Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5131 | clevermous | 1 | /* |
2 | SDL - Simple DirectMedia Layer |
||
3 | Copyright (C) 1997, 1998, 1999, 2000, 2001 Sam Lantinga |
||
4 | |||
5 | This library is free software; you can redistribute it and/or |
||
6 | modify it under the terms of the GNU Library General Public |
||
7 | License as published by the Free Software Foundation; either |
||
8 | version 2 of the License, or (at your option) any later version. |
||
9 | |||
10 | This library is distributed in the hope that it will be useful, |
||
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
13 | Library General Public License for more details. |
||
14 | |||
15 | You should have received a copy of the GNU Library General Public |
||
16 | License along with this library; if not, write to the Free |
||
17 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
||
18 | |||
19 | Sam Lantinga |
||
20 | slouken@devolution.com |
||
21 | */ |
||
22 | |||
23 | |||
24 | |||
25 | #if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT) |
||
26 | |||
27 | #include "SDL_types.h" |
||
28 | |||
29 | #ifdef __ELF__ |
||
30 | #define ASM_VAR(X) _##X |
||
31 | #else |
||
32 | #define ASM_VAR(X) X |
||
33 | #endif |
||
34 | |||
35 | #define static |
||
36 | static unsigned int ASM_VAR(MMX_0080w)[] = {0x00800080, 0x00800080}; |
||
37 | static unsigned int ASM_VAR(MMX_00FFw)[] = {0x00ff00ff, 0x00ff00ff}; |
||
38 | static unsigned int ASM_VAR(MMX_FF00w)[] = {0xff00ff00, 0xff00ff00}; |
||
39 | |||
40 | static unsigned short ASM_VAR(MMX_Ycoeff)[] = {0x004a, 0x004a, 0x004a, 0x004a}; |
||
41 | |||
42 | static unsigned short ASM_VAR(MMX_UbluRGB)[] = {0x0072, 0x0072, 0x0072, 0x0072}; |
||
43 | static unsigned short ASM_VAR(MMX_VredRGB)[] = {0x0059, 0x0059, 0x0059, 0x0059}; |
||
44 | static unsigned short ASM_VAR(MMX_UgrnRGB)[] = {0xffea, 0xffea, 0xffea, 0xffea}; |
||
45 | static unsigned short ASM_VAR(MMX_VgrnRGB)[] = {0xffd2, 0xffd2, 0xffd2, 0xffd2}; |
||
46 | |||
47 | static unsigned short ASM_VAR(MMX_Ublu5x5)[] = {0x0081, 0x0081, 0x0081, 0x0081}; |
||
48 | static unsigned short ASM_VAR(MMX_Vred5x5)[] = {0x0066, 0x0066, 0x0066, 0x0066}; |
||
49 | static unsigned short ASM_VAR(MMX_Ugrn555)[] = {0xffe7, 0xffe7, 0xffe7, 0xffe7}; |
||
50 | static unsigned short ASM_VAR(MMX_Vgrn555)[] = {0xffcc, 0xffcc, 0xffcc, 0xffcc}; |
||
51 | static unsigned short ASM_VAR(MMX_Ugrn565)[] = {0xffe8, 0xffe8, 0xffe8, 0xffe8}; |
||
52 | static unsigned short ASM_VAR(MMX_Vgrn565)[] = {0xffcd, 0xffcd, 0xffcd, 0xffcd}; |
||
53 | |||
54 | static unsigned short ASM_VAR(MMX_red555)[] = {0x7c00, 0x7c00, 0x7c00, 0x7c00}; |
||
55 | static unsigned short ASM_VAR(MMX_red565)[] = {0xf800, 0xf800, 0xf800, 0xf800}; |
||
56 | static unsigned short ASM_VAR(MMX_grn555)[] = {0x03e0, 0x03e0, 0x03e0, 0x03e0}; |
||
57 | static unsigned short ASM_VAR(MMX_grn565)[] = {0x07e0, 0x07e0, 0x07e0, 0x07e0}; |
||
58 | static unsigned short ASM_VAR(MMX_blu5x5)[] = {0x001f, 0x001f, 0x001f, 0x001f}; |
