2 * AltiVec acceleration for colorspace conversion
4 * copyright (C) 2004 Marc Hoffman <marc.hoffman@analog.com>
6 * This file is part of FFmpeg.
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.
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.
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
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.
30 * Lots of optimizations to be done here.
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.
35 * 2. The inefficient use of chroma loading needs a bit of brushing up.
37 * 3. Analysis of pipeline stalls needs to be done. Use shark to identify
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.
47 * ADDED altivec_yuv2packedX vertical scl + RGB converter
50 * PERFORMANCE ANALYSIS
52 * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo
54 * The AltiVec version uses 10% of the processor or ~100Mips for D1 video
57 * 720 * 480 * 30 ~10MPS
59 * so we have roughly 10 clocks per pixel. This is too high, something has
62 * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the
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.
72 * Now memcpy is the largest mips consumer in the system, probably due
73 * to the inefficient X11 stuff.
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.)
80 * GL2 libraries work now with patch for RGB32.
82 * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor.
84 * Integrated luma prescaling adjustment for saturation/contrast/brightness
95 #include "libswscale/rgb2rgb.h"
96 #include "libswscale/swscale.h"
97 #include "libswscale/swscale_internal.h"
98 #include "libavutil/attributes.h"
99 #include "libavutil/cpu.h"
100 #include "libavutil/pixdesc.h"
101 #include "yuv2rgb_altivec.h"
103 #undef PROFILE_THE_BEAST
106 typedef unsigned char ubyte;
107 typedef signed char sbyte;
109 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
110 * homogeneous vector registers x0,x1,x2 are interleaved with the
111 * following technique:
113 * o0 = vec_mergeh(x0, x1);
114 * o1 = vec_perm(o0, x2, perm_rgb_0);
115 * o2 = vec_perm(o0, x2, perm_rgb_1);
116 * o3 = vec_mergel(x0, x1);
117 * o4 = vec_perm(o3, o2, perm_rgb_2);
118 * o5 = vec_perm(o3, o2, perm_rgb_3);
120 * perm_rgb_0: o0(RG).h v1(B) --> o1*
122 * rgbr|gbrg|brgb|rgbr
123 * 0010 0100 1001 0010
124 * 0102 3145 2673 894A
126 * perm_rgb_1: o0(RG).h v1(B) --> o2
128 * gbrg|brgb|bbbb|bbbb
129 * 0100 1001 1111 1111
130 * B5CD 6EF7 89AB CDEF
132 * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4*
134 * gbrg|brgb|rgbr|gbrg
135 * 1111 1111 0010 0100
136 * 89AB CDEF 0182 3945
138 * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5*
140 * brgb|rgbr|gbrg|brgb
141 * 1001 0010 0100 1001
142 * a67b 89cA BdCD eEFf
145 static const vector unsigned char
146 perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05,
147 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a },
148 perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17,
149 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f },
150 perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
151 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 },
152 perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a,
153 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f };
155 #define vec_merge3(x2, x1, x0, y0, y1, y2) \
157 __typeof__(x0) o0, o2, o3; \
158 o0 = vec_mergeh(x0, x1); \
159 y0 = vec_perm(o0, x2, perm_rgb_0); \
160 o2 = vec_perm(o0, x2, perm_rgb_1); \
161 o3 = vec_mergel(x0, x1); \
162 y1 = vec_perm(o3, o2, perm_rgb_2); \
