[0]; \
432 y2i += instrides_scl[0]; \
433 } \
434 return srcSliceH; \
435 }
436
437 #define out_abgr(a, b, c, ptr) \
438 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), c, b, a, ptr)
439 #define out_bgra(a, b, c, ptr) \
440 vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), ptr)
441 #define out_rgba(a, b, c, ptr) \
442 vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), ptr)
443 #define out_argb(a, b, c, ptr) \
444 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) vec_splat((__typeof__(a)){ 255 }, 0)), a, b, c, ptr)
445 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
446 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
447
448 DEFCSP420_CVT(yuv2_abgr, out_abgr)
449 DEFCSP420_CVT(yuv2_bgra, out_bgra)
450 DEFCSP420_CVT(yuv2_rgba, out_rgba)
451 DEFCSP420_CVT(yuv2_argb, out_argb)
452 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
453 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
454
455 // uyvy|uyvy|uyvy|uyvy
456 // 0123 4567 89ab cdef
457 static const vector unsigned char
458 demux_u = { 0x10, 0x00, 0x10, 0x00,
459 0x10, 0x04, 0x10, 0x04,
460 0x10, 0x08, 0x10, 0x08,
461 0x10, 0x0c, 0x10, 0x0c },
462 demux_v = { 0x10, 0x02, 0x10, 0x02,
463 0x10, 0x06, 0x10, 0x06,
464 0x10, 0x0A, 0x10, 0x0A,
465 0x10, 0x0E, 0x10, 0x0E },
466 demux_y = { 0x10, 0x01, 0x10, 0x03,
467 0x10, 0x05, 0x10, 0x07,
468 0x10, 0x09, 0x10, 0x0B,
469 0x10, 0x0D, 0x10, 0x0F };
470
471 /*
472 * this is so I can play live CCIR raw video
473 */
altivec_uyvy_rgb32(SwsContext * c,const unsigned char ** in,int * instrides,int srcSliceY,int srcSliceH,unsigned char ** oplanes,int * outstrides)474 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
475 int *instrides, int srcSliceY, int srcSliceH,
476 unsigned char **oplanes, int *outstrides)
477 {
478 int w = c->srcW;
479 int h = srcSliceH;
480 int i, j;
481 vector unsigned char uyvy;
482 vector signed short Y, U, V;
483 vector signed short R0, G0, B0, R1, G1, B1;
484 vector unsigned char R, G, B;
485 vector unsigned char *out;
486 const ubyte *img;
487
488 img = in[0];
489 out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
490
491 for (i = 0; i < h; i++)
492 for (j = 0; j < w / 16; j++) {
493 uyvy = vec_ld(0, img);
494
495 U = (vector signed short)
496 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
497 V = (vector signed short)
498 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
499 Y = (vector signed short)
500 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
501
502 cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
503
504 uyvy = vec_ld(16, img);
505
506 U = (vector signed short)
507 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
508 V = (vector signed short)
509 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
510 Y = (vector signed short)
511 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
512
513 cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
514
515 R = vec_packclp(R0, R1);
516 G = vec_packclp(G0, G1);
517 B = vec_packclp(B0, B1);
518
519 // vec_mstbgr24 (R,G,B, out);
520 out_rgba(R, G, B, out);
521
522 img += 32;
523 }
524 return srcSliceH;
525 }
526
527 #endif /* HAVE_ALTIVEC */
528
529 /* Ok currently the acceleration routine only supports
530 * inputs of widths a multiple of 16
531 * and heights a multiple 2
532 *
533 * So we just fall back to the C codes for this.
534 */
ff_yuv2rgb_init_ppc(SwsContext * c)535 av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c)
536 {
537 #if HAVE_ALTIVEC
538 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
539 return NULL;
540
541 /*
542 * and this seems not to matter too much I tried a bunch of
543 * videos with abnormal widths and MPlayer crashes elsewhere.
544 * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
545 * boom with X11 bad match.
