Mercurial > sdl-ios-xcode

comparison src/video/ps3/spulibs/yuv2rgb.c @ 3156:4ead4cef6b7b gsoc2009_ps3

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Rewritten Makefile (fixed bug) and cleanup for yuv2rgb.
author Martin Lowinski <martin@goldtopf.org>
date Fri, 14 Aug 2009 15:49:44 +0000
parents
children cce01ba54174
comparison
equal deleted inserted replaced
3155:abc49915ccb2 3156:4ead4cef6b7b
1 /*
2 * SDL - Simple DirectMedia Layer
3 * CELL BE Support for PS3 Framebuffer
4 * Copyright (C) 2008, 2009 International Business Machines Corporation
5 *
6 * This library is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU Lesser General Public License as published
8 * by the Free Software Foundation; either version 2.1 of the License, or
9 * (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301
19 * USA
20 *
21 * Martin Lowinski <lowinski [at] de [dot] ibm [ibm] com>
22 * Dirk Herrendoerfer <d.herrendoerfer [at] de [dot] ibm [dot] com>
23 * SPE code based on research by:
24 * Rene Becker
25 * Thimo Emmerich
26 */
27
28 #include "spu_common.h"
29
30 #include <spu_intrinsics.h>
31 #include <spu_mfcio.h>
32
33 // Debugging
34 #define DEBUG
35
36 // Test environment for /2 resolutions
37 #define TESTING
38
39 #ifdef DEBUG
40 #define deprintf(fmt, args... ) \
41 fprintf( stdout, fmt, ##args ); \
42 fflush( stdout );
43 #else
44 #define deprintf( fmt, args... )
45 #endif
46
47 struct yuv2rgb_parms_t parms_converter __attribute__((aligned(128)));
48
49 /* A maximum of 8 lines Y, therefore 4 lines V, 4 lines U are stored
50 * there might be the need to retrieve misaligned data, adjust
51 * incoming v and u plane to be able to handle this (add 128)
52 */
53 unsigned char y_plane[2][(MAX_HDTV_WIDTH + 128) * 4] __attribute__((aligned(128)));
54 unsigned char v_plane[2][(MAX_HDTV_WIDTH + 128) * 2] __attribute__((aligned(128)));
55 unsigned char u_plane[2][(MAX_HDTV_WIDTH + 128) * 2] __attribute__((aligned(128)));
56
57 /* A maximum of 4 lines BGRA are stored, 4 byte per pixel */
58 unsigned char bgra[4 * MAX_HDTV_WIDTH * 4] __attribute__((aligned(128)));
59
60 /* some vectors needed by the float to int conversion */
61 static const vector float vec_255 = { 255.0f, 255.0f, 255.0f, 255.0f };
62 static const vector float vec_0_1 = { 0.1f, 0.1f, 0.1f, 0.1f };
63
64 void yuv_to_rgb_w16();
65 void yuv_to_rgb_w32();
66
67 void yuv_to_rgb_w2_line(unsigned char* y_addr, unsigned char* v_addr, unsigned char* u_addr, unsigned char* bgra_addr, unsigned int width);
68 void yuv_to_rgb_w32_line(unsigned char* y_addr, unsigned char* v_addr, unsigned char* u_addr, unsigned char* bgra_addr_, unsigned int width);
69
70
71 int main(unsigned long long spe_id __attribute__((unused)), unsigned long long argp __attribute__ ((unused)))
72 {
73 deprintf("[SPU] yuv2rgb_spu is up... (on SPE #%llu)\n", spe_id);
74 uint32_t ea_mfc, mbox;
75 // send ready message
76 spu_write_out_mbox(SPU_READY);
77
78 while (1) {
79 /* Check mailbox */
80 mbox = spu_read_in_mbox();
81 deprintf("[SPU] Message is %u\n", mbox);
82 switch (mbox) {
83 case SPU_EXIT:
84 deprintf("[SPU] yuv2rgb_converter goes down...\n");
85 return 0;
86 case SPU_START:
87 break;
88 default:
89 deprintf("[SPU] Cannot handle message\n");
90 continue;
91 }
92
93 /* Tag Manager setup */
94 unsigned int tag_id;
95 tag_id = mfc_multi_tag_reserve(1);
96 if (tag_id == MFC_TAG_INVALID) {
97 deprintf("[SPU] Failed to reserve mfc tags on yuv2rgb_converter\n");
98 return 0;
99 }
100
101 /* DMA transfer for the input parameters */
102 ea_mfc = spu_read_in_mbox();
103 deprintf("[SPU] Message on yuv2rgb_converter is %u\n", ea_mfc);
104 spu_mfcdma32(&parms_converter, (unsigned int)ea_mfc, sizeof(struct yuv2rgb_parms_t), tag_id, MFC_GET_CMD);
105 DMA_WAIT_TAG(tag_id);
106
107 /* There are alignment issues that involve handling of special cases
