1 /*****************************************************************************
2 * transforms_yuvmmx.h: MMX YUV transformation assembly
3 *****************************************************************************
4 * Copyright (C) 1999-2004 VideoLAN (Centrale Réseaux) and its contributors
7 * Authors: Olie Lho <ollie@sis.com.tw>
8 * Gaël Hendryckx <jimmy@via.ecp.fr>
9 * Samuel Hocevar <sam@zoy.org>
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
24 *****************************************************************************/
26 /* hope these constant values are cache line aligned */
27 #if __GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 3)
28 #define USED_U64(foo) \
29 static const uint64_t foo __asm__ (#foo) __attribute__((used))
31 #define USED_U64(foo) \
32 static const uint64_t foo __asm__ (#foo) __attribute__((unused))
34 USED_U64(mmx_80w) = 0x0080008000800080ULL;
35 USED_U64(mmx_10w) = 0x1010101010101010ULL;
36 USED_U64(mmx_00ffw) = 0x00ff00ff00ff00ffULL;
37 USED_U64(mmx_Y_coeff) = 0x253f253f253f253fULL;
39 USED_U64(mmx_U_green) = 0xf37df37df37df37dULL;
40 USED_U64(mmx_U_blue) = 0x4093409340934093ULL;
41 USED_U64(mmx_V_red) = 0x3312331233123312ULL;
42 USED_U64(mmx_V_green) = 0xe5fce5fce5fce5fcULL;
44 USED_U64(mmx_mask_f8) = 0xf8f8f8f8f8f8f8f8ULL;
45 USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL;
48 #define MMX_INIT_16 " \n\
49 movd (%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
50 movd (%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
51 pxor %%mm4, %%mm4 # zero mm4 \n\
52 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
53 #movl $0, (%3) # cache preload for image \n\
56 #define MMX_INIT_16_GRAY " \n\
57 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
58 #movl $0, (%3) # cache preload for image \n\
61 #define MMX_INIT_32 " \n\
62 movd (%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
63 movl $0, (%3) # cache preload for image \n\
64 movd (%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
65 pxor %%mm4, %%mm4 # zero mm4 \n\
66 movq (%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
70 * Do the multiply part of the conversion for even and odd pixels,
72 * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
73 * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
74 * mm6 -> Y even, mm7 -> Y odd
77 #define MMX_YUV_MUL " \n\
78 # convert the chroma part \n\
79 punpcklbw %%mm4, %%mm0 # scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 \n\
80 punpcklbw %%mm4, %%mm1 # scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 \n\
81 psubsw mmx_80w, %%mm0 # Cb -= 128 \n\
82 psubsw mmx_80w, %%mm1 # Cr -= 128 \n\
83 psllw $3, %%mm0 # Promote precision \n\
84 psllw $3, %%mm1 # Promote precision \n\
85 movq %%mm0, %%mm2 # Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 \n\
86 movq %%mm1, %%mm3 # Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 \n\
87 pmulhw mmx_U_green, %%mm2 # Mul Cb with green coeff -> Cb green \n\
88 pmulhw mmx_V_green, %%mm3 # Mul Cr with green coeff -> Cr green \n\
89 pmulhw mmx_U_blue, %%mm0 # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 \n\
90 pmulhw mmx_V_red, %%mm1 # Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 \n\
91 paddsw %%mm3, %%mm2 # Cb green + Cr green -> Cgreen \n\
93 # convert the luma part \n\
94 psubusb mmx_10w, %%mm6 # Y -= 16 \n\
95 movq %%mm6, %%mm7 # Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
96 pand mmx_00ffw, %%mm6 # get Y even 00 Y6 00 Y4 00 Y2 00 Y0 \n\
97 psrlw $8, %%mm7 # get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 \n\
98 psllw $3, %%mm6 # Promote precision \n\
99 psllw $3, %%mm7 # Promote precision \n\
100 pmulhw mmx_Y_coeff, %%mm6 # Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 \n\
