Parent Directory | Revision Log
Revision 1.6 - (view) (download)
1 : | edgomez | 1.3 | ;/***************************************************************************** |
2 : | edgomez | 1.2 | ; * |
3 : | edgomez | 1.3 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - RGB colorspace conversions - | ||
5 : | edgomez | 1.2 | ; * |
6 : | edgomez | 1.3 | ; * Copyright(C) 2002-2003 Michael Militzer <isibaar@xvid.org> |
7 : | ; * 2002-2003 Peter Ross <pross@xvid.org> | ||
8 : | edgomez | 1.2 | ; * |
9 : | edgomez | 1.3 | ; * This program is free software ; you can redistribute it and/or modify |
10 : | ; * it under the terms of the GNU General Public License as published by | ||
11 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
12 : | ; * (at your option) any later version. | ||
13 : | edgomez | 1.2 | ; * |
14 : | edgomez | 1.3 | ; * This program is distributed in the hope that it will be useful, |
15 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
16 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
17 : | ; * GNU General Public License for more details. | ||
18 : | edgomez | 1.2 | ; * |
19 : | edgomez | 1.3 | ; * You should have received a copy of the GNU General Public License |
20 : | ; * along with this program ; if not, write to the Free Software | ||
21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
22 : | edgomez | 1.2 | ; * |
23 : | edgomez | 1.3 | ; ****************************************************************************/ |
24 : | edgomez | 1.2 | |
25 : | edgomez | 1.3 | BITS 32 |
26 : | edgomez | 1.2 | |
27 : | edgomez | 1.3 | %macro cglobal 1 |
28 : | edgomez | 1.2 | %ifdef PREFIX |
29 : | edgomez | 1.5 | %ifdef MARK_FUNCS |
30 : | edgomez | 1.6 | global _%1:function %1.endfunc-%1 |
31 : | %define %1 _%1:function %1.endfunc-%1 | ||
32 : | edgomez | 1.5 | %else |
33 : | global _%1 | ||
34 : | %define %1 _%1 | ||
35 : | %endif | ||
36 : | edgomez | 1.2 | %else |
37 : | edgomez | 1.5 | %ifdef MARK_FUNCS |
38 : | edgomez | 1.6 | global %1:function %1.endfunc-%1 |
39 : | edgomez | 1.5 | %else |
40 : | global %1 | ||
41 : | %endif | ||
42 : | edgomez | 1.2 | %endif |
43 : | %endmacro | ||
44 : | |||
45 : | edgomez | 1.3 | ;============================================================================= |
46 : | ; Some constants | ||
47 : | ;============================================================================= | ||
48 : | edgomez | 1.2 | |
49 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
