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 "MANGLE(mmx_80w)", %%mm0;" /* Cb -= 128 */ \
60 "psubsw "MANGLE(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 "MANGLE(mmx_U_green)", %%mm2;" /* Mul Cb with green coeff -> Cb green */ \
69 "pmulhw "MANGLE(mmx_V_green)", %%mm3;" /* Mul Cr with green coeff -> Cr green */ \
71 "pmulhw "MANGLE(mmx_U_blue)", %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */\
72 "pmulhw "MANGLE(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 "MANGLE(mmx_10w)", %%mm6;" /* Y -= 16 */\
79 "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */\
80 "pand "MANGLE(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 "MANGLE(mmx_Y_coeff)", %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */\
88 "pmulhw "MANGLE(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;
139 int internal_h_size= h_size;
140 int aligned_h_size= (h_size+7)&~7;
142 if(rgb_stride >= aligned_h_size*2) internal_h_size= aligned_h_size;
144 b5Dither= dither8[y&1];
145 g6Dither= dither4[y&1];
146 g5Dither= dither8[y&1];
147 r5Dither= dither8[(y+1)&1];
149 /* load data for start of next scan line */
150 __asm__ __volatile__ (
151 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
152 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
153 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
155 : : "r" (_py), "r" (_pu), "r" (_pv));
157 for (x = internal_h_size >> 3; --x >= 0; ) {
158 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
159 pixels in each iteration */
161 __asm__ __volatile__ (
162 /* no speed diference on my p3@500 with prefetch,
163 * if it is faster for anyone with -benchmark then tell me
164 PREFETCH" 64(%0) \n\t"
165 PREFETCH" 64(%1) \n\t"
166 PREFETCH" 64(%2) \n\t"
171 "paddusb "MANGLE(b5Dither)", %%mm0;"
172 "paddusb "MANGLE(g6Dither)", %%mm2;"
173 "paddusb "MANGLE(r5Dither)", %%mm1;"
175 /* mask unneeded bits off */
176 "pand "MANGLE(mmx_redmask)", %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
177 "pand "MANGLE(mmx_grnmask)", %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */
178 "pand "MANGLE(mmx_redmask)", %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
180 "psrlw $3,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
181 "pxor %%mm4, %%mm4;" /* zero mm4 */
183 "movq %%mm0, %%mm5;" /* Copy B7-B0 */
184 "movq %%mm2, %%mm7;" /* Copy G7-G0 */
186 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
187 "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
188 "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
190 "psllw $3, %%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
191 "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
193 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
194 MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
196 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
197 "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
198 "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
200 "psllw $3, %%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
201 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
203 "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
204 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
206 MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
207 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
226 __asm__ __volatile__ (EMMS);
229 static inline void RENAME(yuv420_rgb15) (uint8_t * image, uint8_t * py,
230 uint8_t * pu, uint8_t * pv,
231 int h_size, int v_size,
232 int rgb_stride, int y_stride, int uv_stride)
237 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
239 for (y = v_size; --y >= 0; ) {
240 uint8_t *_image = image;
244 int internal_h_size= h_size;
245 int aligned_h_size= (h_size+7)&~7;
247 if(rgb_stride >= aligned_h_size*2) internal_h_size= aligned_h_size;
249 b5Dither= dither8[y&1];
250 g6Dither= dither4[y&1];
251 g5Dither= dither8[y&1];
252 r5Dither= dither8[(y+1)&1];
254 /* load data for start of next scan line */
255 __asm__ __volatile__ (
256 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
257 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
258 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
260 : : "r" (_py), "r" (_pu), "r" (_pv));
262 for (x = internal_h_size >> 3; --x >= 0; ) {
263 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
264 pixels in each iteration */
266 __asm__ __volatile__ (
270 "paddusb "MANGLE(b5Dither)", %%mm0 \n\t"
271 "paddusb "MANGLE(g5Dither)", %%mm2 \n\t"
272 "paddusb "MANGLE(r5Dither)", %%mm1 \n\t"
275 /* mask unneeded bits off */
276 "pand "MANGLE(mmx_redmask)", %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