163 y2 = vec_perm(o3, o2, perm_rgb_3); \
166 #define vec_mstbgr24(x0, x1, x2, ptr) \
168 __typeof__(x0) _0, _1, _2; \
169 vec_merge3(x0, x1, x2, _0, _1, _2); \
170 vec_st(_0, 0, ptr++); \
171 vec_st(_1, 0, ptr++); \
172 vec_st(_2, 0, ptr++); \
175 #define vec_mstrgb24(x0, x1, x2, ptr) \
177 __typeof__(x0) _0, _1, _2; \
178 vec_merge3(x2, x1, x0, _0, _1, _2); \
179 vec_st(_0, 0, ptr++); \
180 vec_st(_1, 0, ptr++); \
181 vec_st(_2, 0, ptr++); \
184 /* pack the pixels in rgb0 format
188 #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \
191 _0 = vec_mergeh(x0, x1); \
192 _1 = vec_mergeh(x2, x3); \
193 _2 = (T) vec_mergeh((vector unsigned short) _0, \
194 (vector unsigned short) _1); \
195 _3 = (T) vec_mergel((vector unsigned short) _0, \
196 (vector unsigned short) _1); \
197 vec_st(_2, 0 * 16, (T *) ptr); \
198 vec_st(_3, 1 * 16, (T *) ptr); \
199 _0 = vec_mergel(x0, x1); \
200 _1 = vec_mergel(x2, x3); \
201 _2 = (T) vec_mergeh((vector unsigned short) _0, \
202 (vector unsigned short) _1); \
203 _3 = (T) vec_mergel((vector unsigned short) _0, \
204 (vector unsigned short) _1); \
205 vec_st(_2, 2 * 16, (T *) ptr); \
206 vec_st(_3, 3 * 16, (T *) ptr); \
212 * 1 -0.3441 -0.7142 |x| Cb|
219 * typical YUV conversion works on Y: 0-255 this version has been
220 * optimized for JPEG decoding.
224 (vector signed short) \
225 vec_perm(x, (__typeof__(x)) { 0 }, \
226 ((vector unsigned char) { \
227 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
228 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
231 (vector signed short) \
232 vec_perm(x, (__typeof__(x)) { 0 }, \
233 ((vector unsigned char) { \
234 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
235 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
237 #define vec_clip_s16(x) \
238 vec_max(vec_min(x, ((vector signed short) { \
239 235, 235, 235, 235, 235, 235, 235, 235 })), \
240 ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
242 #define vec_packclp(x, y) \
243 (vector unsigned char) \
244 vec_packs((vector unsigned short) \
245 vec_max(x, ((vector signed short) { 0 })), \
246 (vector unsigned short) \
247 vec_max(y, ((vector signed short) { 0 })))
249 //#define out_pixels(a, b, c, ptr) vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, a, a, ptr)
251 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
252 vector signed short U, vector signed short V,
253 vector signed short *R, vector signed short *G,
254 vector signed short *B)
256 vector signed short vx, ux, uvx;
258 Y = vec_mradds(Y, c->CY, c->OY);
259 U = vec_sub(U, (vector signed short)
260 vec_splat((vector signed short) { 128 }, 0));
261 V = vec_sub(V, (vector signed short)
262 vec_splat((vector signed short) { 128 }, 0));
264 // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
265 ux = vec_sl(U, c->CSHIFT);
266 *B = vec_mradds(ux, c->CBU, Y);
268 // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
269 vx = vec_sl(V, c->CSHIFT);
270 *R = vec_mradds(vx, c->CRV, Y);
272 // uvx = ((CGU * u) + (CGV * v)) >> 15;
273 uvx = vec_mradds(U, c->CGU, Y);
274 *G = vec_mradds(V, c->CGV, uvx);
278 * ------------------------------------------------------------------------------
280 * ------------------------------------------------------------------------------
283 #define DEFCSP420_CVT(name, out_pixels) \
284 static int altivec_ ## name(SwsContext *c, const unsigned char **in, \
285 int *instrides, int srcSliceY, int srcSliceH, \
286 unsigned char **oplanes, int *outstrides) \
291 int instrides_scl[3]; \
292 vector unsigned char y0, y1; \