||
59 | |||
60 | /** |
||
61 | This MMX assembler is my first assembler/MMX program ever. |
||
62 | Thus it maybe buggy. |
||
63 | Send patches to: |
||
64 | mvogt@rhrk.uni-kl.de |
||
65 | |||
66 | After it worked fine I have "obfuscated" the code a bit to have |
||
67 | more parallism in the MMX units. This means I moved |
||
68 | initilisation around and delayed other instruction. |
||
69 | Performance measurement did not show that this brought any advantage |
||
70 | but in theory it _should_ be faster this way. |
||
71 | |||
72 | The overall performanve gain to the C based dither was 30%-40%. |
||
73 | The MMX routine calculates 256bit=8RGB values in each cycle |
||
74 | (4 for row1 & 4 for row2) |
||
75 | |||
76 | The red/green/blue.. coefficents are taken from the mpeg_play |
||
77 | player. They look nice, but I dont know if you can have |
||
78 | better values, to avoid integer rounding errors. |
||
79 | |||
80 | |||
81 | IMPORTANT: |
||
82 | ========== |
||
83 | |||
84 | It is a requirement that the cr/cb/lum are 8 byte aligned and |
||
85 | the out are 16byte aligned or you will/may get segfaults |
||
86 | |||
87 | */ |
||
88 | |||
89 | void ColorRGBDitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix, |
||
90 | unsigned char *lum, unsigned char *cr, |
||
91 | unsigned char *cb, unsigned char *out, |
||
92 | int rows, int cols, int mod ) |
||
93 | { |
||
94 | Uint32 *row1; |
||
95 | Uint32 *row2; |
||
96 | |||
97 | unsigned char* y = lum +cols*rows; // Pointer to the end |
||
98 | int x=0; |
||
99 | row1 = (Uint32 *)out; // 32 bit target |
||
100 | row2 = (Uint32 *)out+cols+mod; // start of second row |
||
101 | mod = (mod+cols+mod)*4; // increment for row1 in byte |
||
102 | |||
103 | __asm__ __volatile__ ( |
||
104 | /* We don't really care about PIC - the code should be rewritten to use |
||
105 | relative addressing for the static tables, so right now we take the |
||
106 | COW hit on the pages this code resides. Big deal. |
||
107 | This spill is just to reduce register pressure in the PIC case. */ |
||
108 | "pushl %%ebx\n" |
||
109 | "movl %0, %%ebx\n" |
||
110 | |||
111 | ".align 8\n" |
||
112 | "1:\n" |
||
113 | |||
114 | // create Cr (result in mm1) |
||
115 | "movd (%%ebx), %%mm1\n" // 0 0 0 0 v3 v2 v1 v0 |
||
116 | "pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00 |
||
117 | "movd (%2), %%mm2\n" // 0 0 0 0 l3 l2 l1 l0 |
||
118 | "punpcklbw %%mm7,%%mm1\n" // 0 v3 0 v2 00 v1 00 v0 |
||
119 | "punpckldq %%mm1,%%mm1\n" // 00 v1 00 v0 00 v1 00 v0 |
||
120 | "psubw _MMX_0080w,%%mm1\n" // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0 |
||
121 | |||
122 | // create Cr_g (result in mm0) |
||
123 | "movq %%mm1,%%mm0\n" // r1 r1 r0 r0 r1 r1 r0 r0 |
||
124 | "pmullw _MMX_VgrnRGB,%%mm0\n"// red*-46dec=0.7136*64 |
||
125 | "pmullw _MMX_VredRGB,%%mm1\n"// red*89dec=1.4013*64 |
||
126 | "psraw $6, %%mm0\n" // red=red/64 |
||
