Add 8088flex TMV video decoder.
[ffmpeg.git] / libavcodec / sh4 / idct_sh4.c
1 /*
2  * idct for sh4
3  *
4  * Copyright (c) 2001-2003 BERO <bero@geocities.co.jp>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 #include "libavcodec/dsputil.h"
24 #include "sh4.h"
25
26 #define c1      1.38703984532214752434  /* sqrt(2)*cos(1*pi/16) */
27 #define c2      1.30656296487637657577  /* sqrt(2)*cos(2*pi/16) */
28 #define c3      1.17587560241935884520  /* sqrt(2)*cos(3*pi/16) */
29 #define c4      1.00000000000000000000  /* sqrt(2)*cos(4*pi/16) */
30 #define c5      0.78569495838710234903  /* sqrt(2)*cos(5*pi/16) */
31 #define c6      0.54119610014619712324  /* sqrt(2)*cos(6*pi/16) */
32 #define c7      0.27589937928294311353  /* sqrt(2)*cos(7*pi/16) */
33
34 static const float even_table[] __attribute__ ((aligned(8))) = {
35         c4, c4, c4, c4,
36         c2, c6,-c6,-c2,
37         c4,-c4,-c4, c4,
38         c6,-c2, c2,-c6
39 };
40
41 static const float odd_table[] __attribute__ ((aligned(8))) = {
42         c1, c3, c5, c7,
43         c3,-c7,-c1,-c5,
44         c5,-c1, c7, c3,
45         c7,-c5, c3,-c1
46 };
47
48 #undef  c1
49 #undef  c2
50 #undef  c3
51 #undef  c4
52 #undef  c5
53 #undef  c6
54 #undef  c7
55
56 #if 1
57
58 #define         load_matrix(table) \
59     do { \
60         const float *t = table; \
61         __asm__ volatile( \
62         "       fschg\n" \
63         "       fmov   @%0+,xd0\n" \
64         "       fmov   @%0+,xd2\n" \
65         "       fmov   @%0+,xd4\n" \
66         "       fmov   @%0+,xd6\n" \
67         "       fmov   @%0+,xd8\n" \
68         "       fmov   @%0+,xd10\n" \
69         "       fmov   @%0+,xd12\n" \
70         "       fmov   @%0+,xd14\n" \
71         "       fschg\n" \
72         : "+r"(t) \
73         ); \
74     } while (0)
75
76 #define         ftrv() \
77                 __asm__ volatile("ftrv xmtrx,fv0" \
78                 : "+f"(fr0),"+f"(fr1),"+f"(fr2),"+f"(fr3));
79
80 #define         DEFREG        \
81         register float fr0 __asm__("fr0"); \
82         register float fr1 __asm__("fr1"); \
83         register float fr2 __asm__("fr2"); \
84         register float fr3 __asm__("fr3")
85
86 #else
87
88 /* generic C code for check */
89
90 static void ftrv_(const float xf[],float fv[])
91 {
92         float f0,f1,f2,f3;
93         f0 = fv[0];
94         f1 = fv[1];
95         f2 = fv[2];
96         f3 = fv[3];
97         fv[0] = xf[0]*f0 + xf[4]*f1 + xf[ 8]*f2 + xf[12]*f3;
98         fv[1] = xf[1]*f0 + xf[5]*f1 + xf[ 9]*f2 + xf[13]*f3;
99         fv[2] = xf[2]*f0 + xf[6]*f1 + xf[10]*f2 + xf[14]*f3;
100         fv[3] = xf[3]*f0 + xf[7]*f1 + xf[11]*f2 + xf[15]*f3;
101 }
102
103 static void load_matrix_(float xf[],const float table[])
104 {
105         int i;
106         for(i=0;i<16;i++) xf[i]=table[i];
107 }
108
109 #define         ftrv()                  ftrv_(xf,fv)
110 #define         load_matrix(table)      load_matrix_(xf,table)
111
112 #define         DEFREG \
113         float fv[4],xf[16]
114
115 #define         fr0     fv[0]
116 #define         fr1     fv[1]
117 #define         fr2     fv[2]
118 #define         fr3     fv[3]
119
120 #endif
121
122 #if 1
123 #define         DESCALE(x,n)    (x)*(1.0f/(1<<(n)))
124 #else
125 #define         DESCALE(x,n)    (((int)(x)+(1<<(n-1)))>>(n))
126 #endif
127
128 /* this code work worse on gcc cvs. 3.2.3 work fine */
129
130
131 #if 1
132 //optimized
133
134 void idct_sh4(DCTELEM *block)
135 {