546 *
547 */
548 if ((c->srcW & 0xf) != 0)
549 return NULL;
550
551 switch (c->srcFormat) {
552 case AV_PIX_FMT_YUV410P:
553 case AV_PIX_FMT_YUV420P:
554 /*case IMGFMT_CLPL: ??? */
555 case AV_PIX_FMT_GRAY8:
556 case AV_PIX_FMT_NV12:
557 case AV_PIX_FMT_NV21:
558 if ((c->srcH & 0x1) != 0)
559 return NULL;
560
561 switch (c->dstFormat) {
562 case AV_PIX_FMT_RGB24:
563 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
564 return altivec_yuv2_rgb24;
565 case AV_PIX_FMT_BGR24:
566 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
567 return altivec_yuv2_bgr24;
568 case AV_PIX_FMT_ARGB:
569 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
570 return altivec_yuv2_argb;
571 case AV_PIX_FMT_ABGR:
572 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
573 return altivec_yuv2_abgr;
574 case AV_PIX_FMT_RGBA:
575 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
576 return altivec_yuv2_rgba;
577 case AV_PIX_FMT_BGRA:
578 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
579 return altivec_yuv2_bgra;
580 default: return NULL;
581 }
582 break;
583
584 case AV_PIX_FMT_UYVY422:
585 switch (c->dstFormat) {
586 case AV_PIX_FMT_BGR32:
587 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
588 return altivec_uyvy_rgb32;
589 default: return NULL;
590 }
591 break;
592 }
593 #endif /* HAVE_ALTIVEC */
594
595 return NULL;
596 }
597
ff_yuv2rgb_init_tables_ppc(SwsContext * c,const int inv_table[4],int brightness,int contrast,int saturation)598 av_cold void ff_yuv2rgb_init_tables_ppc(SwsContext *c,
599 const int inv_table[4],
600 int brightness,
601 int contrast,
602 int saturation)
603 {
604 #if HAVE_ALTIVEC
605 union {
606 DECLARE_ALIGNED(16, signed short, tmp)[8];
607 vector signed short vec;
608 } buf;
609
610 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
611 return;
612
613 buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy
614 buf.tmp[1] = -256 * brightness; // oy
615 buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv
616 buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu
617 buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu
618 buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv
619
620 c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
621 c->CY = vec_splat((vector signed short) buf.vec, 0);
622 c->OY = vec_splat((vector signed short) buf.vec, 1);
623 c->CRV = vec_splat((vector signed short) buf.vec, 2);
624 c->CBU = vec_splat((vector signed short) buf.vec, 3);
625 c->CGU = vec_splat((vector signed short) buf.vec, 4);
626 c->CGV = vec_splat((vector signed short) buf.vec, 5);
627 return;
628 #endif /* HAVE_ALTIVEC */
629 }
630
631 #if HAVE_ALTIVEC
632
yuv2packedX_altivec(SwsContext * c,const int16_t * lumFilter,const int16_t ** lumSrc,int lumFilterSize,const int16_t * chrFilter,const int16_t ** chrUSrc,const int16_t ** chrVSrc,int chrFilterSize,const int16_t ** alpSrc,uint8_t * dest,int dstW,int dstY,enum AVPixelFormat target)633 static av_always_inline void yuv2packedX_altivec(SwsContext *c,
634 const int16_t *lumFilter,
635 const int16_t **lumSrc,
636 int lumFilterSize,
637 const int16_t *chrFilter,
638 const int16_t **chrUSrc,
639 const int16_t **chrVSrc,
640 int chrFilterSize,
641 const int16_t **alpSrc,
642 uint8_t *dest,
643 int dstW, int dstY,
644 enum AVPixelFormat target)
645 {
646 int i, j;
647 vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
648 vector signed short R0, G0, B0, R1, G1, B1;
649
650 vector unsigned char R, G, B;
651 vector unsigned char *out, *nout;
652
653 vector signed short RND = vec_splat_s16(1 << 3);
654 vector unsigned short SCL = vec_splat_u16(4);
655 DECLARE_ALIGNED(16, unsigned int, scratch)[16];
656
657 vector signed short *YCoeffs, *CCoeffs;
658
659 YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
660 CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
661
662 out = (vector unsigned char *) dest;
663
664 for (i = 0; i < dstW; i += 16) {
665 Y0 = RND;
666 Y1 = RND;
667 /* extract 16 coeffs from lumSrc */
668 for (j = 0; j < lumFilterSize; j++) {
669 X0 = vec_ld(0, &lumSrc[j][i]);
670 X1 = vec_ld(16, &lumSrc[j][i]);
671 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
672 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
673 }
674
675 U = RND;
676 V = RND;
677 /* extract 8 coeffs from U,V */
678 for (j = 0; j < chrFilterSize; j++) {
679 X = vec_ld(0, &chrUSrc[j][i / 2]);
680 U = vec_mradds(X, CCoeffs[j], U);
681 X = vec_ld(0, &chrVSrc[j][i / 2]);
682 V = vec_mradds(X, CCoeffs[j], V);
683 }
684
685 /* scale and clip signals */
686 Y0 = vec_sra(Y0, SCL);
687 Y1 = vec_sra(Y1, SCL);
688 U = vec_sra(U, SCL);
689 V = vec_sra(V, SCL);
690
691 Y0 = vec_clip_s16(Y0);
692 Y1 = vec_clip_s16(Y1);
693 U = vec_clip_s16(U);
