3 * yuv2rgb_mmx.c, Software YUV to RGB coverter with Intel MMX "technology"
5 * Copyright (C) 2000, Silicon Integrated System Corp.
8 * Author: Olie Lho <ollie@sis.com.tw>
10 * This file is part of mpeg2dec, a free MPEG-2 video decoder
12 * mpeg2dec is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2, or (at your option)
17 * mpeg2dec is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with GNU Make; see the file COPYING. If not, write to
24 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 * 15,24 bpp and dithering from Michael Niedermayer (michaelni@gmx.at)
27 * MMX/MMX2 Template stuff from Michael Niedermayer (needed for fast movntq support)
35 /* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
42 #define MOVNTQ "movntq"
43 #define SFENCE "sfence"
50 /* Do the multiply part of the conversion for even and odd pixels,
52 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
53 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
54 mm6 -> Y even, mm7 -> Y odd */\
55 /* convert the chroma part */\
56 "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
57 "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
59 "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ \
60 "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ \
62 "psllw $3, %%mm0;" /* Promote precision */ \
63 "psllw $3, %%mm1;" /* Promote precision */ \
65 "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
66 "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
68 "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ \
69 "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ \
71 "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */\
72 "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */\
74 "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */\
76 /* convert the luma part */\
77 "psubusb mmx_10w, %%mm6;" /* Y -= 16 */\
79 "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */\
80 "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */\
82 "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */\
84 "psllw $3, %%mm6;" /* Promote precision */\
85 "psllw $3, %%mm7;" /* Promote precision */\
87 "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */\
88 "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */\
90 /* Do the addition part of the conversion for even and odd pixels,
92 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
93 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
94 mm6 -> Y even, mm7 -> Y odd */\
95 "movq %%mm0, %%mm3;" /* Copy Cblue */\
96 "movq %%mm1, %%mm4;" /* Copy Cred */\
97 "movq %%mm2, %%mm5;" /* Copy Cgreen */\
99 "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */\
100 "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */\
102 "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */\
103 "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */\
105 "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */\
106 "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */\
108 /* Limit RGB even to 0..255 */\
109 "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0 B6 B4 B2 B0 */\
110 "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0 R6 R4 R2 R0 */\
111 "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0 G6 G4 G2 G0 */\
113 /* Limit RGB odd to 0..255 */\
114 "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1 B7 B5 B3 B1 */\
115 "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1 R7 R5 R3 R1 */\
116 "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1 G7 G5 G3 G1 */\
118 /* Interleave RGB even and odd */\
119 "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */\
120 "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */\
121 "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */\
124 static inline void RENAME(yuv420_rgb16) (uint8_t * image, uint8_t * py,
125 uint8_t * pu, uint8_t * pv,
126 int h_size, int v_size,
127 int rgb_stride, int y_stride, int uv_stride)
132 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
134 for (y = v_size; --y >= 0; ) {
135 uint8_t *_image = image;