136         DEFREG;
137
138         int i;
139         float        tblock[8*8],*fblock;
140         int ofs1,ofs2,ofs3;
141         int fpscr;
142
143         fp_single_enter(fpscr);
144
145         /* row */
146
147         /* even part */
148         load_matrix(even_table);
149
150         fblock = tblock+4;
151         i = 8;
152         do {
153                 fr0 = block[0];
154                 fr1 = block[2];
155                 fr2 = block[4];
156                 fr3 = block[6];
157                 block+=8;
158                 ftrv();
159                 *--fblock = fr3;
160                 *--fblock = fr2;
161                 *--fblock = fr1;
162                 *--fblock = fr0;
163                 fblock+=8+4;
164         } while(--i);
165         block-=8*8;
166         fblock-=8*8+4;
167
168         load_matrix(odd_table);
169
170         i = 8;
171
172         do {
173                 float t0,t1,t2,t3;
174                 fr0 = block[1];
175                 fr1 = block[3];
176                 fr2 = block[5];
177                 fr3 = block[7];
178                 block+=8;
179                 ftrv();
180                 t0 = *fblock++;
181                 t1 = *fblock++;
182                 t2 = *fblock++;
183                 t3 = *fblock++;
184                 fblock+=4;
185                 *--fblock = t0 - fr0;
186                 *--fblock = t1 - fr1;
187                 *--fblock = t2 - fr2;
188                 *--fblock = t3 - fr3;
189                 *--fblock = t3 + fr3;
190                 *--fblock = t2 + fr2;
191                 *--fblock = t1 + fr1;
192                 *--fblock = t0 + fr0;
193                 fblock+=8;
194         } while(--i);
195         block-=8*8;
196         fblock-=8*8;
197
198         /* col */
199
200         /* even part */
201         load_matrix(even_table);
202
203         ofs1 = sizeof(float)*2*8;
204         ofs2 = sizeof(float)*4*8;
205         ofs3 = sizeof(float)*6*8;
206
207         i = 8;
208
209 #define        OA(fblock,ofs)   *(float*)((char*)fblock + ofs)
210
211         do {
212                 fr0 = OA(fblock,   0);
213                 fr1 = OA(fblock,ofs1);
214                 fr2 = OA(fblock,ofs2);
215                 fr3 = OA(fblock,ofs3);
216                 ftrv();
217                 OA(fblock,0   ) = fr0;
218                 OA(fblock,ofs1) = fr1;
219                 OA(fblock,ofs2) = fr2;
220                 OA(fblock,ofs3) = fr3;
221                 fblock++;
222         } while(--i);
223         fblock-=8;
224
225         load_matrix(odd_table);
226
227         i=8;
228         do {
229                 float t0,t1,t2,t3;
230                 t0 = OA(fblock,   0); /* [8*0] */
231                 t1 = OA(fblock,ofs1); /* [8*2] */
232                 t2 = OA(fblock,ofs2); /* [8*4] */
233                 t3 = OA(fblock,ofs3); /* [8*6] */
234                 fblock+=8;
235                 fr0 = OA(fblock,   0); /* [8*1] */
236                 fr1 = OA(fblock,ofs1); /* [8*3] */
237                 fr2 = OA(fblock,ofs2); /* [8*5] */
238                 fr3 = OA(fblock,ofs3); /* [8*7] */
239                 fblock+=-8+1;
240                 ftrv();
241                 block[8*0] = DESCALE(t0 + fr0,3);
242                 block[8*7] = DESCALE(t0 - fr0,3);
243                 block[8*1] = DESCALE(t1 + fr1,3);
244                 block[8*6] = DESCALE(t1 - fr1,3);
245                 block[8*2] = DESCALE(t2 + fr2,3);