108 * a width of 32 results in a width of 16 in the chrominance
109 * --> choose the proper handling to optimize the performance
110 */
111 deprintf("[SPU] Convert %ix%i from YUV to RGB\n", parms_converter.src_pixel_width, parms_converter.src_pixel_height);
112 if (!(parms_converter.src_pixel_width & 0x1f)) {
113 deprintf("[SPU] Using yuv_to_rgb_w16\n");
114 yuv_to_rgb_w16();
115 } else {
116 deprintf("[SPU] Using yuv_to_rgb_w32\n");
117 yuv_to_rgb_w32();
118 }
119
120 mfc_multi_tag_release(tag_id, 1);
121 deprintf("[SPU] yuv2rgb_spu... done!\n");
122 /* Send FIN message */
123 spu_write_out_mbox(SPU_FIN);
124 }
125
126 return 0;
127 }
128
129
130 /*
131 * float_to_char()
132 *
133 * converts a float to a character using saturated
134 * arithmetic
135 *
136 * @param s float for conversion
137 * @returns converted character
138 */
139 inline static unsigned char float_to_char(float s) {
140 vector float vec_s = spu_splats(s);
141 vector unsigned int select_1 = spu_cmpgt(vec_0_1, vec_s);
142 vec_s = spu_sel(vec_s, vec_0_1, select_1);
143
144 vector unsigned int select_2 = spu_cmpgt(vec_s, vec_255);
145 vec_s = spu_sel(vec_s, vec_255, select_2);
146 return (unsigned char) spu_extract(vec_s,0);
147 }
148
149
150 /*
151 * vfloat_to_vuint()
152 *
153 * converts a float vector to an unsinged int vector using saturated
154 * arithmetic
155 *
156 * @param vec_s float vector for conversion
157 * @returns converted unsigned int vector
158 */
159 inline static vector unsigned int vfloat_to_vuint(vector float vec_s) {
160 vector unsigned int select_1 = spu_cmpgt(vec_0_1, vec_s);
161 vec_s = spu_sel(vec_s, vec_0_1, select_1);
162
163 vector unsigned int select_2 = spu_cmpgt(vec_s, vec_255);
164 vec_s = spu_sel(vec_s, vec_255, select_2);
165 return spu_convtu(vec_s,0);
166 }
167
168
169 void yuv_to_rgb_w16() {
170 // Pixel dimensions of the picture
171 uint32_t width, height;
172
173 // Extract parameters
174 width = parms_converter.src_pixel_width;
175 height = parms_converter.src_pixel_height;
176
177 // Plane data management
178 // Y
179 unsigned char* ram_addr_y = parms_converter.y_plane;
180 // V
181 unsigned char* ram_addr_v = parms_converter.v_plane;
182 // U
183 unsigned char* ram_addr_u = parms_converter.u_plane;
184
185 // BGRA
186 unsigned char* ram_addr_bgra = parms_converter.dstBuffer;
187
188 // Strides
189 unsigned int stride_y = width;
190 unsigned int stride_vu = width>>1;
191
192 // Buffer management
193 unsigned int buf_idx = 0;
194 unsigned int size_4lines_y = stride_y<<2;
195 unsigned int size_2lines_y = stride_y<<1;
196 unsigned int size_2lines_vu = stride_vu<<1;
197
198 // 2*width*4byte_per_pixel
199 unsigned int size_2lines_bgra = width<<3;
200
201
202 // start double-buffered processing
203 // 4 lines y
204 spu_mfcdma32(y_plane[buf_idx], (unsigned int) ram_addr_y, size_4lines_y, RETR_BUF+buf_idx, MFC_GET_CMD);
205
206 // 2 lines v
207 spu_mfcdma32(v_plane[buf_idx], (unsigned int) ram_addr_v, size_2lines_vu, RETR_BUF+buf_idx, MFC_GET_CMD);
208
209 // 2 lines u
210 spu_mfcdma32(u_plane[buf_idx], (unsigned int) ram_addr_u, size_2lines_vu, RETR_BUF+buf_idx, MFC_GET_CMD);
211
212 // Wait for these transfers to be completed
213 DMA_WAIT_TAG((RETR_BUF + buf_idx));
214
215 unsigned int i;
216 for(i=0; i<(height>>2)-1; i++) {
217
218 buf_idx^=1;
219
220 // 4 lines y
221 spu_mfcdma32(y_plane[buf_idx], (unsigned int) ram_addr_y+size_4lines_y, size_4lines_y, RETR_BUF+buf_idx, MFC_GET_CMD);
222
223 // 2 lines v
224 spu_mfcdma32(v_plane[buf_idx], (unsigned int) ram_addr_v+size_2lines_vu, size_2lines_vu, RETR_BUF+buf_idx, MFC_GET_CMD);
225
226 // 2 lines u
227 spu_mfcdma32(u_plane[buf_idx], (unsigned int) ram_addr_u+size_2lines_vu, size_2lines_vu, RETR_BUF+buf_idx, MFC_GET_CMD);
228
229 DMA_WAIT_TAG((RETR_BUF + buf_idx));
230
231 buf_idx^=1;
232
233
234 // Convert YUV to BGRA, store it back (first two lines)