294 vector signed char u, v; \
296 vector signed short Y0, Y1, Y2, Y3; \
297 vector signed short U, V; \
298 vector signed short vx, ux, uvx; \
299 vector signed short vx0, ux0, uvx0; \
300 vector signed short vx1, ux1, uvx1; \
301 vector signed short R0, G0, B0; \
302 vector signed short R1, G1, B1; \
303 vector unsigned char R, G, B; \
305 const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \
306 vector unsigned char align_perm; \
308 vector signed short lCY = c->CY; \
309 vector signed short lOY = c->OY; \
310 vector signed short lCRV = c->CRV; \
311 vector signed short lCBU = c->CBU; \
312 vector signed short lCGU = c->CGU; \
313 vector signed short lCGV = c->CGV; \
314 vector unsigned short lCSHIFT = c->CSHIFT; \
316 const ubyte *y1i = in[0]; \
317 const ubyte *y2i = in[0] + instrides[0]; \
318 const ubyte *ui = in[1]; \
319 const ubyte *vi = in[2]; \
321 vector unsigned char *oute, *outo; \
323 /* loop moves y{1, 2}i by w */ \
324 instrides_scl[0] = instrides[0] * 2 - w; \
325 /* loop moves ui by w / 2 */ \
326 instrides_scl[1] = instrides[1] - w / 2; \
327 /* loop moves vi by w / 2 */ \
328 instrides_scl[2] = instrides[2] - w / 2; \
330 for (i = 0; i < h / 2; i++) { \
331 oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
332 (srcSliceY + i * 2)); \
333 outo = oute + (outstrides[0] >> 4); \
334 vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
335 vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
337 for (j = 0; j < w / 16; j++) { \
338 y1ivP = (const vector unsigned char *) y1i; \
339 y2ivP = (const vector unsigned char *) y2i; \
340 uivP = (const vector unsigned char *) ui; \
341 vivP = (const vector unsigned char *) vi; \
343 align_perm = vec_lvsl(0, y1i); \
344 y0 = (vector unsigned char) \
345 vec_perm(y1ivP[0], y1ivP[1], align_perm); \
347 align_perm = vec_lvsl(0, y2i); \
348 y1 = (vector unsigned char) \
349 vec_perm(y2ivP[0], y2ivP[1], align_perm); \
351 align_perm = vec_lvsl(0, ui); \
352 u = (vector signed char) \
353 vec_perm(uivP[0], uivP[1], align_perm); \
355 align_perm = vec_lvsl(0, vi); \
356 v = (vector signed char) \
357 vec_perm(vivP[0], vivP[1], align_perm); \
359 u = (vector signed char) \
361 (vector signed char) \
362 vec_splat((vector signed char) { 128 }, 0)); \
363 v = (vector signed char) \
365 (vector signed char) \
366 vec_splat((vector signed char) { 128 }, 0)); \
368 U = vec_unpackh(u); \
369 V = vec_unpackh(v); \
376 Y0 = vec_mradds(Y0, lCY, lOY); \
377 Y1 = vec_mradds(Y1, lCY, lOY); \
378 Y2 = vec_mradds(Y2, lCY, lOY); \
379 Y3 = vec_mradds(Y3, lCY, lOY); \
381 /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \
382 ux = vec_sl(U, lCSHIFT); \
383 ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
384 ux0 = vec_mergeh(ux, ux); \
385 ux1 = vec_mergel(ux, ux); \
387 /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \
388 vx = vec_sl(V, lCSHIFT); \
389 vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
390 vx0 = vec_mergeh(vx, vx); \
391 vx1 = vec_mergel(vx, vx); \
393 /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \
394 uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
395 uvx = vec_mradds(V, lCGV, uvx); \
396 uvx0 = vec_mergeh(uvx, uvx); \
397 uvx1 = vec_mergel(uvx, uvx); \
399 R0 = vec_add(Y0, vx0); \
400 G0 = vec_add(Y0, uvx0); \
401 B0 = vec_add(Y0, ux0); \
402 R1 = vec_add(Y1, vx1); \
403 G1 = vec_add(Y1, uvx1); \
404 B1 = vec_add(Y1, ux1); \
406 R = vec_packclp(R0, R1); \
407 G = vec_packclp(G0, G1); \
408 B = vec_packclp(B0, B1); \
410 out_pixels(R, G, B, oute); \