127 | "psraw $6, %%mm1\n" // red=red/64 |
||
128 | |||
129 | // create L1 L2 (result in mm2,mm4) |
||
130 | // L2=lum+cols |
||
131 | "movq (%2,%4),%%mm3\n" // 0 0 0 0 L3 L2 L1 L0 |
||
132 | "punpckldq %%mm3,%%mm2\n" // L3 L2 L1 L0 l3 l2 l1 l0 |
||
133 | "movq %%mm2,%%mm4\n" // L3 L2 L1 L0 l3 l2 l1 l0 |
||
134 | "pand _MMX_FF00w,%%mm2\n" // L3 0 L1 0 l3 0 l1 0 |
||
135 | "pand _MMX_00FFw,%%mm4\n" // 0 L2 0 L0 0 l2 0 l0 |
||
136 | "psrlw $8,%%mm2\n" // 0 L3 0 L1 0 l3 0 l1 |
||
137 | |||
138 | // create R (result in mm6) |
||
139 | "movq %%mm2,%%mm5\n" // 0 L3 0 L1 0 l3 0 l1 |
||
140 | "movq %%mm4,%%mm6\n" // 0 L2 0 L0 0 l2 0 l0 |
||
141 | "paddsw %%mm1, %%mm5\n" // lum1+red:x R3 x R1 x r3 x r1 |
||
142 | "paddsw %%mm1, %%mm6\n" // lum1+red:x R2 x R0 x r2 x r0 |
||
143 | "packuswb %%mm5,%%mm5\n" // R3 R1 r3 r1 R3 R1 r3 r1 |
||
144 | "packuswb %%mm6,%%mm6\n" // R2 R0 r2 r0 R2 R0 r2 r0 |
||
145 | "pxor %%mm7,%%mm7\n" // 00 00 00 00 00 00 00 00 |
||
146 | "punpcklbw %%mm5,%%mm6\n" // R3 R2 R1 R0 r3 r2 r1 r0 |
||
147 | |||
148 | // create Cb (result in mm1) |
||
149 | "movd (%1), %%mm1\n" // 0 0 0 0 u3 u2 u1 u0 |
||
150 | "punpcklbw %%mm7,%%mm1\n" // 0 u3 0 u2 00 u1 00 u0 |
||
151 | "punpckldq %%mm1,%%mm1\n" // 00 u1 00 u0 00 u1 00 u0 |
||
152 | "psubw _MMX_0080w,%%mm1\n" // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0 |
||
153 | // create Cb_g (result in mm5) |
||
154 | "movq %%mm1,%%mm5\n" // u1 u1 u0 u0 u1 u1 u0 u0 |
||
155 | "pmullw _MMX_UgrnRGB,%%mm5\n" // blue*-109dec=1.7129*64 |
||
156 | "pmullw _MMX_UbluRGB,%%mm1\n" // blue*114dec=1.78125*64 |
||
157 | "psraw $6, %%mm5\n" // blue=red/64 |
||
158 | "psraw $6, %%mm1\n" // blue=blue/64 |
||
159 | |||
160 | // create G (result in mm7) |
||
161 | "movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1 |
||
162 | "movq %%mm4,%%mm7\n" // 0 L2 0 L0 0 l2 0 l1 |
||
163 | "paddsw %%mm5, %%mm3\n" // lum1+Cb_g:x G3t x G1t x g3t x g1t |
||
164 | "paddsw %%mm5, %%mm7\n" // lum1+Cb_g:x G2t x G0t x g2t x g0t |
||
165 | "paddsw %%mm0, %%mm3\n" // lum1+Cr_g:x G3 x G1 x g3 x g1 |
||
166 | "paddsw %%mm0, %%mm7\n" // lum1+blue:x G2 x G0 x g2 x g0 |
||
167 | "packuswb %%mm3,%%mm3\n" // G3 G1 g3 g1 G3 G1 g3 g1 |
||
168 | "packuswb %%mm7,%%mm7\n" // G2 G0 g2 g0 G2 G0 g2 g0 |
||
169 | "punpcklbw %%mm3,%%mm7\n" // G3 G2 G1 G0 g3 g2 g1 g0 |
||
170 | |||
171 | // create B (result in mm5) |
||
172 | "movq %%mm2,%%mm3\n" // 0 L3 0 L1 0 l3 0 l1 |
||
173 | "movq %%mm4,%%mm5\n" // 0 L2 0 L0 0 l2 0 l1 |
||
174 | "paddsw %%mm1, %%mm3\n" // lum1+blue:x B3 x B1 x b3 x b1 |
||
175 | "paddsw %%mm1, %%mm5\n" // lum1+blue:x B2 x B0 x b2 x b0 |
||
176 | "packuswb %%mm3,%%mm3\n" // B3 B1 b3 b1 B3 B1 b3 b1 |
||
177 | "packuswb %%mm5,%%mm5\n" // B2 B0 b2 b0 B2 B0 b2 b0 |
||
178 | "punpcklbw %%mm3,%%mm5\n" // B3 B2 B1 B0 b3 b2 b1 b0 |
||
179 | |||
180 | // fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb) |
||
181 | |||
182 | "pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0 |