277 "pand "MANGLE(mmx_redmask)", %%mm2;" /* g7g6g5g4 g3_0_0_0 g7g6g5g4 g3_0_0_0 */
278 "pand "MANGLE(mmx_redmask)", %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
280 "psrlw $3,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
281 "psrlw $1,%%mm1;" /* 0_r7r6r5 r4r3_0_0 0_r7r6r5 r4r3_0_0 */
282 "pxor %%mm4, %%mm4;" /* zero mm4 */
284 "movq %%mm0, %%mm5;" /* Copy B7-B0 */
285 "movq %%mm2, %%mm7;" /* Copy G7-G0 */
287 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
288 "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3_0_0_0 */
289 "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
291 "psllw $2, %%mm2;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
292 "por %%mm2, %%mm0;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
294 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
295 MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
297 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
298 "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 0_g7g6g5 g4g3_0_0 */
299 "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
301 "psllw $2, %%mm7;" /* 0_0_0_0 0_0_g7g6 g5g4g3_0 0_0_0_0 */
302 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
304 "por %%mm7, %%mm5;" /* 0_r7r6r5 r4r3g7g6 g5g4g3b7 b6b5b4b3 */
305 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
307 MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
308 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
327 __asm__ __volatile__ (EMMS);
330 static inline void RENAME(yuv420_rgb24) (uint8_t * image, uint8_t * py,
331 uint8_t * pu, uint8_t * pv,
332 int h_size, int v_size,
333 int rgb_stride, int y_stride, int uv_stride)
338 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
340 for (y = v_size; --y >= 0; ) {
341 uint8_t *_image = image;
345 int internal_h_size= h_size;
346 int aligned_h_size= (h_size+7)&~7;
348 if(rgb_stride >= aligned_h_size*3) internal_h_size= aligned_h_size;
350 /* load data for start of next scan line */
351 __asm__ __volatile__ (
352 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
353 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
354 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
356 : : "r" (_py), "r" (_pu), "r" (_pv));
358 for (x = internal_h_size >> 3; --x >= 0; ) {
359 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
360 pixels in each iteration */
362 __asm__ __volatile__ (
364 /* mm0=B, %%mm2=G, %%mm1=R */
366 "movq "MANGLE(M24A)", %%mm4 \n\t"
367 "movq "MANGLE(M24C)", %%mm7 \n\t"
368 "pshufw $0x50, %%mm0, %%mm5 \n\t" /* B3 B2 B3 B2 B1 B0 B1 B0 */
369 "pshufw $0x50, %%mm2, %%mm3 \n\t" /* G3 G2 G3 G2 G1 G0 G1 G0 */
370 "pshufw $0x00, %%mm1, %%mm6 \n\t" /* R1 R0 R1 R0 R1 R0 R1 R0 */
372 "pand %%mm4, %%mm5 \n\t" /* B2 B1 B0 */
373 "pand %%mm4, %%mm3 \n\t" /* G2 G1 G0 */
374 "pand %%mm7, %%mm6 \n\t" /* R1 R0 */
376 "psllq $8, %%mm3 \n\t" /* G2 G1 G0 */
377 "por %%mm5, %%mm6 \n\t"
378 "por %%mm3, %%mm6 \n\t"
379 MOVNTQ" %%mm6, (%3) \n\t"
381 "psrlq $8, %%mm2 \n\t" /* 00 G7 G6 G5 G4 G3 G2 G1 */
382 "pshufw $0xA5, %%mm0, %%mm5 \n\t" /* B5 B4 B5 B4 B3 B2 B3 B2 */
383 "pshufw $0x55, %%mm2, %%mm3 \n\t" /* G4 G3 G4 G3 G4 G3 G4 G3 */
384 "pshufw $0xA5, %%mm1, %%mm6 \n\t" /* R5 R4 R5 R4 R3 R2 R3 R2 */
386 "pand "MANGLE(M24B)", %%mm5 \n\t" /* B5 B4 B3 */
387 "pand %%mm7, %%mm3 \n\t" /* G4 G3 */
388 "pand %%mm4, %%mm6 \n\t" /* R4 R3 R2 */
390 "por %%mm5, %%mm3 \n\t" /* B5 G4 B4 G3 B3 */
391 "por %%mm3, %%mm6 \n\t"
392 MOVNTQ" %%mm6, 8(%3) \n\t"
394 "pshufw $0xFF, %%mm0, %%mm5 \n\t" /* B7 B6 B7 B6 B7 B6 B6 B7 */
395 "pshufw $0xFA, %%mm2, %%mm3 \n\t" /* 00 G7 00 G7 G6 G5 G6 G5 */
396 "pshufw $0xFA, %%mm1, %%mm6 \n\t" /* R7 R6 R7 R6 R5 R4 R5 R4 */
397 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
399 "pand %%mm7, %%mm5 \n\t" /* B7 B6 */
400 "pand %%mm4, %%mm3 \n\t" /* G7 G6 G5 */
401 "pand "MANGLE(M24B)", %%mm6 \n\t" /* R7 R6 R5 */
402 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
404 "por %%mm5, %%mm3 \n\t"
405 "por %%mm3, %%mm6 \n\t"
406 MOVNTQ" %%mm6, 16(%3) \n\t"
407 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
408 "pxor %%mm4, %%mm4 \n\t"
412 "pxor %%mm4, %%mm4 \n\t"
413 "movq %%mm0, %%mm5 \n\t" /* B */
414 "movq %%mm1, %%mm6 \n\t" /* R */
415 "punpcklbw %%mm2, %%mm0 \n\t" /* GBGBGBGB 0 */
416 "punpcklbw %%mm4, %%mm1 \n\t" /* 0R0R0R0R 0 */
417 "punpckhbw %%mm2, %%mm5 \n\t" /* GBGBGBGB 2 */
418 "punpckhbw %%mm4, %%mm6 \n\t" /* 0R0R0R0R 2 */
419 "movq %%mm0, %%mm7 \n\t" /* GBGBGBGB 0 */
420 "movq %%mm5, %%mm3 \n\t" /* GBGBGBGB 2 */
421 "punpcklwd %%mm1, %%mm7 \n\t" /* 0RGB0RGB 0 */