50 : | ; RGB->YV12 yuv constants | ||
51 : | ;----------------------------------------------------------------------------- | ||
52 : | edgomez | 1.2 | |
53 : | %define Y_R 0.257 | ||
54 : | %define Y_G 0.504 | ||
55 : | %define Y_B 0.098 | ||
56 : | %define Y_ADD 16 | ||
57 : | |||
58 : | %define U_R 0.148 | ||
59 : | %define U_G 0.291 | ||
60 : | %define U_B 0.439 | ||
61 : | %define U_ADD 128 | ||
62 : | |||
63 : | %define V_R 0.439 | ||
64 : | %define V_G 0.368 | ||
65 : | %define V_B 0.071 | ||
66 : | %define V_ADD 128 | ||
67 : | |||
68 : | edgomez | 1.3 | ; Scaling used during conversion |
69 : | %define SCALEBITS 6 | ||
70 : | |||
71 : | ;============================================================================= | ||
72 : | ; Read only data | ||
73 : | ;============================================================================= | ||
74 : | |||
75 : | %ifdef FORMAT_COFF | ||
76 : | edgomez | 1.4 | SECTION .rodata |
77 : | edgomez | 1.3 | %else |
78 : | edgomez | 1.4 | SECTION .rodata align=16 |
79 : | edgomez | 1.3 | %endif |
80 : | ALIGN 16 | ||
81 : | |||
82 : | ;----------------------------------------------------------------------------- | ||
83 : | edgomez | 1.2 | ; RGB->YV12 multiplication matrices |
84 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
85 : | ; FIX(Y_B) FIX(Y_G) FIX(Y_R) Ignored | ||
86 : | edgomez | 1.2 | |
87 : | edgomez | 1.3 | y_mul: dw 25, 129, 66, 0 |
88 : | u_mul: dw 112, -74, -38, 0 | ||
89 : | v_mul: dw -18, -94, 112, 0 | ||
90 : | edgomez | 1.2 | |
91 : | |||
92 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
93 : | ; YV12->RGB data | ||
94 : | ;----------------------------------------------------------------------------- | ||
95 : | edgomez | 1.2 | |
96 : | edgomez | 1.3 | Y_SUB: dw 16, 16, 16, 16 |
97 : | U_SUB: dw 128, 128, 128, 128 | ||
98 : | V_SUB: dw 128, 128, 128, 128 | ||
99 : | edgomez | 1.2 | |
100 : | edgomez | 1.3 | Y_MUL: dw 74, 74, 74, 74 |
101 : | edgomez | 1.2 | |
102 : | edgomez | 1.3 | UG_MUL: dw 25, 25, 25, 25 |
103 : | VG_MUL: dw 52, 52, 52, 52 | ||
104 : | edgomez | 1.2 | |
105 : | edgomez | 1.3 | UB_MUL: dw 129, 129, 129, 129 |
106 : | VR_MUL: dw 102, 102, 102, 102 | ||
107 : | edgomez | 1.2 | |
108 : | edgomez | 1.3 | BRIGHT: db 128, 128, 128, 128, 128, 128, 128, 128 |
109 : | edgomez | 1.2 | |
110 : | edgomez | 1.3 | ;============================================================================= |
111 : | ; Helper macros used with the colorspace_mmx.inc file | ||
