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