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
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/pixdesc.h"
100 #include "yuv2rgb_altivec.h"
104 #undef PROFILE_THE_BEAST
107 typedef unsigned char ubyte;
108 typedef signed char sbyte;
110 /* RGB interleaver, 16 planar pels 8-bit samples per channel in
111 * homogeneous vector registers x0,x1,x2 are interleaved with the
112 * following technique:
114 * o0 = vec_mergeh(x0, x1);
115 * o1 = vec_perm(o0, x2, perm_rgb_0);
116 * o2 = vec_perm(o0, x2, perm_rgb_1);
117 * o3 = vec_mergel(x0, x1);
118 * o4 = vec_perm(o3, o2, perm_rgb_2);
119 * o5 = vec_perm(o3, o2, perm_rgb_3);
121 * perm_rgb_0: o0(RG).h v1(B) --> o1*
123 * rgbr|gbrg|brgb|rgbr
124 * 0010 0100 1001 0010
125 * 0102 3145 2673 894A
127 * perm_rgb_1: o0(RG).h v1(B) --> o2
129 * gbrg|brgb|bbbb|bbbb
130 * 0100 1001 1111 1111
131 * B5CD 6EF7 89AB CDEF
133 * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4*
135 * gbrg|brgb|rgbr|gbrg
136 * 1111 1111 0010 0100
137 * 89AB CDEF 0182 3945
139 * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5*
141 * brgb|rgbr|gbrg|brgb
142 * 1001 0010 0100 1001
143 * 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.
225 (vector signed short) \
226 vec_perm(x, (__typeof__(x)) { 0 }, \
227 ((vector unsigned char) { \
228 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \
229 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 }))
232 (vector signed short) \
233 vec_perm(x, (__typeof__(x)) { 0 }, \
234 ((vector unsigned char) { \
235 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \
236 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F }))
238 #define vec_unh(x)(vector signed short) vec_mergeh(x,(__typeof__(x)) { 0 })
239 #define vec_unl(x)(vector signed short) vec_mergel(x,(__typeof__(x)) { 0 })
242 #define vec_clip_s16(x) \
243 vec_max(vec_min(x, ((vector signed short) { \
244 235, 235, 235, 235, 235, 235, 235, 235 })), \
245 ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 }))
247 #define vec_packclp(x, y) \
248 (vector unsigned char) \
249 vec_packs((vector unsigned short) \
250 vec_max(x, ((vector signed short) { 0 })), \
251 (vector unsigned short) \
252 vec_max(y, ((vector signed short) { 0 })))
254 static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y,
255 vector signed short U, vector signed short V,
256 vector signed short *R, vector signed short *G,
257 vector signed short *B)
259 vector signed short vx, ux, uvx;
261 Y = vec_mradds(Y, c->CY, c->OY);
262 U = vec_sub(U, (vector signed short)
263 vec_splat((vector signed short) { 128 }, 0));
264 V = vec_sub(V, (vector signed short)
265 vec_splat((vector signed short) { 128 }, 0));
267 // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15;
268 ux = vec_sl(U, c->CSHIFT);
269 *B = vec_mradds(ux, c->CBU, Y);
271 // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15;
272 vx = vec_sl(V, c->CSHIFT);
273 *R = vec_mradds(vx, c->CRV, Y);
275 // uvx = ((CGU * u) + (CGV * v)) >> 15;
276 uvx = vec_mradds(U, c->CGU, Y);
277 *G = vec_mradds(V, c->CGV, uvx);
281 * ------------------------------------------------------------------------------
283 * ------------------------------------------------------------------------------
286 #define DEFCSP420_CVT(name, out_pixels) \
287 static int altivec_ ## name(SwsContext *c, const unsigned char **in, \
288 int *instrides, int srcSliceY, int srcSliceH, \
289 unsigned char **oplanes, int *outstrides) \
294 int instrides_scl[3]; \
295 vector unsigned char y0, y1; \
297 vector signed char u, v; \