412 R0 = vec_add(Y2, vx0); \
413 G0 = vec_add(Y2, uvx0); \
414 B0 = vec_add(Y2, ux0); \
415 R1 = vec_add(Y3, vx1); \
416 G1 = vec_add(Y3, uvx1); \
417 B1 = vec_add(Y3, ux1); \
418 R = vec_packclp(R0, R1); \
419 G = vec_packclp(G0, G1); \
420 B = vec_packclp(B0, B1); \
423 out_pixels(R, G, B, outo); \
431 ui += instrides_scl[1]; \
432 vi += instrides_scl[2]; \
433 y1i += instrides_scl[0]; \
434 y2i += instrides_scl[0]; \
439 #define out_abgr(a, b, c, ptr) \
440 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
441 #define out_bgra(a, b, c, ptr) \
442 vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
443 #define out_rgba(a, b, c, ptr) \
444 vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
445 #define out_argb(a, b, c, ptr) \
446 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
447 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
448 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
450 DEFCSP420_CVT(yuv2_abgr, out_abgr)
451 DEFCSP420_CVT(yuv2_bgra, out_bgra)
452 DEFCSP420_CVT(yuv2_rgba, out_rgba)
453 DEFCSP420_CVT(yuv2_argb, out_argb)
454 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
455 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
457 // uyvy|uyvy|uyvy|uyvy
458 // 0123 4567 89ab cdef
459 static const vector unsigned char
460 demux_u = { 0x10, 0x00, 0x10, 0x00,
461 0x10, 0x04, 0x10, 0x04,
462 0x10, 0x08, 0x10, 0x08,
463 0x10, 0x0c, 0x10, 0x0c },
464 demux_v = { 0x10, 0x02, 0x10, 0x02,
465 0x10, 0x06, 0x10, 0x06,
466 0x10, 0x0A, 0x10, 0x0A,
467 0x10, 0x0E, 0x10, 0x0E },
468 demux_y = { 0x10, 0x01, 0x10, 0x03,
469 0x10, 0x05, 0x10, 0x07,
470 0x10, 0x09, 0x10, 0x0B,
471 0x10, 0x0D, 0x10, 0x0F };
474 * this is so I can play live CCIR raw video
476 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
477 int *instrides, int srcSliceY, int srcSliceH,
478 unsigned char **oplanes, int *outstrides)
483 vector unsigned char uyvy;
484 vector signed short Y, U, V;
485 vector signed short R0, G0, B0, R1, G1, B1;
486 vector unsigned char R, G, B;
487 vector unsigned char *out;
491 out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
493 for (i = 0; i < h; i++)
494 for (j = 0; j < w / 16; j++) {
495 uyvy = vec_ld(0, img);
497 U = (vector signed short)
498 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
499 V = (vector signed short)
500 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
501 Y = (vector signed short)
502 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
504 cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
506 uyvy = vec_ld(16, img);
508 U = (vector signed short)
509 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
510 V = (vector signed short)
511 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
512 Y = (vector signed short)
513 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
515 cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
517 R = vec_packclp(R0, R1);
518 G = vec_packclp(G0, G1);
519 B = vec_packclp(B0, B1);
521 // vec_mstbgr24 (R,G,B, out);
522 out_rgba(R, G, B, out);
529 /* Ok currently the acceleration routine only supports
530 * inputs of widths a multiple of 16
531 * and heights a multiple 2
533 * So we just fall back to the C codes for this.
535 av_cold SwsFunc ff_yuv2rgb_init_altivec(SwsContext *c)
537 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
541 * and this seems not to matter too much I tried a bunch of
542 * videos with abnormal widths and MPlayer crashes elsewhere.
543 * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
544 * boom with X11 bad match.