140 b5Dither= dither8[y&1];
141 g6Dither= dither4[y&1];
142 g5Dither= dither8[y&1];
143 r5Dither= dither8[(y+1)&1];
145 /* load data for start of next scan line */
146 __asm__ __volatile__ (
147 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
148 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
149 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
151 : : "r" (_py), "r" (_pu), "r" (_pv));
153 for (x = h_size >> 3; --x >= 0; ) {
154 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
155 pixels in each iteration */
157 __asm__ __volatile__ (
158 /* no speed diference on my p3@500 with prefetch,
159 * if it is faster for anyone with -benchmark then tell me
160 PREFETCH" 64(%0) \n\t"
161 PREFETCH" 64(%1) \n\t"
162 PREFETCH" 64(%2) \n\t"
167 "paddusb b5Dither, %%mm0;"
168 "paddusb g6Dither, %%mm2;"
169 "paddusb r5Dither, %%mm1;"
171 /* mask unneeded bits off */
172 "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
173 "pand mmx_grnmask, %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */
174 "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
176 "psrlw $3,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
177 "pxor %%mm4, %%mm4;" /* zero mm4 */
179 "movq %%mm0, %%mm5;" /* Copy B7-B0 */
180 "movq %%mm2, %%mm7;" /* Copy G7-G0 */
182 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
183 "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
184 "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
186 "psllw $3, %%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
187 "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
189 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
190 MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
192 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
193 "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
194 "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
196 "psllw $3, %%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
197 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
199 "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
200 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
202 MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
203 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
222 __asm__ __volatile__ (EMMS);
225 static inline void RENAME(yuv420_rgb15) (uint8_t * image, uint8_t * py,
226 uint8_t * pu, uint8_t * pv,
227 int h_size, int v_size,
228 int rgb_stride, int y_stride, int uv_stride)
233 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
235 for (y = v_size; --y >= 0; ) {
236 uint8_t *_image = image;
241 b5Dither= dither8[y&1];
242 g6Dither= dither4[y&1];
243 g5Dither= dither8[y&1];
244 r5Dither= dither8[(y+1)&1];
246 /* load data for start of next scan line */
247 __asm__ __volatile__ (
248 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
249 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
250 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
252 : : "r" (_py), "r" (_pu), "r" (_pv));
254 for (x = h_size >> 3; --x >= 0; ) {
255 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
256 pixels in each iteration */
258 __asm__ __volatile__ (
262 "paddusb b5Dither, %%mm0 \n\t"
263 "paddusb g5Dither, %%mm2 \n\t"
264 "paddusb r5Dither, %%mm1 \n\t"
267 /* mask unneeded bits off */
268 "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
269 "pand mmx_redmask, %%mm2;" /* g7g6g5g4 g3_0_0_0 g7g6g5g4 g3_0_0_0 */
270 "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
272 "psrlw $3,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
273 "psrlw $1,%%mm1;" /* 0_r7r6r5 r4r3_0_0 0_r7r6r5 r4r3_0_0 */
274 "pxor %%mm4, %%mm4;" /* zero mm4 */
276 "movq %%mm0, %%mm5;" /* Copy B7-B0 */
277 "movq %%mm2, %%mm7;" /* Copy G7-G0 */
279 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
280 "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3_0_0_0 */
281 "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
283 "psllw $2, %%mm2;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
284 "por %%mm2, %%mm0;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