299 vector signed short Y0, Y1, Y2, Y3; \
300 vector signed short U, V; \
301 vector signed short vx, ux, uvx; \
302 vector signed short vx0, ux0, uvx0; \
303 vector signed short vx1, ux1, uvx1; \
304 vector signed short R0, G0, B0; \
305 vector signed short R1, G1, B1; \
306 vector unsigned char R, G, B; \
308 const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \
309 vector unsigned char align_perm; \
311 vector signed short lCY = c->CY; \
312 vector signed short lOY = c->OY; \
313 vector signed short lCRV = c->CRV; \
314 vector signed short lCBU = c->CBU; \
315 vector signed short lCGU = c->CGU; \
316 vector signed short lCGV = c->CGV; \
317 vector unsigned short lCSHIFT = c->CSHIFT; \
319 const ubyte *y1i = in[0]; \
320 const ubyte *y2i = in[0] + instrides[0]; \
321 const ubyte *ui = in[1]; \
322 const ubyte *vi = in[2]; \
324 vector unsigned char *oute, *outo; \
326 /* loop moves y{1, 2}i by w */ \
327 instrides_scl[0] = instrides[0] * 2 - w; \
328 /* loop moves ui by w / 2 */ \
329 instrides_scl[1] = instrides[1] - w / 2; \
330 /* loop moves vi by w / 2 */ \
331 instrides_scl[2] = instrides[2] - w / 2; \
333 for (i = 0; i < h / 2; i++) { \
334 oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \
335 (srcSliceY + i * 2)); \
336 outo = oute + (outstrides[0] >> 4); \
337 vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \
338 vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \
340 for (j = 0; j < w / 16; j++) { \
341 y1ivP = (const vector unsigned char *) y1i; \
342 y2ivP = (const vector unsigned char *) y2i; \
343 uivP = (const vector unsigned char *) ui; \
344 vivP = (const vector unsigned char *) vi; \
346 align_perm = vec_lvsl(0, y1i); \
347 y0 = (vector unsigned char) \
348 vec_perm(y1ivP[0], y1ivP[1], align_perm); \
350 align_perm = vec_lvsl(0, y2i); \
351 y1 = (vector unsigned char) \
352 vec_perm(y2ivP[0], y2ivP[1], align_perm); \
354 align_perm = vec_lvsl(0, ui); \
355 u = (vector signed char) \
356 vec_perm(uivP[0], uivP[1], align_perm); \
358 align_perm = vec_lvsl(0, vi); \
359 v = (vector signed char) \
360 vec_perm(vivP[0], vivP[1], align_perm); \
362 u = (vector signed char) \
364 (vector signed char) \
365 vec_splat((vector signed char) { 128 }, 0)); \
366 v = (vector signed char) \
368 (vector signed char) \
369 vec_splat((vector signed char) { 128 }, 0)); \
371 U = vec_unpackh(u); \
372 V = vec_unpackh(v); \
379 Y0 = vec_mradds(Y0, lCY, lOY); \
380 Y1 = vec_mradds(Y1, lCY, lOY); \
381 Y2 = vec_mradds(Y2, lCY, lOY); \
382 Y3 = vec_mradds(Y3, lCY, lOY); \
384 /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \
385 ux = vec_sl(U, lCSHIFT); \
386 ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \
387 ux0 = vec_mergeh(ux, ux); \
388 ux1 = vec_mergel(ux, ux); \
390 /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \
391 vx = vec_sl(V, lCSHIFT); \
392 vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \
393 vx0 = vec_mergeh(vx, vx); \
394 vx1 = vec_mergel(vx, vx); \
396 /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \
397 uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \
398 uvx = vec_mradds(V, lCGV, uvx); \
399 uvx0 = vec_mergeh(uvx, uvx); \
400 uvx1 = vec_mergel(uvx, uvx); \
402 R0 = vec_add(Y0, vx0); \
403 G0 = vec_add(Y0, uvx0); \
404 B0 = vec_add(Y0, ux0); \
405 R1 = vec_add(Y1, vx1); \
406 G1 = vec_add(Y1, uvx1); \
407 B1 = vec_add(Y1, ux1); \
409 R = vec_packclp(R0, R1); \
410 G = vec_packclp(G0, G1); \
411 B = vec_packclp(B0, B1); \
413 out_pixels(R, G, B, oute); \
415 R0 = vec_add(Y2, vx0); \
416 G0 = vec_add(Y2, uvx0); \