246                 block[8*5] = DESCALE(t2 - fr2,3);
247                 block[8*3] = DESCALE(t3 + fr3,3);
248                 block[8*4] = DESCALE(t3 - fr3,3);
249                 block++;
250         } while(--i);
251
252         fp_single_leave(fpscr);
253 }
254 #else
255 void idct_sh4(DCTELEM *block)
256 {
257         DEFREG;
258
259         int i;
260         float   tblock[8*8],*fblock;
261
262         /* row */
263
264         /* even part */
265         load_matrix(even_table);
266
267         fblock = tblock;
268         i = 8;
269         do {
270                 fr0 = block[0];
271                 fr1 = block[2];
272                 fr2 = block[4];
273                 fr3 = block[6];
274                 block+=8;
275                 ftrv();
276                 fblock[0] = fr0;
277                 fblock[2] = fr1;
278                 fblock[4] = fr2;
279                 fblock[6] = fr3;
280                 fblock+=8;
281         } while(--i);
282         block-=8*8;
283         fblock-=8*8;
284
285         load_matrix(odd_table);
286
287         i = 8;
288
289         do {
290                 float t0,t1,t2,t3;
291                 fr0 = block[1];
292                 fr1 = block[3];
293                 fr2 = block[5];
294                 fr3 = block[7];
295                 block+=8;
296                 ftrv();
297                 t0 = fblock[0];
298                 t1 = fblock[2];
299                 t2 = fblock[4];
300                 t3 = fblock[6];
301                 fblock[0] = t0 + fr0;
302                 fblock[7] = t0 - fr0;
303                 fblock[1] = t1 + fr1;
304                 fblock[6] = t1 - fr1;
305                 fblock[2] = t2 + fr2;
306                 fblock[5] = t2 - fr2;
307                 fblock[3] = t3 + fr3;
308                 fblock[4] = t3 - fr3;
309                 fblock+=8;
310         } while(--i);
311         block-=8*8;
312         fblock-=8*8;
313
314         /* col */
315
316         /* even part */
317         load_matrix(even_table);
318
319         i = 8;
320
321         do {
322                 fr0 = fblock[8*0];
323                 fr1 = fblock[8*2];
324                 fr2 = fblock[8*4];
325                 fr3 = fblock[8*6];
326                 ftrv();
327                 fblock[8*0] = fr0;
328                 fblock[8*2] = fr1;
329                 fblock[8*4] = fr2;
330                 fblock[8*6] = fr3;
331                 fblock++;
332         } while(--i);
333         fblock-=8;
334
335         load_matrix(odd_table);
336
337         i=8;
338         do {
339                 float t0,t1,t2,t3;
340                 fr0 = fblock[8*1];
341                 fr1 = fblock[8*3];
342                 fr2 = fblock[8*5];
343                 fr3 = fblock[8*7];
344                 ftrv();
345                 t0 = fblock[8*0];
346                 t1 = fblock[8*2];
347                 t2 = fblock[8*4];
348                 t3 = fblock[8*6];
349                 fblock++;
350                 block[8*0] = DESCALE(t0 + fr0,3);
351                 block[8*7] = DESCALE(t0 - fr0,3);
352                 block[8*1] = DESCALE(t1 + fr1,3);
353                 block[8*6] = DESCALE(t1 - fr1,3);
354                 block[8*2] = DESCALE(t2 + fr2,3);
355                 block[8*5] = DESCALE(t2 - fr2,3);
356                 block[8*3] = DESCALE(t3 + fr3,3);
357                 block[8*4] = DESCALE(t3 - fr3,3);
358                 block++;
359         } while(--i);
360 }
361 #endif