235 #ifndef TESTING
236 yuv_to_rgb_w16_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
237
238 // Next two lines
239 yuv_to_rgb_w16_line(y_plane[buf_idx] + size_2lines_y,
240 v_plane[buf_idx] + stride_vu,
241 u_plane[buf_idx] + stride_vu,
242 bgra + size_2lines_bgra,
243 width);
244 #else
245 yuv_to_rgb_w2_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
246
247 // Next two lines
248 yuv_to_rgb_w2_line(y_plane[buf_idx] + size_2lines_y,
249 v_plane[buf_idx] + stride_vu,
250 u_plane[buf_idx] + stride_vu,
251 bgra + size_2lines_bgra,
252 width);
253 #endif
254
255 // Wait for previous storing transfer to be completed
256 DMA_WAIT_TAG(STR_BUF);
257
258 // Store converted lines in two steps->max transfer size 16384
259 spu_mfcdma32(bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
260 ram_addr_bgra += size_2lines_bgra;
261 spu_mfcdma32(bgra+size_2lines_bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
262 ram_addr_bgra += size_2lines_bgra;
263
264 // Move 4 lines
265 ram_addr_y += size_4lines_y;
266 ram_addr_v += size_2lines_vu;
267 ram_addr_u += size_2lines_vu;
268
269 buf_idx^=1;
270 }
271
272 #ifndef TESTING
273 // Convert YUV to BGRA, store it back (first two lines)
274 yuv_to_rgb_w16_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
275
276 // Next two lines
277 yuv_to_rgb_w16_line(y_plane[buf_idx] + size_2lines_y,
278 v_plane[buf_idx] + stride_vu,
279 u_plane[buf_idx] + stride_vu,
280 bgra + size_2lines_bgra,
281 width);
282 #else
283 // Convert YUV to BGRA, store it back (first two lines)
284 yuv_to_rgb_w2_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
285
286 // Next two lines
287 yuv_to_rgb_w2_line(y_plane[buf_idx] + size_2lines_y,
288 v_plane[buf_idx] + stride_vu,
289 u_plane[buf_idx] + stride_vu,
290 bgra + size_2lines_bgra,
291 width);
292 #endif
293
294 // Wait for previous storing transfer to be completed
295 DMA_WAIT_TAG(STR_BUF);
296 spu_mfcdma32(bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
297 ram_addr_bgra += size_2lines_bgra;
298 spu_mfcdma32(bgra+size_2lines_bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
299
300 // wait for previous storing transfer to be completed
301 DMA_WAIT_TAG(STR_BUF);
302
303 }
304
305
306 void yuv_to_rgb_w32() {
307 // Pixel dimensions of the picture
308 uint32_t width, height;
309
310 // Extract parameters
311 width = parms_converter.src_pixel_width;
312 height = parms_converter.src_pixel_height;
313
314 // Plane data management
315 // Y
316 unsigned char* ram_addr_y = parms_converter.y_plane;
317 // V
318 unsigned char* ram_addr_v = parms_converter.v_plane;
319 // U
320 unsigned char* ram_addr_u = parms_converter.u_plane;
321
322 // BGRA
323 unsigned char* ram_addr_bgra = parms_converter.dstBuffer;
324
325 // Strides
326 unsigned int stride_y = width;
327 unsigned int stride_vu = width>>1;
328
329 // Buffer management
330 unsigned int buf_idx = 0;
331 unsigned int size_4lines_y = stride_y<<2;
332 unsigned int size_2lines_y = stride_y<<1;
333 unsigned int size_2lines_vu = stride_vu<<1;
334
335 // 2*width*4byte_per_pixel
336 unsigned int size_2lines_bgra = width<<3;
337
338 // start double-buffered processing
339 // 4 lines y
340 spu_mfcdma32(y_plane[buf_idx], (unsigned int) ram_addr_y, size_4lines_y, RETR_BUF + buf_idx, MFC_GET_CMD);
341 // 2 lines v
342 spu_mfcdma32(v_plane[buf_idx], (unsigned int) ram_addr_v, size_2lines_vu, RETR_BUF + buf_idx, MFC_GET_CMD);
343 // 2 lines u
344 spu_mfcdma32(u_plane[buf_idx], (unsigned int) ram_addr_u, size_2lines_vu, RETR_BUF + buf_idx, MFC_GET_CMD);
345
346 // Wait for these transfers to be completed
347 DMA_WAIT_TAG((RETR_BUF + buf_idx));
348
349 unsigned int i;
350 for(i=0; i < (height>>2)-1; i++) {
351 buf_idx^=1;
352 // 4 lines y
353 spu_mfcdma32(y_plane[buf_idx], (unsigned int) ram_addr_y+size_4lines_y, size_4lines_y, RETR_BUF + buf_idx, MFC_GET_CMD);