286 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
287 MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
289 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
290 "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 0_g7g6g5 g4g3_0_0 */
291 "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
293 "psllw $2, %%mm7;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
294 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
296 "por %%mm7, %%mm5;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
297 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
299 MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
300 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
319 __asm__ __volatile__ (EMMS);
322 static inline void RENAME(yuv420_rgb24) (uint8_t * image, uint8_t * py,
323 uint8_t * pu, uint8_t * pv,
324 int h_size, int v_size,
325 int rgb_stride, int y_stride, int uv_stride)
330 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
332 for (y = v_size; --y >= 0; ) {
333 uint8_t *_image = image;
338 /* load data for start of next scan line */
339 __asm__ __volatile__ (
340 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
341 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
342 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
344 : : "r" (_py), "r" (_pu), "r" (_pv));
346 for (x = h_size >> 3; --x >= 0; ) {
347 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
348 pixels in each iteration */
350 __asm__ __volatile__ (
352 /* mm0=B, %%mm2=G, %%mm1=R */
354 "movq M24A, %%mm4 \n\t"
355 "movq M24C, %%mm7 \n\t"
356 "pshufw $0x50, %%mm0, %%mm5 \n\t" /* B3 B2 B3 B2 B1 B0 B1 B0 */
357 "pshufw $0x50, %%mm2, %%mm3 \n\t" /* G3 G2 G3 G2 G1 G0 G1 G0 */
358 "pshufw $0x00, %%mm1, %%mm6 \n\t" /* R1 R0 R1 R0 R1 R0 R1 R0 */
360 "pand %%mm4, %%mm5 \n\t" /* B2 B1 B0 */
361 "pand %%mm4, %%mm3 \n\t" /* G2 G1 G0 */
362 "pand %%mm7, %%mm6 \n\t" /* R1 R0 */
364 "psllq $8, %%mm3 \n\t" /* G2 G1 G0 */
365 "por %%mm5, %%mm6 \n\t"
366 "por %%mm3, %%mm6 \n\t"
367 MOVNTQ" %%mm6, (%3) \n\t"
369 "psrlq $8, %%mm2 \n\t" /* 00 G7 G6 G5 G4 G3 G2 G1 */
370 "pshufw $0xA5, %%mm0, %%mm5 \n\t" /* B5 B4 B5 B4 B3 B2 B3 B2 */
371 "pshufw $0x55, %%mm2, %%mm3 \n\t" /* G4 G3 G4 G3 G4 G3 G4 G3 */
372 "pshufw $0xA5, %%mm1, %%mm6 \n\t" /* R5 R4 R5 R4 R3 R2 R3 R2 */
374 "pand M24B, %%mm5 \n\t" /* B5 B4 B3 */
375 "pand %%mm7, %%mm3 \n\t" /* G4 G3 */
376 "pand %%mm4, %%mm6 \n\t" /* R4 R3 R2 */
378 "por %%mm5, %%mm3 \n\t" /* B5 G4 B4 G3 B3 */
379 "por %%mm3, %%mm6 \n\t"
380 MOVNTQ" %%mm6, 8(%3) \n\t"
382 "pshufw $0xFF, %%mm0, %%mm5 \n\t" /* B7 B6 B7 B6 B7 B6 B6 B7 */
383 "pshufw $0xFA, %%mm2, %%mm3 \n\t" /* 00 G7 00 G7 G6 G5 G6 G5 */
384 "pshufw $0xFA, %%mm1, %%mm6 \n\t" /* R7 R6 R7 R6 R5 R4 R5 R4 */
385 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
387 "pand %%mm7, %%mm5 \n\t" /* B7 B6 */
388 "pand %%mm4, %%mm3 \n\t" /* G7 G6 G5 */
389 "pand M24B, %%mm6 \n\t" /* R7 R6 R5 */
390 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
392 "por %%mm5, %%mm3 \n\t"
393 "por %%mm3, %%mm6 \n\t"
394 MOVNTQ" %%mm6, 16(%3) \n\t"
395 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
396 "pxor %%mm4, %%mm4 \n\t"
400 "pxor %%mm4, %%mm4 \n\t"
401 "movq %%mm0, %%mm5 \n\t" /* B */
402 "movq %%mm1, %%mm6 \n\t" /* R */
403 "punpcklbw %%mm2, %%mm0 \n\t" /* GBGBGBGB 0 */
404 "punpcklbw %%mm4, %%mm1 \n\t" /* 0R0R0R0R 0 */
405 "punpckhbw %%mm2, %%mm5 \n\t" /* GBGBGBGB 2 */
406 "punpckhbw %%mm4, %%mm6 \n\t" /* 0R0R0R0R 2 */
407 "movq %%mm0, %%mm7 \n\t" /* GBGBGBGB 0 */
408 "movq %%mm5, %%mm3 \n\t" /* GBGBGBGB 2 */
409 "punpcklwd %%mm1, %%mm7 \n\t" /* 0RGB0RGB 0 */
410 "punpckhwd %%mm1, %%mm0 \n\t" /* 0RGB0RGB 1 */
411 "punpcklwd %%mm6, %%mm5 \n\t" /* 0RGB0RGB 2 */
412 "punpckhwd %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */
414 "movq %%mm7, %%mm2 \n\t" /* 0RGB0RGB 0 */
415 "movq %%mm0, %%mm6 \n\t" /* 0RGB0RGB 1 */
416 "movq %%mm5, %%mm1 \n\t" /* 0RGB0RGB 2 */
417 "movq %%mm3, %%mm4 \n\t" /* 0RGB0RGB 3 */
419 "psllq $40, %%mm7 \n\t" /* RGB00000 0 */
