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.
31 #include "../config.h"
33 //#include "libmpeg2/mpeg2.h"
34 //#include "libmpeg2/mpeg2_internal.h"
39 /* hope these constant values are cache line aligned */
40 uint64_t mmx_80w = 0x0080008000800080;
41 uint64_t mmx_10w = 0x1010101010101010;
42 uint64_t mmx_00ffw = 0x00ff00ff00ff00ff;
43 uint64_t mmx_Y_coeff = 0x253f253f253f253f;
45 /* hope these constant values are cache line aligned */
46 uint64_t mmx_U_green = 0xf37df37df37df37d;
47 uint64_t mmx_U_blue = 0x4093409340934093;
48 uint64_t mmx_V_red = 0x3312331233123312;
49 uint64_t mmx_V_green = 0xe5fce5fce5fce5fc;
51 /* hope these constant values are cache line aligned */
52 uint64_t mmx_redmask = 0xf8f8f8f8f8f8f8f8;
53 uint64_t mmx_grnmask = 0xfcfcfcfcfcfcfcfc;
54 uint64_t mmx_grnshift = 0x03;
55 uint64_t mmx_blueshift = 0x03;
58 /* use this for K7 and p3 only */
59 #define MOVNTQ "movntq"
61 /* for MMX-only processors */
65 #if !defined( HAVE_MMX2) && defined( HAVE_3DNOW)
72 static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
73 uint8_t * pu, uint8_t * pv,
74 int h_size, int v_size,
75 int rgb_stride, int y_stride, int uv_stride)
80 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
82 for (y = v_size; --y >= 0; ) {
83 uint8_t *_image = image;
88 /* load data for start of next scan line */
89 __asm__ __volatile__ (
90 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
91 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
92 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
94 : : "r" (_py), "r" (_pu), "r" (_pv));
96 for (x = h_size >> 3; --x >= 0; ) {
97 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
98 pixels in each iteration */
100 __asm__ __volatile__ (
101 /* Do the multiply part of the conversion for even and odd pixels,
103 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
104 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
105 mm6 -> Y even, mm7 -> Y odd */
106 /* convert the chroma part */
107 "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */
108 "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */
110 "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */
111 "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */
113 "psllw $3, %%mm0;" /* Promote precision */
114 "psllw $3, %%mm1;" /* Promote precision */
116 "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */
117 "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */
119 "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */
120 "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */
122 "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */
123 "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */
125 "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */
127 /* convert the luma part */
128 "psubusb mmx_10w, %%mm6;" /* Y -= 16 */
130 "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
131 "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */
133 "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */
135 "psllw $3, %%mm6;" /* Promote precision */
136 "psllw $3, %%mm7;" /* Promote precision */
138 "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */
139 "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */
141 /* Do the addition part of the conversion for even and odd pixels,
143 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
144 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
145 mm6 -> Y even, mm7 -> Y odd */
146 "movq %%mm0, %%mm3;" /* Copy Cblue */
147 "movq %%mm1, %%mm4;" /* Copy Cred */
148 "movq %%mm2, %%mm5;" /* Copy Cgreen */
150 "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */
151 "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */
153 "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */
154 "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */
156 "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */
157 "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */
159 /* Limit RGB even to 0..255 */
160 "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0 B6 B4 B2 B0 */
161 "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0 R6 R4 R2 R0 */
162 "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0 G6 G4 G2 G0 */
164 /* Limit RGB odd to 0..255 */
165 "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1 B7 B5 B3 B1 */
166 "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1 R7 R5 R3 R1 */
167 "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1 G7 G5 G3 G1 */
169 /* Interleave RGB even and odd */
170 "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
171 "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
172 "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */
174 /* mask unneeded bits off */
175 "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
176 "pand mmx_grnmask, %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */
177 "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
179 "psrlw mmx_blueshift,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
180 "pxor %%mm4, %%mm4;" /* zero mm4 */
182 "movq %%mm0, %%mm5;" /* Copy B7-B0 */
183 "movq %%mm2, %%mm7;" /* Copy G7-G0 */
185 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
186 "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
187 "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
189 "psllw mmx_blueshift,%%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
190 "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
192 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
193 MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
195 /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
196 "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
197 "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
199 "psllw mmx_blueshift,%%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
200 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
202 "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
203 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
205 MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
206 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
225 __asm__ __volatile__ (EMMS);
228 static void yuv420_argb32_mmx (uint8_t * image, uint8_t * py,
229 uint8_t * pu, uint8_t * pv,
230 int h_size, int v_size,
231 int rgb_stride, int y_stride, int uv_stride)
236 __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
238 for (y = v_size; --y >= 0; ) {
239 uint8_t *_image = image;
244 /* load data for start of next scan line */
247 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
248 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
249 "movq (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
250 : : "r" (_py), "r" (_pu), "r" (_pv)
253 for (x = h_size >> 3; --x >= 0; ) {
254 /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
255 pixels in each iteration */
256 __asm__ __volatile__ (
257 /* Do the multiply part of the conversion for even and odd pixels,
259 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
260 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
261 mm6 -> Y even, mm7 -> Y odd */
263 /* convert the chroma part */
264 "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */
265 "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */
267 "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */
268 "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */
270 "psllw $3, %%mm0;" /* Promote precision */
271 "psllw $3, %%mm1;" /* Promote precision */
273 "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */
274 "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */
276 "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */
277 "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */
279 "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */
280 "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */
282 "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */
284 /* convert the luma part */
285 "psubusb mmx_10w, %%mm6;" /* Y -= 16 */
287 "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
288 "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */
290 "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */
292 "psllw $3, %%mm6;" /* Promote precision */
293 "psllw $3, %%mm7;" /* Promote precision */
295 "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */
296 "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */
298 /* Do the addition part of the conversion for even and odd pixels,
300 mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
301 mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
302 mm6 -> Y even, mm7 -> Y odd */
304 "movq %%mm0, %%mm3;" /* Copy Cblue */
305 "movq %%mm1, %%mm4;" /* Copy Cred */
306 "movq %%mm2, %%mm5;" /* Copy Cgreen */
308 "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */
309 "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */
311 "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */
312 "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */
314 "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */
315 "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */
317 /* Limit RGB even to 0..255 */
318 "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0 B6 B4 B2 B0 */
319 "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0 R6 R4 R2 R0 */
320 "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0 G6 G4 G2 G0 */
322 /* Limit RGB odd to 0..255 */
323 "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1 B7 B5 B3 B1 */
324 "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1 R7 R5 R3 R1 */
325 "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1 G7 G5 G3 G1 */
327 /* Interleave RGB even and odd */
328 "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
329 "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
330 "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */
332 /* convert RGB plane to RGB packed format,
333 mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
334 mm4 -> GB, mm5 -> AR pixel 4-7,
335 mm6 -> GB, mm7 -> AR pixel 0-3 */
336 "pxor %%mm3, %%mm3;" /* zero mm3 */
338 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
339 "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
341 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
342 "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
344 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
345 "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */
347 "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */
348 MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */
350 "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
351 "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
353 "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */
354 MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */
356 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
357 "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */
359 "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */
360 MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */
362 "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
363 "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
365 "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */
366 MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */
368 "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
369 "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
371 "pxor %%mm4, %%mm4;" /* zero mm4 */
372 "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
374 : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
393 __asm__ __volatile__ (EMMS);
396 yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode)
398 // if (bpp == 15 || bpp == 16) {
399 if (bpp == 16 && mode == MODE_RGB) return yuv420_rgb16_mmx;
400 if (bpp == 32 && mode == MODE_RGB) return yuv420_argb32_mmx;
401 return NULL; // Fallback to C.