354 deprintf("4lines = %d\n", size_4lines_y);
355 // 2 lines v
356 spu_mfcdma32(v_plane[buf_idx], (unsigned int) ram_addr_v+size_2lines_vu, size_2lines_vu, RETR_BUF + buf_idx, MFC_GET_CMD);
357 deprintf("2lines = %d\n", size_2lines_vu);
358 // 2 lines u
359 spu_mfcdma32(u_plane[buf_idx], (unsigned int) ram_addr_u+size_2lines_vu, size_2lines_vu, RETR_BUF + buf_idx, MFC_GET_CMD);
360 deprintf("2lines = %d\n", size_2lines_vu);
361
362 DMA_WAIT_TAG((RETR_BUF + buf_idx));
363
364 buf_idx^=1;
365
366 // Convert YUV to BGRA, store it back (first two lines)
367 yuv_to_rgb_w32_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
368
369 // Next two lines
370 yuv_to_rgb_w32_line(y_plane[buf_idx] + size_2lines_y,
371 v_plane[buf_idx] + stride_vu,
372 u_plane[buf_idx] + stride_vu,
373 bgra + size_2lines_bgra,
374 width);
375
376 // Wait for previous storing transfer to be completed
377 DMA_WAIT_TAG(STR_BUF);
378
379 // Store converted lines in two steps->max transfer size 16384
380 spu_mfcdma32(bgra, (unsigned int)ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
381 ram_addr_bgra += size_2lines_bgra;
382 spu_mfcdma32(bgra + size_2lines_bgra, (unsigned int)ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
383 ram_addr_bgra += size_2lines_bgra;
384
385 // Move 4 lines
386 ram_addr_y += size_4lines_y;
387 ram_addr_v += size_2lines_vu;
388 ram_addr_u += size_2lines_vu;
389
390 buf_idx^=1;
391 }
392
393 // Convert YUV to BGRA, store it back (first two lines)
394 yuv_to_rgb_w32_line(y_plane[buf_idx], v_plane[buf_idx], u_plane[buf_idx], bgra, width);
395
396 // Next two lines
397 yuv_to_rgb_w32_line(y_plane[buf_idx] + size_2lines_y,
398 v_plane[buf_idx] + stride_vu,
399 u_plane[buf_idx] + stride_vu,
400 bgra + size_2lines_bgra,
401 width);
402
403 // Wait for previous storing transfer to be completed
404 DMA_WAIT_TAG(STR_BUF);
405 spu_mfcdma32(bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
406 ram_addr_bgra += size_2lines_bgra;
407 spu_mfcdma32(bgra + size_2lines_bgra, (unsigned int) ram_addr_bgra, size_2lines_bgra, STR_BUF, MFC_PUT_CMD);
408
409 // Wait for previous storing transfer to be completed
410 DMA_WAIT_TAG(STR_BUF);
411 }
412
413
414 /* Some vectors needed by the yuv 2 rgb conversion algorithm */
415 const vector float vec_minus_128 = { -128.0f, -128.0f, -128.0f, -128.0f };
416 const vector unsigned char vec_null = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
417 const vector unsigned char vec_char2int_first = { 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x13 };
418 const vector unsigned char vec_char2int_second = { 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x17 };
419 const vector unsigned char vec_char2int_third = { 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x1A, 0x00, 0x00, 0x00, 0x1B };
420 const vector unsigned char vec_char2int_fourth = { 0x00, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x1D, 0x00, 0x00, 0x00, 0x1E, 0x00, 0x00, 0x00, 0x1F };
421
422 const vector float vec_R_precalc_coeff = {1.403f, 1.403f, 1.403f, 1.403f};
423 const vector float vec_Gu_precalc_coeff = {-0.344f, -0.344f, -0.344f, -0.344f};
424 const vector float vec_Gv_precalc_coeff = {-0.714f, -0.714f, -0.714f, -0.714f};
425 const vector float vec_B_precalc_coeff = {1.773f, 1.773f, 1.773f, 1.773f};
426
427 const vector unsigned int vec_alpha = { 255 << 24, 255 << 24, 255 << 24, 255 << 24 };
428
429 const vector unsigned char vec_select_floats_upper = { 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x04, 0x05, 0x06, 0x07 };
430 const vector unsigned char vec_select_floats_lower = { 0x08, 0x09, 0x0A, 0x0B, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x0C, 0x0D, 0x0E, 0x0F };
431
432
433 #ifdef TESTING
434 /*
435 * yuv_to_rgb_w2()
436 *
437 * - converts x * 4 pixels from YUV to RGB
438 * - two lines of YUV are taken as input.