112 : | ;============================================================================= | ||
113 : | edgomez | 1.2 | |
114 : | ;------------------------------------------------------------------------------ | ||
115 : | ; BGR_TO_YV12( BYTES ) | ||
116 : | ; | ||
117 : | ; BYTES 3=bgr(24bit), 4=bgra(32-bit) | ||
118 : | ; | ||
119 : | ; bytes=3/4, pixels = 2, vpixels=2 | ||
120 : | ;------------------------------------------------------------------------------ | ||
121 : | edgomez | 1.3 | |
122 : | edgomez | 1.2 | %macro BGR_TO_YV12_INIT 2 |
123 : | edgomez | 1.3 | movq mm7, [y_mul] |
124 : | edgomez | 1.2 | %endmacro |
125 : | |||
126 : | |||
127 : | %macro BGR_TO_YV12 2 | ||
128 : | edgomez | 1.3 | ; y_out |
129 : | pxor mm4, mm4 | ||
130 : | pxor mm5, mm5 | ||
131 : | movd mm0, [edi] ; x_ptr[0...] | ||
132 : | movd mm2, [edi+edx] ; x_ptr[x_stride...] | ||
133 : | punpcklbw mm0, mm4 ; [ |b |g |r ] | ||
134 : | punpcklbw mm2, mm5 ; [ |b |g |r ] | ||
135 : | movq mm6, mm0 ; = [ |b4|g4|r4] | ||
136 : | paddw mm6, mm2 ; +[ |b4|g4|r4] | ||
137 : | pmaddwd mm0, mm7 ; *= Y_MUL | ||
138 : | pmaddwd mm2, mm7 ; *= Y_MUL | ||
139 : | movq mm4, mm0 ; [r] | ||
140 : | movq mm5, mm2 ; [r] | ||
141 : | psrlq mm4, 32 ; +[g] | ||
142 : | psrlq mm5, 32 ; +[g] | ||
143 : | paddd mm0, mm4 ; +[b] | ||
144 : | paddd mm2, mm5 ; +[b] | ||
145 : | |||
146 : | pxor mm4, mm4 | ||
147 : | pxor mm5, mm5 | ||
148 : | movd mm1, [edi+%1] ; src[%1...] | ||
149 : | movd mm3, [edi+edx+%1] ; src[x_stride+%1...] | ||
150 : | punpcklbw mm1, mm4 ; [ |b |g |r ] | ||
151 : | punpcklbw mm3, mm5 ; [ |b |g |r ] | ||
152 : | paddw mm6, mm1 ; +[ |b4|g4|r4] | ||
153 : | paddw mm6, mm3 ; +[ |b4|g4|r4] | ||
154 : | pmaddwd mm1, mm7 ; *= Y_MUL | ||
155 : | pmaddwd mm3, mm7 ; *= Y_MUL | ||
156 : | movq mm4, mm1 ; [r] | ||
157 : | movq mm5, mm3 ; [r] | ||
158 : | psrlq mm4, 32 ; +[g] | ||
159 : | psrlq mm5, 32 ; +[g] | ||
160 : | paddd mm1, mm4 ; +[b] | ||
161 : | paddd mm3, mm5 ; +[b] | ||
162 : | |||
163 : | push edx | ||
164 : | |||
165 : | movd edx, mm0 | ||
166 : | shr edx, 8 | ||
167 : | add edx, Y_ADD | ||
168 : | mov [esi], dl ; y_ptr[0] | ||
169 : | |||
170 : | movd edx, mm1 | ||
171 : | shr edx, 8 | ||
172 : | add edx, Y_ADD | ||
173 : | mov [esi + 1], dl ; y_ptr[1] | ||
174 : | |||
175 : | movd edx, mm2 | ||
176 : | shr edx, 8 | ||
177 : | add edx, Y_ADD | ||
178 : | mov [esi + eax + 0], dl ; y_ptr[y_stride + 0] | ||
179 : | |||
180 : | movd edx, mm3 | ||
181 : | shr edx, 8 | ||