417 B0 = vec_add(Y2, ux0); \
418 R1 = vec_add(Y3, vx1); \
419 G1 = vec_add(Y3, uvx1); \
420 B1 = vec_add(Y3, ux1); \
421 R = vec_packclp(R0, R1); \
422 G = vec_packclp(G0, G1); \
423 B = vec_packclp(B0, B1); \
426 out_pixels(R, G, B, outo); \
434 ui += instrides_scl[1]; \
435 vi += instrides_scl[2]; \
436 y1i += instrides_scl[0]; \
437 y2i += instrides_scl[0]; \
442 #define out_abgr(a, b, c, ptr) \
443 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr)
444 #define out_bgra(a, b, c, ptr) \
445 vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr)
446 #define out_rgba(a, b, c, ptr) \
447 vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr)
448 #define out_argb(a, b, c, ptr) \
449 vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr)
450 #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr)
451 #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr)
453 DEFCSP420_CVT(yuv2_abgr, out_abgr)
454 DEFCSP420_CVT(yuv2_bgra, out_bgra)
455 DEFCSP420_CVT(yuv2_rgba, out_rgba)
456 DEFCSP420_CVT(yuv2_argb, out_argb)
457 DEFCSP420_CVT(yuv2_rgb24, out_rgb24)
458 DEFCSP420_CVT(yuv2_bgr24, out_bgr24)
460 // uyvy|uyvy|uyvy|uyvy
461 // 0123 4567 89ab cdef
462 static const vector unsigned char
463 demux_u = { 0x10, 0x00, 0x10, 0x00,
464 0x10, 0x04, 0x10, 0x04,
465 0x10, 0x08, 0x10, 0x08,
466 0x10, 0x0c, 0x10, 0x0c },
467 demux_v = { 0x10, 0x02, 0x10, 0x02,
468 0x10, 0x06, 0x10, 0x06,
469 0x10, 0x0A, 0x10, 0x0A,
470 0x10, 0x0E, 0x10, 0x0E },
471 demux_y = { 0x10, 0x01, 0x10, 0x03,
472 0x10, 0x05, 0x10, 0x07,
473 0x10, 0x09, 0x10, 0x0B,
474 0x10, 0x0D, 0x10, 0x0F };
477 * this is so I can play live CCIR raw video
479 static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in,
480 int *instrides, int srcSliceY, int srcSliceH,
481 unsigned char **oplanes, int *outstrides)
486 vector unsigned char uyvy;
487 vector signed short Y, U, V;
488 vector signed short R0, G0, B0, R1, G1, B1;
489 vector unsigned char R, G, B;
490 vector unsigned char *out;
494 out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]);
496 for (i = 0; i < h; i++)
497 for (j = 0; j < w / 16; j++) {
498 uyvy = vec_ld(0, img);
500 U = (vector signed short)
501 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
502 V = (vector signed short)
503 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
504 Y = (vector signed short)
505 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
507 cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0);
509 uyvy = vec_ld(16, img);
511 U = (vector signed short)
512 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u);
513 V = (vector signed short)
514 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v);
515 Y = (vector signed short)
516 vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y);
518 cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1);
520 R = vec_packclp(R0, R1);
521 G = vec_packclp(G0, G1);
522 B = vec_packclp(B0, B1);
524 // vec_mstbgr24 (R,G,B, out);
525 out_rgba(R, G, B, out);
532 #endif /* HAVE_ALTIVEC */
534 /* Ok currently the acceleration routine only supports
535 * inputs of widths a multiple of 16
536 * and heights a multiple 2
538 * So we just fall back to the C codes for this.
540 av_cold SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c)
543 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
547 * and this seems not to matter too much I tried a bunch of
548 * videos with abnormal widths and MPlayer crashes elsewhere.
549 * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
550 * boom with X11 bad match.