439 * - width has to be a multiple of 2 (= 4 pixel)
440 *
441 * @param y_addr address of the y plane (local store)
442 * @param v_addr address of the v plane (local store)
443 * @param u_addr address of the u plane (local store)
444 * @param bgra_addr_char address of the bgra output buffer (local store)
445 * @param width the width of a line in pixel
446 */
447 void yuv_to_rgb_w2_line(unsigned char* y_addr, unsigned char* v_addr, unsigned char* u_addr, unsigned char* bgra_addr_char, unsigned int width) {
448 // each pixel is stored as an integer
449 unsigned int* bgra_addr = (unsigned int*) bgra_addr_char;
450
451 unsigned int x;
452 // Go through each line in steps of 2, because every U and V value is connected to 4 pixels Y (YUV 4:2:0)
453 for(x = 0; x < width; x+=2) {
454 // Get the 4 Y, 1 U and 1 V values
455 const unsigned char Y_1 = *(y_addr + x);
456 const unsigned char Y_2 = *(y_addr + x + 1);
457 const unsigned char Y_3 = *(y_addr + x + width);
458 const unsigned char Y_4 = *(y_addr + x + width + 1);
459 const unsigned char U = *(u_addr + (x >> 1));
460 const unsigned char V = *(v_addr + (x >> 1));
461
462 // Start converting
463 float V_minus_128 = (float)((float)V - 128.0f);
464 float U_minus_128 = (float)((float)U - 128.0f);
465
466 float R_precalculate = 1.403f * V_minus_128;
467 float G_precalculate = -(0.344f * U_minus_128 + 0.714f * V_minus_128);
468 float B_precalculate = 1.773f * U_minus_128;
469
470 // Cast the results
471 const unsigned char R_1 = float_to_char((Y_1 + R_precalculate));
472 const unsigned char R_2 = float_to_char((Y_2 + R_precalculate));
473 const unsigned char R_3 = float_to_char((Y_3 + R_precalculate));
474 const unsigned char R_4 = float_to_char((Y_4 + R_precalculate));
475 const unsigned char G_1 = float_to_char((Y_1 + G_precalculate));
476 const unsigned char G_2 = float_to_char((Y_2 + G_precalculate));
477 const unsigned char G_3 = float_to_char((Y_3 + G_precalculate));
478 const unsigned char G_4 = float_to_char((Y_4 + G_precalculate));
479 const unsigned char B_1 = float_to_char((Y_1 + B_precalculate));
480 const unsigned char B_2 = float_to_char((Y_2 + B_precalculate));
481 const unsigned char B_3 = float_to_char((Y_3 + B_precalculate));
482 const unsigned char B_4 = float_to_char((Y_4 + B_precalculate));
483
484 // Write back
485 *(bgra_addr + x) = (B_1 << 0)| (G_1 << 8) | (R_1 << 16) | (255 << 24);
486 *(bgra_addr + x + 1) = (B_2 << 0)| (G_2 << 8) | (R_2 << 16) | (255 << 24);
487 *(bgra_addr + x + width) = (B_3 << 0)| (G_3 << 8) | (R_3 << 16) | (255 << 24);
488 *(bgra_addr + x + width + 1) = (B_4 << 0)| (G_4 << 8) | (R_4 << 16) | (255 << 24);
489 }
490 }
491 #endif
492
493
494 /*
495 * yuv_to_rgb_w32()
496 *
497 * processes to line of yuv-input, width has to be a multiple of 32
498 * two lines of yuv are taken as input
499 *
500 * @param y_addr address of the y plane in local store
501 * @param v_addr address of the v plane in local store
502 * @param u_addr address of the u plane in local store
503 * @param bgra_addr_ address of the bgra output buffer
504 * @param width the width in pixel
505 */
506 void yuv_to_rgb_w32_line(unsigned char* y_addr, unsigned char* v_addr, unsigned char* u_addr, unsigned char* bgra_addr_, unsigned int width) {
507 // each pixel is stored as an integer
508 unsigned int* bgra_addr = (unsigned int*) bgra_addr_;
509
510 unsigned int x;
511 for(x = 0; x < width; x+=32) {
512 // Gehe zweischrittig durch die zeile, da jeder u und v wert fuer 4 pixel(zwei hoch, zwei breit) gilt
513
514 const vector unsigned char vchar_Y_1 = *((vector unsigned char*)(y_addr + x));
515 const vector unsigned char vchar_Y_2 = *((vector unsigned char*)(y_addr + x + 16));
516 const vector unsigned char vchar_Y_3 = *((vector unsigned char*)(y_addr + x + width));
517 const vector unsigned char vchar_Y_4 = *((vector unsigned char*)(y_addr + x + width + 16));
518 const vector unsigned char vchar_U = *((vector unsigned char*)(u_addr + (x >> 1)));
519 const vector unsigned char vchar_V = *((vector unsigned char*)(v_addr + (x >> 1)));
520
521 const vector float vfloat_U_1 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_U, vec_char2int_first), 0),vec_minus_128);
522 const vector float vfloat_U_2 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_U, vec_char2int_second), 0),vec_minus_128);