101 pmulhw mmx_Y_coeff, %%mm7 # Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 \n\
105 * Do the addition part of the conversion for even and odd pixels,
107 * mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
108 * mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
109 * mm6 -> Y even, mm7 -> Y odd
112 #define MMX_YUV_ADD " \n\
113 # Do horizontal and vertical scaling \n\
114 movq %%mm0, %%mm3 # Copy Cblue \n\
115 movq %%mm1, %%mm4 # Copy Cred \n\
116 movq %%mm2, %%mm5 # Copy Cgreen \n\
117 paddsw %%mm6, %%mm0 # Y even + Cblue 00 B6 00 B4 00 B2 00 B0 \n\
118 paddsw %%mm7, %%mm3 # Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 \n\
119 paddsw %%mm6, %%mm1 # Y even + Cred 00 R6 00 R4 00 R2 00 R0 \n\
120 paddsw %%mm7, %%mm4 # Y odd + Cred 00 R7 00 R5 00 R3 00 R1 \n\
121 paddsw %%mm6, %%mm2 # Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 \n\
122 paddsw %%mm7, %%mm5 # Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 \n\
124 # Limit RGB even to 0..255 \n\
125 packuswb %%mm0, %%mm0 # B6 B4 B2 B0 / B6 B4 B2 B0 \n\
126 packuswb %%mm1, %%mm1 # R6 R4 R2 R0 / R6 R4 R2 R0 \n\
127 packuswb %%mm2, %%mm2 # G6 G4 G2 G0 / G6 G4 G2 G0 \n\
129 # Limit RGB odd to 0..255 \n\
130 packuswb %%mm3, %%mm3 # B7 B5 B3 B1 / B7 B5 B3 B1 \n\
131 packuswb %%mm4, %%mm4 # R7 R5 R3 R1 / R7 R5 R3 R1 \n\
132 packuswb %%mm5, %%mm5 # G7 G5 G3 G1 / G7 G5 G3 G1 \n\
134 # Interleave RGB even and odd \n\
135 punpcklbw %%mm3, %%mm0 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
136 punpcklbw %%mm4, %%mm1 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
137 punpcklbw %%mm5, %%mm2 # G7 G6 G5 G4 G3 G2 G1 G0 \n\
141 * Grayscale case, only use Y
144 #define MMX_YUV_GRAY " \n\
145 # convert the luma part \n\
146 psubusb mmx_10w, %%mm6 \n\
147 movq %%mm6, %%mm7 \n\
148 pand mmx_00ffw, %%mm6 \n\
152 pmulhw mmx_Y_coeff, %%mm6 \n\
153 pmulhw mmx_Y_coeff, %%mm7 \n\
154 packuswb %%mm6, %%mm6 \n\
155 packuswb %%mm7, %%mm7 \n\
156 punpcklbw %%mm7, %%mm6 \n\
159 #define MMX_UNPACK_16_GRAY " \n\
160 movq %%mm6, %%mm5 \n\
161 pand mmx_mask_f8, %%mm6 \n\
162 pand mmx_mask_fc, %%mm5 \n\
163 movq %%mm6, %%mm7 \n\
165 pxor %%mm3, %%mm3 \n\
166 movq %%mm7, %%mm2 \n\
167 movq %%mm5, %%mm0 \n\
168 punpcklbw %%mm3, %%mm5 \n\
169 punpcklbw %%mm6, %%mm7 \n\
173 punpckhbw %%mm3, %%mm0 \n\
174 punpckhbw %%mm6, %%mm2 \n\
176 movq 8(%0), %%mm6 \n\
178 movq %%mm2, 8(%3) \n\
183 * convert RGB plane to RGB 15 bits,
184 * mm0 -> B, mm1 -> R, mm2 -> G,
185 * mm4 -> GB, mm5 -> AR pixel 4-7,
186 * mm6 -> GB, mm7 -> AR pixel 0-3
189 #define MMX_UNPACK_15 " \n\
190 # mask unneeded bits off \n\
191 pand mmx_mask_f8, %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______ \n\
192 psrlw $3,%%mm0 # ______b7 b6b5b4b3 ______b7 b6b5b4b3 \n\
193 pand mmx_mask_f8, %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______ \n\
194 pand mmx_mask_f8, %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______ \n\
195 psrlw $1,%%mm1 # __r7r6r5 r4r3____ __r7r6r5 r4r3____ \n\
196 pxor %%mm4, %%mm4 # zero mm4 \n\
197 movq %%mm0, %%mm5 # Copy B7-B0 \n\
198 movq %%mm2, %%mm7 # Copy G7-G0 \n\
200 # convert rgb24 plane to rgb15 pack for pixel 0-3 \n\
201 punpcklbw %%mm4, %%mm2 # ________ ________ g7g6g5g4 g3______ \n\
202 punpcklbw %%mm1, %%mm0 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
203 psllw $2,%%mm2 # ________ ____g7g6 g5g4g3__ ________ \n\
204 por %%mm2, %%mm0 # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3 \n\
205 movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
206 movq %%mm0, (%3) # store pixel 0-3 \n\
208 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
209 punpckhbw %%mm4, %%mm7 # ________ ________ g7g6g5g4 g3______ \n\