547 if ((c->srcW & 0xf) != 0)
550 switch (c->srcFormat) {
551 case AV_PIX_FMT_YUV410P:
552 case AV_PIX_FMT_YUV420P:
553 /*case IMGFMT_CLPL: ??? */
554 case AV_PIX_FMT_GRAY8:
555 case AV_PIX_FMT_NV12:
556 case AV_PIX_FMT_NV21:
557 if ((c->srcH & 0x1) != 0)
560 switch (c->dstFormat) {
561 case AV_PIX_FMT_RGB24:
562 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
563 return altivec_yuv2_rgb24;
564 case AV_PIX_FMT_BGR24:
565 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
566 return altivec_yuv2_bgr24;
567 case AV_PIX_FMT_ARGB:
568 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
569 return altivec_yuv2_argb;
570 case AV_PIX_FMT_ABGR:
571 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
572 return altivec_yuv2_abgr;
573 case AV_PIX_FMT_RGBA:
574 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
575 return altivec_yuv2_rgba;
576 case AV_PIX_FMT_BGRA:
577 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
578 return altivec_yuv2_bgra;
579 default: return NULL;
583 case AV_PIX_FMT_UYVY422:
584 switch (c->dstFormat) {
585 case AV_PIX_FMT_BGR32:
586 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
587 return altivec_uyvy_rgb32;
588 default: return NULL;
595 av_cold void ff_yuv2rgb_init_tables_altivec(SwsContext *c,
596 const int inv_table[4],
602 DECLARE_ALIGNED(16, signed short, tmp)[8];
603 vector signed short vec;
606 buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy
607 buf.tmp[1] = -256 * brightness; // oy
608 buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv
609 buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu
610 buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu
611 buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv
613 c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
614 c->CY = vec_splat((vector signed short) buf.vec, 0);
615 c->OY = vec_splat((vector signed short) buf.vec, 1);
616 c->CRV = vec_splat((vector signed short) buf.vec, 2);
617 c->CBU = vec_splat((vector signed short) buf.vec, 3);
618 c->CGU = vec_splat((vector signed short) buf.vec, 4);
619 c->CGV = vec_splat((vector signed short) buf.vec, 5);
623 static av_always_inline void ff_yuv2packedX_altivec(SwsContext *c,
624 const int16_t *lumFilter,
625 const int16_t **lumSrc,
627 const int16_t *chrFilter,
628 const int16_t **chrUSrc,
629 const int16_t **chrVSrc,
631 const int16_t **alpSrc,
634 enum AVPixelFormat target)
637 vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
638 vector signed short R0, G0, B0, R1, G1, B1;
640 vector unsigned char R, G, B;
641 vector unsigned char *out, *nout;
643 vector signed short RND = vec_splat_s16(1 << 3);
644 vector unsigned short SCL = vec_splat_u16(4);
645 DECLARE_ALIGNED(16, unsigned int, scratch)[16];
647 vector signed short *YCoeffs, *CCoeffs;
649 YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
650 CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
652 out = (vector unsigned char *) dest;
654 for (i = 0; i < dstW; i += 16) {
657 /* extract 16 coeffs from lumSrc */
658 for (j = 0; j < lumFilterSize; j++) {
659 X0 = vec_ld(0, &lumSrc[j][i]);
660 X1 = vec_ld(16, &lumSrc[j][i]);
661 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
662 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
667 /* extract 8 coeffs from U,V */
668 for (j = 0; j < chrFilterSize; j++) {
669 X = vec_ld(0, &chrUSrc[j][i / 2]);
670 U = vec_mradds(X, CCoeffs[j], U);
671 X = vec_ld(0, &chrVSrc[j][i / 2]);
672 V = vec_mradds(X, CCoeffs[j], V);
675 /* scale and clip signals */
676 Y0 = vec_sra(Y0, SCL);
677 Y1 = vec_sra(Y1, SCL);
681 Y0 = vec_clip_s16(Y0);
682 Y1 = vec_clip_s16(Y1);
687 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
688 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
690 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
691 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
692 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
695 U0 = vec_mergeh(U, U);
696 V0 = vec_mergeh(V, V);
698 U1 = vec_mergel(U, U);
699 V1 = vec_mergel(V, V);
701 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
702 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
704 R = vec_packclp(R0, R1);
705 G = vec_packclp(G0, G1);
706 B = vec_packclp(B0, B1);
709 case AV_PIX_FMT_ABGR:
710 out_abgr(R, G, B, out);
712 case AV_PIX_FMT_BGRA:
713 out_bgra(R, G, B, out);
715 case AV_PIX_FMT_RGBA:
716 out_rgba(R, G, B, out);
718 case AV_PIX_FMT_ARGB:
719 out_argb(R, G, B, out);
721 case AV_PIX_FMT_RGB24:
722 out_rgb24(R, G, B, out);
724 case AV_PIX_FMT_BGR24:
725 out_bgr24(R, G, B, out);
729 /* If this is reached, the caller should have called yuv2packedXinC
731 static int printed_error_message;
732 if (!printed_error_message) {
733 av_log(c, AV_LOG_ERROR,
734 "altivec_yuv2packedX doesn't support %s output\n",
735 av_get_pix_fmt_name(c->dstFormat));
736 printed_error_message = 1;
748 /* extract 16 coeffs from lumSrc */
749 for (j = 0; j < lumFilterSize; j++) {
750 X0 = vec_ld(0, &lumSrc[j][i]);
751 X1 = vec_ld(16, &lumSrc[j][i]);
752 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
753 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
758 /* extract 8 coeffs from U,V */
759 for (j = 0; j < chrFilterSize; j++) {
760 X = vec_ld(0, &chrUSrc[j][i / 2]);
761 U = vec_mradds(X, CCoeffs[j], U);
762 X = vec_ld(0, &chrVSrc[j][i / 2]);
763 V = vec_mradds(X, CCoeffs[j], V);
766 /* scale and clip signals */
767 Y0 = vec_sra(Y0, SCL);
768 Y1 = vec_sra(Y1, SCL);
772 Y0 = vec_clip_s16(Y0);
773 Y1 = vec_clip_s16(Y1);
778 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
779 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
781 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
782 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
783 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
786 U0 = vec_mergeh(U, U);
787 V0 = vec_mergeh(V, V);
789 U1 = vec_mergel(U, U);
790 V1 = vec_mergel(V, V);
792 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
793 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
795 R = vec_packclp(R0, R1);
796 G = vec_packclp(G0, G1);
797 B = vec_packclp(B0, B1);
799 nout = (vector unsigned char *) scratch;
801 case AV_PIX_FMT_ABGR:
802 out_abgr(R, G, B, nout);
804 case AV_PIX_FMT_BGRA:
805 out_bgra(R, G, B, nout);
807 case AV_PIX_FMT_RGBA:
808 out_rgba(R, G, B, nout);
810 case AV_PIX_FMT_ARGB:
811 out_argb(R, G, B, nout);
813 case AV_PIX_FMT_RGB24:
814 out_rgb24(R, G, B, nout);
816 case AV_PIX_FMT_BGR24:
817 out_bgr24(R, G, B, nout);
820 /* Unreachable, I think. */
821 av_log(c, AV_LOG_ERROR,
822 "altivec_yuv2packedX doesn't support %s output\n",
823 av_get_pix_fmt_name(c->dstFormat));
827 memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
831 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
832 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \
833 const int16_t *lumFilter, \
834 const int16_t **lumSrc, \
836 const int16_t *chrFilter, \
837 const int16_t **chrUSrc, \
838 const int16_t **chrVSrc, \
840 const int16_t **alpSrc, \
841 uint8_t *dest, int dstW, int dstY) \
843 ff_yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
844 chrFilter, chrUSrc, chrVSrc, \
845 chrFilterSize, alpSrc, \
846 dest, dstW, dstY, pixfmt); \
849 YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR);
850 YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA);
851 YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB);
852 YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA);
853 YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
854 YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);