523 const vector float vfloat_U_3 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_U, vec_char2int_third), 0),vec_minus_128);
524 const vector float vfloat_U_4 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_U, vec_char2int_fourth), 0),vec_minus_128);
525
526 const vector float vfloat_V_1 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_V, vec_char2int_first), 0),vec_minus_128);
527 const vector float vfloat_V_2 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_V, vec_char2int_second), 0),vec_minus_128);
528 const vector float vfloat_V_3 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_V, vec_char2int_third), 0),vec_minus_128);
529 const vector float vfloat_V_4 = spu_add(spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_V, vec_char2int_fourth), 0),vec_minus_128);
530
531 vector float Y_1 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_1, vec_char2int_first), 0);
532 vector float Y_2 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_1, vec_char2int_second), 0);
533 vector float Y_3 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_1, vec_char2int_third), 0);
534 vector float Y_4 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_1, vec_char2int_fourth), 0);
535 vector float Y_5 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_2, vec_char2int_first), 0);
536 vector float Y_6 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_2, vec_char2int_second), 0);
537 vector float Y_7 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_2, vec_char2int_third), 0);
538 vector float Y_8 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_2, vec_char2int_fourth), 0);
539 vector float Y_9 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_3, vec_char2int_first), 0);
540 vector float Y_10 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_3, vec_char2int_second), 0);
541 vector float Y_11 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_3, vec_char2int_third), 0);
542 vector float Y_12 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_3, vec_char2int_fourth), 0);
543 vector float Y_13 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_4, vec_char2int_first), 0);
544 vector float Y_14 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_4, vec_char2int_second), 0);
545 vector float Y_15 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_4, vec_char2int_third), 0);
546 vector float Y_16 = spu_convtf((vector unsigned int)spu_shuffle(vec_null, vchar_Y_4, vec_char2int_fourth), 0);
547
548 const vector float R1a_precalculate = spu_mul(vec_R_precalc_coeff, vfloat_V_1);
549 const vector float R2a_precalculate = spu_mul(vec_R_precalc_coeff, vfloat_V_2);
550 const vector float R3a_precalculate = spu_mul(vec_R_precalc_coeff, vfloat_V_3);
551 const vector float R4a_precalculate = spu_mul(vec_R_precalc_coeff, vfloat_V_4);
552
553 const vector float R1_precalculate = spu_shuffle(R1a_precalculate, R1a_precalculate, vec_select_floats_upper);
554 const vector float R2_precalculate = spu_shuffle(R1a_precalculate, R1a_precalculate, vec_select_floats_lower);
555 const vector float R3_precalculate = spu_shuffle(R2a_precalculate, R2a_precalculate, vec_select_floats_upper);
556 const vector float R4_precalculate = spu_shuffle(R2a_precalculate, R2a_precalculate, vec_select_floats_lower);
557 const vector float R5_precalculate = spu_shuffle(R3a_precalculate, R3a_precalculate, vec_select_floats_upper);
558 const vector float R6_precalculate = spu_shuffle(R3a_precalculate, R3a_precalculate, vec_select_floats_lower);
559 const vector float R7_precalculate = spu_shuffle(R4a_precalculate, R4a_precalculate, vec_select_floats_upper);
560 const vector float R8_precalculate = spu_shuffle(R4a_precalculate, R4a_precalculate, vec_select_floats_lower);
561
562
563 const vector float G1a_precalculate = spu_madd(vec_Gu_precalc_coeff, vfloat_U_1, spu_mul(vfloat_V_1, vec_Gv_precalc_coeff));
564 const vector float G2a_precalculate = spu_madd(vec_Gu_precalc_coeff, vfloat_U_2, spu_mul(vfloat_V_2, vec_Gv_precalc_coeff));
565 const vector float G3a_precalculate = spu_madd(vec_Gu_precalc_coeff, vfloat_U_3, spu_mul(vfloat_V_3, vec_Gv_precalc_coeff));
566 const vector float G4a_precalculate = spu_madd(vec_Gu_precalc_coeff, vfloat_U_4, spu_mul(vfloat_V_4, vec_Gv_precalc_coeff));