694 V = vec_clip_s16(V);
695
696 /* now we have
697 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
698 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
699 *
700 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
701 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
702 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
703 */
704
705 U0 = vec_mergeh(U, U);
706 V0 = vec_mergeh(V, V);
707
708 U1 = vec_mergel(U, U);
709 V1 = vec_mergel(V, V);
710
711 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
712 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
713
714 R = vec_packclp(R0, R1);
715 G = vec_packclp(G0, G1);
716 B = vec_packclp(B0, B1);
717
718 switch (target) {
719 case AV_PIX_FMT_ABGR:
720 out_abgr(R, G, B, out);
721 break;
722 case AV_PIX_FMT_BGRA:
723 out_bgra(R, G, B, out);
724 break;
725 case AV_PIX_FMT_RGBA:
726 out_rgba(R, G, B, out);
727 break;
728 case AV_PIX_FMT_ARGB:
729 out_argb(R, G, B, out);
730 break;
731 case AV_PIX_FMT_RGB24:
732 out_rgb24(R, G, B, out);
733 break;
734 case AV_PIX_FMT_BGR24:
735 out_bgr24(R, G, B, out);
736 break;
737 default:
738 {
739 /* If this is reached, the caller should have called yuv2packedXinC
740 * instead. */
741 static int printed_error_message;
742 if (!printed_error_message) {
743 av_log(c, AV_LOG_ERROR,
744 "altivec_yuv2packedX doesn't support %s output\n",
745 av_get_pix_fmt_name(c->dstFormat));
746 printed_error_message = 1;
747 }
748 return;
749 }
750 }
751 }
752
753 if (i < dstW) {
754 i -= 16;
755
756 Y0 = RND;
757 Y1 = RND;
758 /* extract 16 coeffs from lumSrc */
759 for (j = 0; j < lumFilterSize; j++) {
760 X0 = vec_ld(0, &lumSrc[j][i]);
761 X1 = vec_ld(16, &lumSrc[j][i]);
762 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
763 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
764 }
765
766 U = RND;
767 V = RND;
768 /* extract 8 coeffs from U,V */
769 for (j = 0; j < chrFilterSize; j++) {
770 X = vec_ld(0, &chrUSrc[j][i / 2]);
771 U = vec_mradds(X, CCoeffs[j], U);
772 X = vec_ld(0, &chrVSrc[j][i / 2]);
773 V = vec_mradds(X, CCoeffs[j], V);
774 }
775
776 /* scale and clip signals */
777 Y0 = vec_sra(Y0, SCL);
778 Y1 = vec_sra(Y1, SCL);
779 U = vec_sra(U, SCL);
780 V = vec_sra(V, SCL);
781
782 Y0 = vec_clip_s16(Y0);
783 Y1 = vec_clip_s16(Y1);
784 U = vec_clip_s16(U);
785 V = vec_clip_s16(V);
786
787 /* now we have
788 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
789 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
790 *
791 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
792 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
793 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
794 */
795
796 U0 = vec_mergeh(U, U);
797 V0 = vec_mergeh(V, V);
798
799 U1 = vec_mergel(U, U);
800 V1 = vec_mergel(V, V);
801
802 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
803 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
804
805 R = vec_packclp(R0, R1);
806 G = vec_packclp(G0, G1);
807 B = vec_packclp(B0, B1);
808
809 nout = (vector unsigned char *) scratch;
810 switch (target) {
811 case AV_PIX_FMT_ABGR:
812 out_abgr(R, G, B, nout);
813 break;
814 case AV_PIX_FMT_BGRA:
815 out_bgra(R, G, B, nout);
816 break;
817 case AV_PIX_FMT_RGBA:
818 out_rgba(R, G, B, nout);
819 break;
820 case AV_PIX_FMT_ARGB:
821 out_argb(R, G, B, nout);
822 break;
823 case AV_PIX_FMT_RGB24:
824 out_rgb24(R, G, B, nout);
825 break;
826 case AV_PIX_FMT_BGR24:
827 out_bgr24(R, G, B, nout);
828 break;
829 default:
830 /* Unreachable, I think. */
831 av_log(c, AV_LOG_ERROR,
832 "altivec_yuv2packedX doesn't support %s output\n",
833 av_get_pix_fmt_name(c->dstFormat));
834 return;
835 }
836
837 memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
838 }
839 }
840
841 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
842 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \
843 const int16_t *lumFilter, \
844 const int16_t **lumSrc, \
845 int lumFilterSize, \
846 const int16_t *chrFilter, \
847 const int16_t **chrUSrc, \
848 const int16_t **chrVSrc, \
849 int chrFilterSize, \
850 const int16_t **alpSrc, \
851 uint8_t *dest, int dstW, int dstY) \
852 { \
853 yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
854 chrFilter, chrUSrc, chrVSrc, \
855 chrFilterSize, alpSrc, \
856 dest, dstW, dstY, pixfmt); \
857 }
858
859 YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR);
860 YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA);
861 YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB);
862 YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA);
863 YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
864 YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);
865
866 #endif /* HAVE_ALTIVEC */
867