182 : | add edx, Y_ADD | ||
183 : | mov [esi + eax + 1], dl ; y_ptr[y_stride + 1] | ||
184 : | |||
185 : | ; u_ptr, v_ptr | ||
186 : | movq mm0, mm6 ; = [ |b4|g4|r4] | ||
187 : | pmaddwd mm6, [v_mul] ; *= V_MUL | ||
188 : | pmaddwd mm0, [u_mul] ; *= U_MUL | ||
189 : | movq mm1, mm0 | ||
190 : | movq mm2, mm6 | ||
191 : | psrlq mm1, 32 | ||
192 : | psrlq mm2, 32 | ||
193 : | paddd mm0, mm1 | ||
194 : | paddd mm2, mm6 | ||
195 : | |||
196 : | movd edx, mm0 | ||
197 : | shr edx, 10 | ||
198 : | add edx, U_ADD | ||
199 : | mov [ebx], dl | ||
200 : | |||
201 : | movd edx, mm2 | ||
202 : | shr edx, 10 | ||
203 : | add edx, V_ADD | ||
204 : | mov [ecx], dl | ||
205 : | edgomez | 1.2 | |
206 : | edgomez | 1.3 | pop edx |
207 : | edgomez | 1.2 | %endmacro |
208 : | |||
209 : | ;------------------------------------------------------------------------------ | ||
210 : | ; YV12_TO_BGR( BYTES ) | ||
211 : | ; | ||
212 : | ; BYTES 3=bgr(24-bit), 4=bgra(32-bit) | ||
213 : | ; | ||
214 : | ; bytes=3/4, pixels = 8, vpixels=2 | ||
215 : | ;------------------------------------------------------------------------------ | ||
216 : | edgomez | 1.3 | |
217 : | edgomez | 1.2 | %macro YV12_TO_BGR_INIT 2 |
218 : | edgomez | 1.3 | pxor mm7, mm7 ; clear mm7 |
219 : | edgomez | 1.2 | %endmacro |
220 : | |||
221 : | %macro YV12_TO_BGR 2 | ||
222 : | %define TEMP_Y1 esp | ||
223 : | %define TEMP_Y2 esp + 8 | ||
224 : | %define TEMP_G1 esp + 16 | ||
225 : | %define TEMP_G2 esp + 24 | ||
226 : | %define TEMP_B1 esp + 32 | ||
227 : | %define TEMP_B2 esp + 40 | ||
228 : | |||
229 : | edgomez | 1.3 | movd mm2, [ebx] ; u_ptr[0] |
230 : | movd mm3, [ecx] ; v_ptr[0] | ||
231 : | punpcklbw mm2, mm7 ; u3u2u1u0 -> mm2 | ||
232 : | punpcklbw mm3, mm7 ; v3v2v1v0 -> mm3 | ||
233 : | psubsw mm2, [U_SUB] ; U - 128 | ||
234 : | psubsw mm3, [V_SUB] ; V - 128 | ||
235 : | movq mm4, mm2 | ||
236 : | movq mm5, mm3 | ||
237 : | pmullw mm2, [UG_MUL] | ||
238 : | pmullw mm3, [VG_MUL] | ||
239 : | movq mm6, mm2 ; u3u2u1u0 -> mm6 | ||
240 : | punpckhwd mm2, mm2 ; u3u3u2u2 -> mm2 | ||
241 : | punpcklwd mm6, mm6 ; u1u1u0u0 -> mm6 | ||
242 : | pmullw mm4, [UB_MUL] ; B_ADD -> mm4 | ||
243 : | movq mm0, mm3 | ||
244 : | punpckhwd mm3, mm3 ; v3v3v2v2 -> mm2 | ||
245 : | punpcklwd mm0, mm0 ; v1v1v0v0 -> mm6 | ||
246 : | paddsw mm2, mm3 | ||
247 : | paddsw mm6, mm0 | ||
248 : | pmullw mm5, [VR_MUL] ; R_ADD -> mm5 | ||
249 : | movq mm0, [esi] ; y7y6y5y4y3y2y1y0 -> mm0 | ||
250 : | movq mm1, mm0 | ||