210 punpckhbw %%mm1, %%mm5 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
211 psllw $2,%%mm7 # ________ ____g7g6 g5g4g3__ ________ \n\
212 movd 4(%1), %%mm0 # Load 4 Cb __ __ __ __ u3 u2 u1 u0 \n\
213 por %%mm7, %%mm5 # r7r6r5r4 r3__g7g6 g5g4g3b7 b6b5b4b3 \n\
214 movd 4(%2), %%mm1 # Load 4 Cr __ __ __ __ v3 v2 v1 v0 \n\
215 movq %%mm5, 8(%3) # store pixel 4-7 \n\
219 * convert RGB plane to RGB 16 bits,
220 * mm0 -> B, mm1 -> R, mm2 -> G,
221 * mm4 -> GB, mm5 -> AR pixel 4-7,
222 * mm6 -> GB, mm7 -> AR pixel 0-3
225 #define MMX_UNPACK_16 " \n\
226 # mask unneeded bits off \n\
227 pand mmx_mask_f8, %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______ \n\
228 pand mmx_mask_fc, %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____ \n\
229 pand mmx_mask_f8, %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______ \n\
230 psrlw $3,%%mm0 # ______b7 b6b5b4b3 ______b7 b6b5b4b3 \n\
231 pxor %%mm4, %%mm4 # zero mm4 \n\
232 movq %%mm0, %%mm5 # Copy B7-B0 \n\
233 movq %%mm2, %%mm7 # Copy G7-G0 \n\
235 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
236 punpcklbw %%mm4, %%mm2 # ________ ________ g7g6g5g4 g3g2____ \n\
237 punpcklbw %%mm1, %%mm0 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
238 psllw $3,%%mm2 # ________ __g7g6g5 g4g3g2__ ________ \n\
239 por %%mm2, %%mm0 # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 \n\
240 movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\
241 movq %%mm0, (%3) # store pixel 0-3 \n\
243 # convert rgb24 plane to rgb16 pack for pixel 0-3 \n\
244 punpckhbw %%mm4, %%mm7 # ________ ________ g7g6g5g4 g3g2____ \n\
245 punpckhbw %%mm1, %%mm5 # r7r6r5r4 r3______ ______b7 b6b5b4b3 \n\
246 psllw $3,%%mm7 # ________ __g7g6g5 g4g3g2__ ________ \n\
247 movd 4(%1), %%mm0 # Load 4 Cb __ __ __ __ u3 u2 u1 u0 \n\
248 por %%mm7, %%mm5 # r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 \n\
249 movd 4(%2), %%mm1 # Load 4 Cr __ __ __ __ v3 v2 v1 v0 \n\
250 movq %%mm5, 8(%3) # store pixel 4-7 \n\
254 * convert RGB plane to RGB packed format,
255 * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
256 * mm4 -> GB, mm5 -> AR pixel 4-7,
257 * mm6 -> GB, mm7 -> AR pixel 0-3
260 #define MMX_UNPACK_32 " \n\
261 pxor %%mm3, %%mm3 # zero mm3 \n\
262 movq %%mm0, %%mm6 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
263 movq %%mm1, %%mm7 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
264 movq %%mm0, %%mm4 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
265 movq %%mm1, %%mm5 # R7 R6 R5 R4 R3 R2 R1 R0 \n\
266 punpcklbw %%mm2, %%mm6 # G3 B3 G2 B2 G1 B1 G0 B0 \n\
267 punpcklbw %%mm3, %%mm7 # 00 R3 00 R2 00 R1 00 R0 \n\
268 punpcklwd %%mm7, %%mm6 # 00 R1 B1 G1 00 R0 B0 G0 \n\
269 movq %%mm6, (%3) # Store ARGB1 ARGB0 \n\
270 movq %%mm0, %%mm6 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
271 punpcklbw %%mm2, %%mm6 # G3 B3 G2 B2 G1 B1 G0 B0 \n\
272 punpckhwd %%mm7, %%mm6 # 00 R3 G3 B3 00 R2 B3 G2 \n\
273 movq %%mm6, 8(%3) # Store ARGB3 ARGB2 \n\
274 punpckhbw %%mm2, %%mm4 # G7 B7 G6 B6 G5 B5 G4 B4 \n\
275 punpckhbw %%mm3, %%mm5 # 00 R7 00 R6 00 R5 00 R4 \n\
276 punpcklwd %%mm5, %%mm4 # 00 R5 B5 G5 00 R4 B4 G4 \n\
277 movq %%mm4, 16(%3) # Store ARGB5 ARGB4 \n\
278 movq %%mm0, %%mm4 # B7 B6 B5 B4 B3 B2 B1 B0 \n\
279 punpckhbw %%mm2, %%mm4 # G7 B7 G6 B6 G5 B5 G4 B4 \n\
280 punpckhwd %%mm5, %%mm4 # 00 R7 G7 B7 00 R6 B6 G6 \n\
281 movq %%mm4, 24(%3) # Store ARGB7 ARGB6 \n\
283 #movd 4(%1), %%mm0 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
284 #movd 4(%2), %%mm1 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
285 #pxor %%mm4, %%mm4 # zero mm4 \n\
286 #movq 8(%0), %%mm6 # Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 \n\