420 "psllq $40, %%mm0 \n\t" /* RGB00000 1 */
421 "psllq $40, %%mm5 \n\t" /* RGB00000 2 */
422 "psllq $40, %%mm3 \n\t" /* RGB00000 3 */
424 "punpckhdq %%mm2, %%mm7 \n\t" /* 0RGBRGB0 0 */
425 "punpckhdq %%mm6, %%mm0 \n\t" /* 0RGBRGB0 1 */
426 "punpckhdq %%mm1, %%mm5 \n\t" /* 0RGBRGB0 2 */
427 "punpckhdq %%mm4, %%mm3 \n\t" /* 0RGBRGB0 3 */
429 "psrlq $8, %%mm7 \n\t" /* 00RGBRGB 0 */
430 "movq %%mm0, %%mm6 \n\t" /* 0RGBRGB0 1 */
431 "psllq $40, %%mm0 \n\t" /* GB000000 1 */
432 "por %%mm0, %%mm7 \n\t" /* GBRGBRGB 0 */
433 MOVNTQ" %%mm7, (%3) \n\t"
435 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
437 "psrlq $24, %%mm6 \n\t" /* 0000RGBR 1 */
438 "movq %%mm5, %%mm1 \n\t" /* 0RGBRGB0 2 */
439 "psllq $24, %%mm5 \n\t" /* BRGB0000 2 */
440 "por %%mm5, %%mm6 \n\t" /* BRGBRGBR 1 */
441 MOVNTQ" %%mm6, 8(%3) \n\t"
443 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
445 "psrlq $40, %%mm1 \n\t" /* 000000RG 2 */
446 "psllq $8, %%mm3 \n\t" /* RGBRGB00 3 */
447 "por %%mm3, %%mm1 \n\t" /* RGBRGBRG 2 */
448 MOVNTQ" %%mm1, 16(%3) \n\t"
450 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
451 "pxor %%mm4, %%mm4 \n\t"
454 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
473 __asm__ __volatile__ (EMMS);
477 static inline void RENAME(yuv420_argb32) (uint8_t * image, uint8_t * py,
478 uint8_t * pu, uint8_t * pv,
479 int h_size, int v_size,
480 int rgb_stride, int y_stride, int uv_stride)
485 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
487 for (y = v_size; --y >= 0; ) {
488 uint8_t *_image = image;
493 /* load data for start of next scan line */
496 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
497 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
498 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
499 : : "r" (_py), "r" (_pu), "r" (_pv)
502 for (x = h_size >> 3; --x >= 0; ) {
503 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
504 pixels in each iteration */
505 __asm__ __volatile__ (
507 /* convert RGB plane to RGB packed format,
508 mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
509 mm4 -> GB, mm5 -> AR pixel 4-7,
510 mm6 -> GB, mm7 -> AR pixel 0-3 */
511 "pxor %%mm3, %%mm3;" /* zero mm3 */
513 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
514 "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
516 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
517 "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
519 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
520 "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */
522 "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */
523 MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */
525 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
526 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
528 "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */
529 MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */
531 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
532 "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */
534 "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */
535 MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */
537 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
538 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
540 "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */
541 MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */
543 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
544 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
546 "pxor %%mm4, %%mm4;" /* zero mm4 */
547 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
549 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
568 __asm__ __volatile__ (EMMS);
571 yuv2rgb_fun RENAME(yuv2rgb_init) (int bpp, int mode)
573 if (bpp == 15 && mode == MODE_RGB) return RENAME(yuv420_rgb15);
574 if (bpp == 16 && mode == MODE_RGB) return RENAME(yuv420_rgb16);
575 if (bpp == 24 && mode == MODE_RGB) return RENAME(yuv420_rgb24);
576 if (bpp == 32 && mode == MODE_RGB) return RENAME(yuv420_argb32);
577 return NULL; // Fallback to C.