251 : | punpckhbw mm1, mm7 ; y7y6y5y4 -> mm1 | ||
252 : | punpcklbw mm0, mm7 ; y3y2y1y0 -> mm0 | ||
253 : | psubsw mm0, [Y_SUB] ; Y - Y_SUB | ||
254 : | psubsw mm1, [Y_SUB] ; Y - Y_SUB | ||
255 : | pmullw mm1, [Y_MUL] | ||
256 : | pmullw mm0, [Y_MUL] | ||
257 : | movq [TEMP_Y2], mm1 ; y7y6y5y4 -> mm3 | ||
258 : | movq [TEMP_Y1], mm0 ; y3y2y1y0 -> mm7 | ||
259 : | psubsw mm1, mm2 ; g7g6g5g4 -> mm1 | ||
260 : | psubsw mm0, mm6 ; g3g2g1g0 -> mm0 | ||
261 : | psraw mm1, SCALEBITS | ||
262 : | psraw mm0, SCALEBITS | ||
263 : | packuswb mm0, mm1 ;g7g6g5g4g3g2g1g0 -> mm0 | ||
264 : | movq [TEMP_G1], mm0 | ||
265 : | movq mm0, [esi+eax] ; y7y6y5y4y3y2y1y0 -> mm0 | ||
266 : | movq mm1, mm0 | ||
267 : | punpckhbw mm1, mm7 ; y7y6y5y4 -> mm1 | ||
268 : | punpcklbw mm0, mm7 ; y3y2y1y0 -> mm0 | ||
269 : | psubsw mm0, [Y_SUB] ; Y - Y_SUB | ||
270 : | psubsw mm1, [Y_SUB] ; Y - Y_SUB | ||
271 : | pmullw mm1, [Y_MUL] | ||
272 : | pmullw mm0, [Y_MUL] | ||
273 : | movq mm3, mm1 | ||
274 : | psubsw mm1, mm2 ; g7g6g5g4 -> mm1 | ||
275 : | movq mm2, mm0 | ||
276 : | psubsw mm0, mm6 ; g3g2g1g0 -> mm0 | ||
277 : | psraw mm1, SCALEBITS | ||
278 : | psraw mm0, SCALEBITS | ||
279 : | packuswb mm0, mm1 ; g7g6g5g4g3g2g1g0 -> mm0 | ||
280 : | movq [TEMP_G2], mm0 | ||
281 : | movq mm0, mm4 | ||
282 : | punpckhwd mm4, mm4 ; u3u3u2u2 -> mm2 | ||
283 : | punpcklwd mm0, mm0 ; u1u1u0u0 -> mm6 | ||
284 : | movq mm1, mm3 ; y7y6y5y4 -> mm1 | ||
285 : | paddsw mm3, mm4 ; b7b6b5b4 -> mm3 | ||
286 : | movq mm7, mm2 ; y3y2y1y0 -> mm7 | ||
287 : | paddsw mm2, mm0 ; b3b2b1b0 -> mm2 | ||
288 : | psraw mm3, SCALEBITS | ||
289 : | psraw mm2, SCALEBITS | ||
290 : | packuswb mm2, mm3 ; b7b6b5b4b3b2b1b0 -> mm2 | ||
291 : | movq [TEMP_B2], mm2 | ||
292 : | movq mm3, [TEMP_Y2] | ||
293 : | movq mm2, [TEMP_Y1] | ||
294 : | movq mm6, mm3 ; TEMP_Y2 -> mm6 | ||
295 : | paddsw mm3, mm4 ; b7b6b5b4 -> mm3 | ||
296 : | movq mm4, mm2 ; TEMP_Y1 -> mm4 | ||
297 : | paddsw mm2, mm0 ; b3b2b1b0 -> mm2 | ||
298 : | psraw mm3, SCALEBITS | ||
299 : | psraw mm2, SCALEBITS | ||
300 : | packuswb mm2, mm3 ; b7b6b5b4b3b2b1b0 -> mm2 | ||
301 : | movq [TEMP_B1], mm2 | ||
302 : | movq mm0, mm5 | ||
303 : | punpckhwd mm5, mm5 ; v3v3v2v2 -> mm5 | ||
304 : | punpcklwd mm0, mm0 ; v1v1v0v0 -> mm0 | ||
305 : | paddsw mm1, mm5 ; r7r6r5r4 -> mm1 | ||
306 : | paddsw mm7, mm0 ; r3r2r1r0 -> mm7 | ||
307 : | psraw mm1, SCALEBITS | ||
308 : | psraw mm7, SCALEBITS | ||
309 : | packuswb mm7, mm1 ; r7r6r5r4r3r2r1r0 -> mm7 (TEMP_R2) | ||