422 "punpckhwd %%mm1, %%mm0 \n\t" /* 0RGB0RGB 1 */
423 "punpcklwd %%mm6, %%mm5 \n\t" /* 0RGB0RGB 2 */
424 "punpckhwd %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */
426 "movq %%mm7, %%mm2 \n\t" /* 0RGB0RGB 0 */
427 "movq %%mm0, %%mm6 \n\t" /* 0RGB0RGB 1 */
428 "movq %%mm5, %%mm1 \n\t" /* 0RGB0RGB 2 */
429 "movq %%mm3, %%mm4 \n\t" /* 0RGB0RGB 3 */
431 "psllq $40, %%mm7 \n\t" /* RGB00000 0 */
432 "psllq $40, %%mm0 \n\t" /* RGB00000 1 */
433 "psllq $40, %%mm5 \n\t" /* RGB00000 2 */
434 "psllq $40, %%mm3 \n\t" /* RGB00000 3 */
436 "punpckhdq %%mm2, %%mm7 \n\t" /* 0RGBRGB0 0 */
437 "punpckhdq %%mm6, %%mm0 \n\t" /* 0RGBRGB0 1 */
438 "punpckhdq %%mm1, %%mm5 \n\t" /* 0RGBRGB0 2 */
439 "punpckhdq %%mm4, %%mm3 \n\t" /* 0RGBRGB0 3 */
441 "psrlq $8, %%mm7 \n\t" /* 00RGBRGB 0 */
442 "movq %%mm0, %%mm6 \n\t" /* 0RGBRGB0 1 */
443 "psllq $40, %%mm0 \n\t" /* GB000000 1 */
444 "por %%mm0, %%mm7 \n\t" /* GBRGBRGB 0 */
445 MOVNTQ" %%mm7, (%3) \n\t"
447 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
449 "psrlq $24, %%mm6 \n\t" /* 0000RGBR 1 */
450 "movq %%mm5, %%mm1 \n\t" /* 0RGBRGB0 2 */
451 "psllq $24, %%mm5 \n\t" /* BRGB0000 2 */
452 "por %%mm5, %%mm6 \n\t" /* BRGBRGBR 1 */
453 MOVNTQ" %%mm6, 8(%3) \n\t"
455 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
457 "psrlq $40, %%mm1 \n\t" /* 000000RG 2 */
458 "psllq $8, %%mm3 \n\t" /* RGBRGB00 3 */
459 "por %%mm3, %%mm1 \n\t" /* RGBRGBRG 2 */
460 MOVNTQ" %%mm1, 16(%3) \n\t"
462 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
463 "pxor %%mm4, %%mm4 \n\t"
466 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
485 __asm__ __volatile__ (EMMS);
489 static inline void RENAME(yuv420_argb32) (uint8_t * image, uint8_t * py,
490 uint8_t * pu, uint8_t * pv,
491 int h_size, int v_size,
492 int rgb_stride, int y_stride, int uv_stride)
497 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
499 for (y = v_size; --y >= 0; ) {
500 uint8_t *_image = image;
504 int internal_h_size= h_size;
505 int aligned_h_size= (h_size+7)&~7;
507 if(rgb_stride >= aligned_h_size*4) internal_h_size= aligned_h_size;
509 /* load data for start of next scan line */
512 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
513 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
514 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
515 : : "r" (_py), "r" (_pu), "r" (_pv)
518 for (x = internal_h_size >> 3; --x >= 0; ) {
519 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
520 pixels in each iteration */
521 __asm__ __volatile__ (
523 /* convert RGB plane to RGB packed format,
524 mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
525 mm4 -> GB, mm5 -> AR pixel 4-7,
526 mm6 -> GB, mm7 -> AR pixel 0-3 */
527 "pxor %%mm3, %%mm3;" /* zero mm3 */
529 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
530 "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
532 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
533 "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
535 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
536 "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */
538 "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */
539 MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */
541 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
542 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
544 "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */
545 MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */
547 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
548 "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */
550 "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */
551 MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */
553 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
554 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
556 "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */
557 MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */
559 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
560 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
562 "pxor %%mm4, %%mm4;" /* zero mm4 */
563 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
565 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
584 __asm__ __volatile__ (EMMS);
587 yuv2rgb_fun RENAME(yuv2rgb_init) (int bpp, int mode)
589 if (bpp == 15 && mode == MODE_RGB) return RENAME(yuv420_rgb15);
590 if (bpp == 16 && mode == MODE_RGB) return RENAME(yuv420_rgb16);
591 if (bpp == 24 && mode == MODE_RGB) return RENAME(yuv420_rgb24);
592 if (bpp == 32 && mode == MODE_RGB) return RENAME(yuv420_argb32);
593 return NULL; // Fallback to C.