||
183 | "pxor %%mm4,%%mm4\n" // 0 0 0 0 0 0 0 0 |
||
184 | "movq %%mm6,%%mm1\n" // R3 R2 R1 R0 r3 r2 r1 r0 |
||
185 | "movq %%mm5,%%mm3\n" // B3 B2 B1 B0 b3 b2 b1 b0 |
||
186 | // process lower lum |
||
187 | "punpcklbw %%mm4,%%mm1\n" // 0 r3 0 r2 0 r1 0 r0 |
||
188 | "punpcklbw %%mm4,%%mm3\n" // 0 b3 0 b2 0 b1 0 b0 |
||
189 | "movq %%mm1,%%mm2\n" // 0 r3 0 r2 0 r1 0 r0 |
||
190 | "movq %%mm3,%%mm0\n" // 0 b3 0 b2 0 b1 0 b0 |
||
191 | "punpcklwd %%mm1,%%mm3\n" // 0 r1 0 b1 0 r0 0 b0 |
||
192 | "punpckhwd %%mm2,%%mm0\n" // 0 r3 0 b3 0 r2 0 b2 |
||
193 | |||
194 | "pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0 |
||
195 | "movq %%mm7,%%mm1\n" // G3 G2 G1 G0 g3 g2 g1 g0 |
||
196 | "punpcklbw %%mm1,%%mm2\n" // g3 0 g2 0 g1 0 g0 0 |
||
197 | "punpcklwd %%mm4,%%mm2\n" // 0 0 g1 0 0 0 g0 0 |
||
198 | "por %%mm3, %%mm2\n" // 0 r1 g1 b1 0 r0 g0 b0 |
||
199 | "movq %%mm2,(%3)\n" // wrote out ! row1 |
||
200 | |||
201 | "pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0 |
||
202 | "punpcklbw %%mm1,%%mm4\n" // g3 0 g2 0 g1 0 g0 0 |
||
203 | "punpckhwd %%mm2,%%mm4\n" // 0 0 g3 0 0 0 g2 0 |
||
204 | "por %%mm0, %%mm4\n" // 0 r3 g3 b3 0 r2 g2 b2 |
||
205 | "movq %%mm4,8(%3)\n" // wrote out ! row1 |
||
206 | |||
207 | // fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb) |
||
208 | // this can be done "destructive" |
||
209 | "pxor %%mm2,%%mm2\n" // 0 0 0 0 0 0 0 0 |
||
210 | "punpckhbw %%mm2,%%mm6\n" // 0 R3 0 R2 0 R1 0 R0 |
||
211 | "punpckhbw %%mm1,%%mm5\n" // G3 B3 G2 B2 G1 B1 G0 B0 |
||
212 | "movq %%mm5,%%mm1\n" // G3 B3 G2 B2 G1 B1 G0 B0 |
||
213 | "punpcklwd %%mm6,%%mm1\n" // 0 R1 G1 B1 0 R0 G0 B0 |
||
214 | "movq %%mm1,(%5)\n" // wrote out ! row2 |
||
215 | "punpckhwd %%mm6,%%mm5\n" // 0 R3 G3 B3 0 R2 G2 B2 |
||
216 | "movq %%mm5,8(%5)\n" // wrote out ! row2 |
||
217 | |||
218 | "addl $4,%2\n" // lum+4 |
||
219 | "leal 16(%3),%3\n" // row1+16 |
||
220 | "leal 16(%5),%5\n" // row2+16 |
||
221 | "addl $2, %%ebx\n" // cr+2 |
||
222 | "addl $2, %1\n" // cb+2 |
||
223 | |||
224 | "addl $4,%6\n" // x+4 |
||
225 | "cmpl %4,%6\n" |
||
226 | |||
227 | "jl 1b\n" |
||
228 | "addl %4, %2\n" // lum += cols |
||
229 | "addl %8, %3\n" // row1+= mod |
||
230 | "addl %8, %5\n" // row2+= mod |
||
231 | "movl $0, %6\n" // x=0 |
||
232 | "cmpl %7, %2\n" |
||
233 | "jl 1b\n" |
||
234 | "emms\n" |
||
235 | "popl %%ebx\n" |
||
236 | : |
||
237 | : "m" (cr), "r"(cb),"r"(lum), |
||
238 | "r"(row1),"r"(cols),"r"(row2),"m"(x),"m"(y),"m"(mod) |
||
239 | : "%ebx" |
||
240 | ); |
||
241 | } |
||
242 | |||
243 | void Color565DitherYV12MMX1X( int *colortab, Uint32 *rgb_2_pix, |
||
244 | unsigned char *lum, unsigned char *cr, |
||
245 | unsigned char *cb, unsigned char *out, |
||
246 | int rows, int cols, int mod ) |
||
247 | { |
||
248 | Uint16 *row1; |
||
249 | Uint16 *row2; |
||
250 | |||
251 | unsigned char* y = lum +cols*rows; /* Pointer to the end */ |
||
252 | int x=0; |
||
253 | row1 = (Uint16 *)out; |