553 if ((c->srcW & 0xf) != 0)
556 switch (c->srcFormat) {
557 case AV_PIX_FMT_YUV410P:
558 case AV_PIX_FMT_YUV420P:
559 /*case IMGFMT_CLPL: ??? */
560 case AV_PIX_FMT_GRAY8:
561 case AV_PIX_FMT_NV12:
562 case AV_PIX_FMT_NV21:
563 if ((c->srcH & 0x1) != 0)
566 switch (c->dstFormat) {
567 case AV_PIX_FMT_RGB24:
568 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n");
569 return altivec_yuv2_rgb24;
570 case AV_PIX_FMT_BGR24:
571 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n");
572 return altivec_yuv2_bgr24;
573 case AV_PIX_FMT_ARGB:
574 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n");
575 return altivec_yuv2_argb;
576 case AV_PIX_FMT_ABGR:
577 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n");
578 return altivec_yuv2_abgr;
579 case AV_PIX_FMT_RGBA:
580 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n");
581 return altivec_yuv2_rgba;
582 case AV_PIX_FMT_BGRA:
583 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n");
584 return altivec_yuv2_bgra;
585 default: return NULL;
589 case AV_PIX_FMT_UYVY422:
590 switch (c->dstFormat) {
591 case AV_PIX_FMT_BGR32:
592 av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n");
593 return altivec_uyvy_rgb32;
594 default: return NULL;
598 #endif /* HAVE_ALTIVEC */
603 av_cold void ff_yuv2rgb_init_tables_ppc(SwsContext *c,
604 const int inv_table[4],
611 DECLARE_ALIGNED(16, signed short, tmp)[8];
612 vector signed short vec;
615 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
618 buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy
619 buf.tmp[1] = -256 * brightness; // oy
620 buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv
621 buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu
622 buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu
623 buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv
625 c->CSHIFT = (vector unsigned short) vec_splat_u16(2);
626 c->CY = vec_splat((vector signed short) buf.vec, 0);
627 c->OY = vec_splat((vector signed short) buf.vec, 1);
628 c->CRV = vec_splat((vector signed short) buf.vec, 2);
629 c->CBU = vec_splat((vector signed short) buf.vec, 3);
630 c->CGU = vec_splat((vector signed short) buf.vec, 4);
631 c->CGV = vec_splat((vector signed short) buf.vec, 5);
633 #endif /* HAVE_ALTIVEC */
638 static av_always_inline void yuv2packedX_altivec(SwsContext *c,
639 const int16_t *lumFilter,
640 const int16_t **lumSrc,
642 const int16_t *chrFilter,
643 const int16_t **chrUSrc,
644 const int16_t **chrVSrc,
646 const int16_t **alpSrc,
649 enum AVPixelFormat target)
652 vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V;
653 vector signed short R0, G0, B0, R1, G1, B1;
655 vector unsigned char R, G, B;
656 vector unsigned char *out, *nout;
658 vector signed short RND = vec_splat_s16(1 << 3);
659 vector unsigned short SCL = vec_splat_u16(4);
660 DECLARE_ALIGNED(16, unsigned int, scratch)[16];
662 vector signed short *YCoeffs, *CCoeffs;
664 YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize;
665 CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize;
667 out = (vector unsigned char *) dest;
669 for (i = 0; i < dstW; i += 16) {
672 /* extract 16 coeffs from lumSrc */
673 for (j = 0; j < lumFilterSize; j++) {
674 X0 = vec_ld(0, &lumSrc[j][i]);
675 X1 = vec_ld(16, &lumSrc[j][i]);
676 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
677 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
682 /* extract 8 coeffs from U,V */
683 for (j = 0; j < chrFilterSize; j++) {
684 X = vec_ld(0, &chrUSrc[j][i / 2]);
685 U = vec_mradds(X, CCoeffs[j], U);
686 X = vec_ld(0, &chrVSrc[j][i / 2]);
687 V = vec_mradds(X, CCoeffs[j], V);
690 /* scale and clip signals */
691 Y0 = vec_sra(Y0, SCL);
692 Y1 = vec_sra(Y1, SCL);
696 Y0 = vec_clip_s16(Y0);
697 Y1 = vec_clip_s16(Y1);
702 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
703 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