310 : | paddsw mm6, mm5 ; r7r6r5r4 -> mm6 | ||
311 : | paddsw mm4, mm0 ; r3r2r1r0 -> mm4 | ||
312 : | psraw mm6, SCALEBITS | ||
313 : | psraw mm4, SCALEBITS | ||
314 : | packuswb mm4, mm6 ; r7r6r5r4r3r2r1r0 -> mm4 (TEMP_R1) | ||
315 : | movq mm0, [TEMP_B1] | ||
316 : | movq mm1, [TEMP_G1] | ||
317 : | movq mm6, mm7 | ||
318 : | movq mm2, mm0 | ||
319 : | punpcklbw mm2, mm4 ; r3b3r2b2r1b1r0b0 -> mm2 | ||
320 : | punpckhbw mm0, mm4 ; r7b7r6b6r5b5r4b4 -> mm0 | ||
321 : | pxor mm7, mm7 | ||
322 : | movq mm3, mm1 | ||
323 : | punpcklbw mm1, mm7 ; 0g30g20g10g0 -> mm1 | ||
324 : | punpckhbw mm3, mm7 ; 0g70g60g50g4 -> mm3 | ||
325 : | movq mm4, mm2 | ||
326 : | punpcklbw mm2, mm1 ; 0r1g1b10r0g0b0 -> mm2 | ||
327 : | punpckhbw mm4, mm1 ; 0r3g3b30r2g2b2 -> mm4 | ||
328 : | movq mm5, mm0 | ||
329 : | punpcklbw mm0, mm3 ; 0r5g5b50r4g4b4 -> mm0 | ||
330 : | punpckhbw mm5, mm3 ; 0r7g7b70r6g6b6 -> mm5 | ||
331 : | %if %1 == 3 ; BGR (24-bit) | ||
332 : | movd [edi], mm2 | ||
333 : | psrlq mm2, 32 | ||
334 : | movd [edi + 3], mm2 | ||
335 : | movd [edi + 6], mm4 | ||
336 : | psrlq mm4, 32 | ||
337 : | movd [edi + 9], mm4 | ||
338 : | movd [edi + 12], mm0 | ||
339 : | psrlq mm0, 32 | ||
340 : | movd [edi + 15], mm0 | ||
341 : | movq mm2, mm5 | ||
342 : | psrlq mm0, 8 ; 000000r5g5 -> mm0 | ||
343 : | psllq mm2, 32 ; 0r6g6b60000 -> mm2 | ||
344 : | psrlq mm5, 32 ; 00000r7g7b7 -> mm5 | ||
345 : | psrlq mm2, 16 ; 000r6g6b600 -> mm2 | ||
346 : | por mm0, mm2 ; 000r6g6b6r5g5 -> mm0 | ||
347 : | psllq mm5, 40 ; r7g7b700000 -> mm5 | ||
348 : | por mm5, mm0 ; r7g7b7r6g6b6r5g5 -> mm5 | ||
349 : | movq [edi + 16], mm5 | ||
350 : | movq mm0, [TEMP_B2] | ||
351 : | movq mm1, [TEMP_G2] | ||
352 : | movq mm2, mm0 | ||
353 : | punpcklbw mm2, mm6 ; r3b3r2b2r1b1r0b0 -> mm2 | ||
354 : | punpckhbw mm0, mm6 ; r7b7r6b6r5b5r4b4 -> mm0 | ||
355 : | movq mm3, mm1 | ||
356 : | punpcklbw mm1, mm7 ; 0g30g20g10g0 -> mm1 | ||
357 : | punpckhbw mm3, mm7 ; 0g70g60g50g4 -> mm3 | ||
358 : | movq mm4, mm2 | ||
359 : | punpcklbw mm2, mm1 ; 0r1g1b10r0g0b0 -> mm2 | ||
360 : | punpckhbw mm4, mm1 ; 0r3g3b30r2g2b2 -> mm4 | ||
361 : | movq mm5, mm0 | ||
362 : | punpcklbw mm0, mm3 ; 0r5g5b50r4g4b4 -> mm0 | ||
363 : | punpckhbw mm5, mm3 ; 0r7g7b70r6g6b6 -> mm5 | ||
364 : | movd [edi+edx], mm2 | ||
365 : | psrlq mm2, 32 | ||
366 : | movd [edi+edx + 3], mm2 | ||
367 : | movd [edi+edx + 6], mm4 | ||
368 : | psrlq mm4, 32 | ||
369 : | movd [edi+edx + 9], mm4 | ||
370 : | movd [edi+edx + 12], mm0 | ||