567
568 const vector float G1_precalculate = spu_shuffle(G1a_precalculate, G1a_precalculate, vec_select_floats_upper);
569 const vector float G2_precalculate = spu_shuffle(G1a_precalculate, G1a_precalculate, vec_select_floats_lower);
570 const vector float G3_precalculate = spu_shuffle(G2a_precalculate, G2a_precalculate, vec_select_floats_upper);
571 const vector float G4_precalculate = spu_shuffle(G2a_precalculate, G2a_precalculate, vec_select_floats_lower);
572 const vector float G5_precalculate = spu_shuffle(G3a_precalculate, G3a_precalculate, vec_select_floats_upper);
573 const vector float G6_precalculate = spu_shuffle(G3a_precalculate, G3a_precalculate, vec_select_floats_lower);
574 const vector float G7_precalculate = spu_shuffle(G4a_precalculate, G4a_precalculate, vec_select_floats_upper);
575 const vector float G8_precalculate = spu_shuffle(G4a_precalculate, G4a_precalculate, vec_select_floats_lower);
576
577
578 const vector float B1a_precalculate = spu_mul(vec_B_precalc_coeff, vfloat_U_1);
579 const vector float B2a_precalculate = spu_mul(vec_B_precalc_coeff, vfloat_U_2);
580 const vector float B3a_precalculate = spu_mul(vec_B_precalc_coeff, vfloat_U_3);
581 const vector float B4a_precalculate = spu_mul(vec_B_precalc_coeff, vfloat_U_4);
582
583 const vector float B1_precalculate = spu_shuffle(B1a_precalculate, B1a_precalculate, vec_select_floats_upper);
584 const vector float B2_precalculate = spu_shuffle(B1a_precalculate, B1a_precalculate, vec_select_floats_lower);
585 const vector float B3_precalculate = spu_shuffle(B2a_precalculate, B2a_precalculate, vec_select_floats_upper);
586 const vector float B4_precalculate = spu_shuffle(B2a_precalculate, B2a_precalculate, vec_select_floats_lower);
587 const vector float B5_precalculate = spu_shuffle(B3a_precalculate, B3a_precalculate, vec_select_floats_upper);
588 const vector float B6_precalculate = spu_shuffle(B3a_precalculate, B3a_precalculate, vec_select_floats_lower);
589 const vector float B7_precalculate = spu_shuffle(B4a_precalculate, B4a_precalculate, vec_select_floats_upper);
590 const vector float B8_precalculate = spu_shuffle(B4a_precalculate, B4a_precalculate, vec_select_floats_lower);
591
592
593 const vector unsigned int R_1 = vfloat_to_vuint(spu_add( Y_1, R1_precalculate));
594 const vector unsigned int R_2 = vfloat_to_vuint(spu_add( Y_2, R2_precalculate));
595 const vector unsigned int R_3 = vfloat_to_vuint(spu_add( Y_3, R3_precalculate));
596 const vector unsigned int R_4 = vfloat_to_vuint(spu_add( Y_4, R4_precalculate));
597 const vector unsigned int R_5 = vfloat_to_vuint(spu_add( Y_5, R5_precalculate));
598 const vector unsigned int R_6 = vfloat_to_vuint(spu_add( Y_6, R6_precalculate));
599 const vector unsigned int R_7 = vfloat_to_vuint(spu_add( Y_7, R7_precalculate));
600 const vector unsigned int R_8 = vfloat_to_vuint(spu_add( Y_8, R8_precalculate));
601 const vector unsigned int R_9 = vfloat_to_vuint(spu_add( Y_9, R1_precalculate));
602 const vector unsigned int R_10 = vfloat_to_vuint(spu_add(Y_10, R2_precalculate));
603 const vector unsigned int R_11 = vfloat_to_vuint(spu_add(Y_11, R3_precalculate));
604 const vector unsigned int R_12 = vfloat_to_vuint(spu_add(Y_12, R4_precalculate));
605 const vector unsigned int R_13 = vfloat_to_vuint(spu_add(Y_13, R5_precalculate));
606 const vector unsigned int R_14 = vfloat_to_vuint(spu_add(Y_14, R6_precalculate));
607 const vector unsigned int R_15 = vfloat_to_vuint(spu_add(Y_15, R7_precalculate));
608 const vector unsigned int R_16 = vfloat_to_vuint(spu_add(Y_16, R8_precalculate));
609
610 const vector unsigned int G_1 = vfloat_to_vuint(spu_add( Y_1, G1_precalculate));
611 const vector unsigned int G_2 = vfloat_to_vuint(spu_add( Y_2, G2_precalculate));
612 const vector unsigned int G_3 = vfloat_to_vuint(spu_add( Y_3, G3_precalculate));
613 const vector unsigned int G_4 = vfloat_to_vuint(spu_add( Y_4, G4_precalculate));
614 const vector unsigned int G_5 = vfloat_to_vuint(spu_add( Y_5, G5_precalculate));
615 const vector unsigned int G_6 = vfloat_to_vuint(spu_add( Y_6, G6_precalculate));
616 const vector unsigned int G_7 = vfloat_to_vuint(spu_add( Y_7, G7_precalculate));
617 const vector unsigned int G_8 = vfloat_to_vuint(spu_add( Y_8, G8_precalculate));
618 const vector unsigned int G_9 = vfloat_to_vuint(spu_add( Y_9, G1_precalculate));
619 const vector unsigned int G_10 = vfloat_to_vuint(spu_add(Y_10, G2_precalculate));
620 const vector unsigned int G_11 = vfloat_to_vuint(spu_add(Y_11, G3_precalculate));
621 const vector unsigned int G_12 = vfloat_to_vuint(spu_add(Y_12, G4_precalculate));
622 const vector unsigned int G_13 = vfloat_to_vuint(spu_add(Y_13, G5_precalculate));
623 const vector unsigned int G_14 = vfloat_to_vuint(spu_add(Y_14, G6_precalculate));
624 const vector unsigned int G_15 = vfloat_to_vuint(spu_add(Y_15, G7_precalculate));
625 const vector unsigned int G_16 = vfloat_to_vuint(spu_add(Y_16, G8_precalculate));
626
627 const vector unsigned int B_1 = vfloat_to_vuint(spu_add( Y_1, B1_precalculate));
628 const vector unsigned int B_2 = vfloat_to_vuint(spu_add( Y_2, B2_precalculate));
629 const vector unsigned int B_3 = vfloat_to_vuint(spu_add( Y_3, B3_precalculate));
630 const vector unsigned int B_4 = vfloat_to_vuint(spu_add( Y_4, B4_precalculate));
631 const vector unsigned int B_5 = vfloat_to_vuint(spu_add( Y_5, B5_precalculate));
632 const vector unsigned int B_6 = vfloat_to_vuint(spu_add( Y_6, B6_precalculate));
633 const vector unsigned int B_7 = vfloat_to_vuint(spu_add( Y_7, B7_precalculate));
634 const vector unsigned int B_8 = vfloat_to_vuint(spu_add( Y_8, B8_precalculate));
635 const vector unsigned int B_9 = vfloat_to_vuint(spu_add( Y_9, B1_precalculate));
636 const vector unsigned int B_10 = vfloat_to_vuint(spu_add(Y_10, B2_precalculate));
637 const vector unsigned int B_11 = vfloat_to_vuint(spu_add(Y_11, B3_precalculate));
638 const vector unsigned int B_12 = vfloat_to_vuint(spu_add(Y_12, B4_precalculate));
639 const vector unsigned int B_13 = vfloat_to_vuint(spu_add(Y_13, B5_precalculate));
640 const vector unsigned int B_14 = vfloat_to_vuint(spu_add(Y_14, B6_precalculate));
641 const vector unsigned int B_15 = vfloat_to_vuint(spu_add(Y_15, B7_precalculate));
642 const vector unsigned int B_16 = vfloat_to_vuint(spu_add(Y_16, B8_precalculate));
643
644 *((vector unsigned int*)(bgra_addr + x)) = spu_or(spu_or(vec_alpha, B_1), spu_or(spu_slqwbyte( R_1, 2),spu_slqwbyte(G_1, 1)));
645 *((vector unsigned int*)(bgra_addr + x + 4)) = spu_or(spu_or(vec_alpha, B_2), spu_or(spu_slqwbyte( R_2, 2),spu_slqwbyte(G_2, 1)));
646 *((vector unsigned int*)(bgra_addr + x + 8)) = spu_or(spu_or(vec_alpha, B_3), spu_or(spu_slqwbyte( R_3, 2),spu_slqwbyte(G_3, 1)));
647 *((vector unsigned int*)(bgra_addr + x + 12)) = spu_or(spu_or(vec_alpha, B_4), spu_or(spu_slqwbyte( R_4, 2),spu_slqwbyte(G_4, 1)));
648 *((vector unsigned int*)(bgra_addr + x + 16)) = spu_or(spu_or(vec_alpha, B_5), spu_or(spu_slqwbyte( R_5, 2),spu_slqwbyte(G_5, 1)));
649 *((vector unsigned int*)(bgra_addr + x + 20)) = spu_or(spu_or(vec_alpha, B_6), spu_or(spu_slqwbyte( R_6, 2),spu_slqwbyte(G_6, 1)));
650 *((vector unsigned int*)(bgra_addr + x + 24)) = spu_or(spu_or(vec_alpha, B_7), spu_or(spu_slqwbyte( R_7, 2),spu_slqwbyte(G_7, 1)));
651 *((vector unsigned int*)(bgra_addr + x + 28)) = spu_or(spu_or(vec_alpha, B_8), spu_or(spu_slqwbyte( R_8, 2),spu_slqwbyte(G_8, 1)));
652 *((vector unsigned int*)(bgra_addr + x + width)) = spu_or(spu_or(vec_alpha, B_9), spu_or(spu_slqwbyte( R_9, 2),spu_slqwbyte(G_9, 1)));
653 *((vector unsigned int*)(bgra_addr + x + width + 4)) = spu_or(spu_or(vec_alpha, B_10), spu_or(spu_slqwbyte(R_10, 2),spu_slqwbyte(G_10, 1)));
654 *((vector unsigned int*)(bgra_addr + x + width + 8)) = spu_or(spu_or(vec_alpha, B_11), spu_or(spu_slqwbyte(R_11, 2),spu_slqwbyte(G_11, 1)));
655 *((vector unsigned int*)(bgra_addr + x + width + 12)) = spu_or(spu_or(vec_alpha, B_12), spu_or(spu_slqwbyte(R_12, 2),spu_slqwbyte(G_12, 1)));
656 *((vector unsigned int*)(bgra_addr + x + width + 16)) = spu_or(spu_or(vec_alpha, B_13), spu_or(spu_slqwbyte(R_13, 2),spu_slqwbyte(G_13, 1)));
657 *((vector unsigned int*)(bgra_addr + x + width + 20)) = spu_or(spu_or(vec_alpha, B_14), spu_or(spu_slqwbyte(R_14, 2),spu_slqwbyte(G_14, 1)));
658 *((vector unsigned int*)(bgra_addr + x + width + 24)) = spu_or(spu_or(vec_alpha, B_15), spu_or(spu_slqwbyte(R_15, 2),spu_slqwbyte(G_15, 1)));
659 *((vector unsigned int*)(bgra_addr + x + width + 28)) = spu_or(spu_or(vec_alpha, B_16), spu_or(spu_slqwbyte(R_16, 2),spu_slqwbyte(G_16, 1)));
660 }
661 }
662