705 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
706 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
707 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
710 U0 = vec_mergeh(U, U);
711 V0 = vec_mergeh(V, V);
713 U1 = vec_mergel(U, U);
714 V1 = vec_mergel(V, V);
716 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
717 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
719 R = vec_packclp(R0, R1);
720 G = vec_packclp(G0, G1);
721 B = vec_packclp(B0, B1);
724 case AV_PIX_FMT_ABGR:
725 out_abgr(R, G, B, out);
727 case AV_PIX_FMT_BGRA:
728 out_bgra(R, G, B, out);
730 case AV_PIX_FMT_RGBA:
731 out_rgba(R, G, B, out);
733 case AV_PIX_FMT_ARGB:
734 out_argb(R, G, B, out);
736 case AV_PIX_FMT_RGB24:
737 out_rgb24(R, G, B, out);
739 case AV_PIX_FMT_BGR24:
740 out_bgr24(R, G, B, out);
744 /* If this is reached, the caller should have called yuv2packedXinC
746 static int printed_error_message;
747 if (!printed_error_message) {
748 av_log(c, AV_LOG_ERROR,
749 "altivec_yuv2packedX doesn't support %s output\n",
750 av_get_pix_fmt_name(c->dstFormat));
751 printed_error_message = 1;
763 /* extract 16 coeffs from lumSrc */
764 for (j = 0; j < lumFilterSize; j++) {
765 X0 = vec_ld(0, &lumSrc[j][i]);
766 X1 = vec_ld(16, &lumSrc[j][i]);
767 Y0 = vec_mradds(X0, YCoeffs[j], Y0);
768 Y1 = vec_mradds(X1, YCoeffs[j], Y1);
773 /* extract 8 coeffs from U,V */
774 for (j = 0; j < chrFilterSize; j++) {
775 X = vec_ld(0, &chrUSrc[j][i / 2]);
776 U = vec_mradds(X, CCoeffs[j], U);
777 X = vec_ld(0, &chrVSrc[j][i / 2]);
778 V = vec_mradds(X, CCoeffs[j], V);
781 /* scale and clip signals */
782 Y0 = vec_sra(Y0, SCL);
783 Y1 = vec_sra(Y1, SCL);
787 Y0 = vec_clip_s16(Y0);
788 Y1 = vec_clip_s16(Y1);
793 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
794 * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7
796 * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15
797 * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7
798 * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7
801 U0 = vec_mergeh(U, U);
802 V0 = vec_mergeh(V, V);
804 U1 = vec_mergel(U, U);
805 V1 = vec_mergel(V, V);
807 cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0);
808 cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1);
810 R = vec_packclp(R0, R1);
811 G = vec_packclp(G0, G1);
812 B = vec_packclp(B0, B1);
814 nout = (vector unsigned char *) scratch;
816 case AV_PIX_FMT_ABGR:
817 out_abgr(R, G, B, nout);
819 case AV_PIX_FMT_BGRA:
820 out_bgra(R, G, B, nout);
822 case AV_PIX_FMT_RGBA:
823 out_rgba(R, G, B, nout);
825 case AV_PIX_FMT_ARGB:
826 out_argb(R, G, B, nout);
828 case AV_PIX_FMT_RGB24:
829 out_rgb24(R, G, B, nout);
831 case AV_PIX_FMT_BGR24:
832 out_bgr24(R, G, B, nout);
835 /* Unreachable, I think. */
836 av_log(c, AV_LOG_ERROR,
837 "altivec_yuv2packedX doesn't support %s output\n",
838 av_get_pix_fmt_name(c->dstFormat));
842 memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4);
846 #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \
847 void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \
848 const int16_t *lumFilter, \
849 const int16_t **lumSrc, \
851 const int16_t *chrFilter, \
852 const int16_t **chrUSrc, \
853 const int16_t **chrVSrc, \
855 const int16_t **alpSrc, \
856 uint8_t *dest, int dstW, int dstY) \
858 yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \
859 chrFilter, chrUSrc, chrVSrc, \
860 chrFilterSize, alpSrc, \
861 dest, dstW, dstY, pixfmt); \
864 YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR);
865 YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA);
866 YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB);
867 YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA);
868 YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24);
869 YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);
871 #endif /* HAVE_ALTIVEC */