371 : | psrlq mm0, 32 | ||
372 : | movd [edi+edx + 15], mm0 | ||
373 : | movq mm2, mm5 | ||
374 : | psrlq mm0, 8 ; 000000r5g5 -> mm0 | ||
375 : | psllq mm2, 32 ; 0r6g6b60000 -> mm2 | ||
376 : | psrlq mm5, 32 ; 00000r7g7b7 -> mm5 | ||
377 : | psrlq mm2, 16 ; 000r6g6b600 -> mm2 | ||
378 : | por mm0, mm2 ; 000r6g6b6r5g5 -> mm0 | ||
379 : | psllq mm5, 40 ; r7g7b700000 -> mm5 | ||
380 : | por mm5, mm0 ; r7g7b7r6g6b6r5g5 -> mm5 | ||
381 : | movq [edi + edx + 16], mm5 | ||
382 : | |||
383 : | %else ; BGRA (32-bit) | ||
384 : | movq [edi], mm2 | ||
385 : | movq [edi + 8], mm4 | ||
386 : | movq [edi + 16], mm0 | ||
387 : | movq [edi + 24], mm5 | ||
388 : | movq mm0, [TEMP_B2] | ||
389 : | movq mm1, [TEMP_G2] | ||
390 : | movq mm2, mm0 | ||
391 : | punpcklbw mm2, mm6 ; r3b3r2b2r1b1r0b0 -> mm2 | ||
392 : | punpckhbw mm0, mm6 ; r7b7r6b6r5b5r4b4 -> mm0 | ||
393 : | movq mm3, mm1 | ||
394 : | punpcklbw mm1, mm7 ; 0g30g20g10g0 -> mm1 | ||
395 : | punpckhbw mm3, mm7 ; 0g70g60g50g4 -> mm3 | ||
396 : | movq mm4, mm2 | ||
397 : | punpcklbw mm2, mm1 ; 0r1g1b10r0g0b0 -> mm2 | ||
398 : | punpckhbw mm4, mm1 ; 0r3g3b30r2g2b2 -> mm4 | ||
399 : | movq mm5, mm0 | ||
400 : | punpcklbw mm0, mm3 ; 0r5g5b50r4g4b4 -> mm0 | ||
401 : | punpckhbw mm5, mm3 ; 0r7g7b70r6g6b6 -> mm5 | ||
402 : | movq [edi + edx], mm2 | ||
403 : | movq [edi + edx + 8], mm4 | ||
404 : | movq [edi + edx + 16], mm0 | ||
405 : | movq [edi + edx + 24], mm5 | ||
406 : | edgomez | 1.2 | %endif |
407 : | |||
408 : | %undef TEMP_Y1 | ||
409 : | %undef TEMP_Y2 | ||
410 : | %undef TEMP_G1 | ||
411 : | %undef TEMP_G2 | ||
412 : | %undef TEMP_B1 | ||
413 : | %undef TEMP_B2 | ||
414 : | %endmacro | ||
415 : | |||
416 : | edgomez | 1.3 | ;============================================================================= |
417 : | ; Code | ||
418 : | ;============================================================================= | ||
419 : | edgomez | 1.2 | |
420 : | edgomez | 1.3 | SECTION .text |
421 : | edgomez | 1.2 | |
422 : | edgomez | 1.3 | %include "colorspace_mmx.inc" |
423 : | edgomez | 1.2 | |
424 : | ; input | ||
425 : | MAKE_COLORSPACE bgr_to_yv12_mmx,0, 3,2,2, BGR_TO_YV12, 3, -1 | ||
426 : | MAKE_COLORSPACE bgra_to_yv12_mmx,0, 4,2,2, BGR_TO_YV12, 4, -1 | ||
427 : | |||
428 : | ; output | ||
429 : | MAKE_COLORSPACE yv12_to_bgr_mmx,48, 3,8,2, YV12_TO_BGR, 3, -1 | ||
430 : | MAKE_COLORSPACE yv12_to_bgra_mmx,48, 4,8,2, YV12_TO_BGR, 4, -1 | ||
431 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |