SH4: remove some commented out blocks of code
[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 #define c1      1.38703984532214752434  /* sqrt(2)*cos(1*pi/16) */
25 #define c2      1.30656296487637657577  /* sqrt(2)*cos(2*pi/16) */
26 #define c3      1.17587560241935884520  /* sqrt(2)*cos(3*pi/16) */
27 #define c4      1.00000000000000000000  /* sqrt(2)*cos(4*pi/16) */
28 #define c5      0.78569495838710234903  /* sqrt(2)*cos(5*pi/16) */
29 #define c6      0.54119610014619712324  /* sqrt(2)*cos(6*pi/16) */
30 #define c7      0.27589937928294311353  /* sqrt(2)*cos(7*pi/16) */
31
32 static const float even_table[] __attribute__ ((aligned(8))) = {
33         c4, c4, c4, c4,
34         c2, c6,-c6,-c2,
35         c4,-c4,-c4, c4,
36         c6,-c2, c2,-c6
37 };
38
39 static const float odd_table[] __attribute__ ((aligned(8))) = {
40         c1, c3, c5, c7,
41         c3,-c7,-c1,-c5,
42         c5,-c1, c7, c3,
43         c7,-c5, c3,-c1
44 };
45
46 #undef  c1
47 #undef  c2
48 #undef  c3
49 #undef  c4
50 #undef  c5
51 #undef  c6
52 #undef  c7
53
54 #if defined(__SH4_SINGLE__) || defined(__SH4_SINGLE_ONLY__)
55
56 #define         load_matrix(table) \
57         __asm__ volatile( \
58         "       fschg\n" \
59         "       fmov   @%0+,xd0\n" \
60         "       fmov   @%0+,xd2\n" \
61         "       fmov   @%0+,xd4\n" \
62         "       fmov   @%0+,xd6\n" \
63         "       fmov   @%0+,xd8\n" \
64         "       fmov   @%0+,xd10\n" \
65         "       fmov   @%0+,xd12\n" \
66         "       fmov   @%0+,xd14\n" \
67         "       fschg\n" \
68         :\
69         : "r"(table)\
70         : "0" \
71         )
72
73 #define         ftrv() \
74                 __asm__ volatile("ftrv xmtrx,fv0" \
75                 : "=f"(fr0),"=f"(fr1),"=f"(fr2),"=f"(fr3) \
76                 :  "0"(fr0), "1"(fr1), "2"(fr2), "3"(fr3) );
77
78 #define         DEFREG        \
79         register float fr0 __asm__("fr0"); \
80         register float fr1 __asm__("fr1"); \
81         register float fr2 __asm__("fr2"); \
82         register float fr3 __asm__("fr3")
83
84 #else
85
86 /* generic C code for check */
87
88 static void ftrv_(const float xf[],float fv[])
89 {
90         float f0,f1,f2,f3;
91         f0 = fv[0];
92         f1 = fv[1];
93         f2 = fv[2];
94         f3 = fv[3];
95         fv[0] = xf[0]*f0 + xf[4]*f1 + xf[ 8]*f2 + xf[12]*f3;
96         fv[1] = xf[1]*f0 + xf[5]*f1 + xf[ 9]*f2 + xf[13]*f3;
97         fv[2] = xf[2]*f0 + xf[6]*f1 + xf[10]*f2 + xf[14]*f3;
98         fv[3] = xf[3]*f0 + xf[7]*f1 + xf[11]*f2 + xf[15]*f3;
99 }
100
101 static void load_matrix_(float xf[],const float table[])
102 {
103         int i;
104         for(i=0;i<16;i++) xf[i]=table[i];
105 }
106
107 #define         ftrv()                  ftrv_(xf,fv)
108 #define         load_matrix(table)      load_matrix_(xf,table)
109
110 #define         DEFREG \
111         float fv[4],xf[16]
112
113 #define         fr0     fv[0]
114 #define         fr1     fv[1]
115 #define         fr2     fv[2]
116 #define         fr3     fv[3]
117
118 #endif
119
120 #if 1
121 #define         DESCALE(x,n)    (x)*(1.0f/(1<<(n)))
122 #else
123 #define         DESCALE(x,n)    (((int)(x)+(1<<(n-1)))>>(n))
124 #endif
125
126 /* this code work worse on gcc cvs. 3.2.3 work fine */
127
128
129 #if 1
130 //optimized
131
132 void idct_sh4(DCTELEM *block)
133 {
134         DEFREG;
135
136         int i;
137         float        tblock[8*8],*fblock;
138         int ofs1,ofs2,ofs3;
139
140 #if defined(__SH4__)
141         __asm__ ("fschg");
142 #endif
143
144         /* row */
145
146         /* even part */
147         load_matrix(even_table);
148
149         fblock = tblock+4;
150         i = 8;
151         do {
152                 fr0 = block[0];
153                 fr1 = block[2];
154                 fr2 = block[4];
155                 fr3 = block[6];
156                 block+=8;
157                 ftrv();
158                 *--fblock = fr3;
159                 *--fblock = fr2;
160                 *--fblock = fr1;
161                 *--fblock = fr0;
162                 fblock+=8+4;
163         } while(--i);
164         block-=8*8;
165         fblock-=8*8+4;
166
167         load_matrix(odd_table);
168
169         i = 8;
170
171         do {
172                 float t0,t1,t2,t3;
173                 fr0 = block[1];
174                 fr1 = block[3];
175                 fr2 = block[5];
176                 fr3 = block[7];
177                 block+=8;
178                 ftrv();
179                 t0 = *fblock++;
180                 t1 = *fblock++;
181                 t2 = *fblock++;
182                 t3 = *fblock++;
183                 fblock+=4;
184                 *--fblock = t0 - fr0;
185                 *--fblock = t1 - fr1;
186                 *--fblock = t2 - fr2;
187                 *--fblock = t3 - fr3;
188                 *--fblock = t3 + fr3;
189                 *--fblock = t2 + fr2;
190                 *--fblock = t1 + fr1;
191                 *--fblock = t0 + fr0;
192                 fblock+=8;
193         } while(--i);
194         block-=8*8;
195         fblock-=8*8;
196
197         /* col */
198
199         /* even part */
200         load_matrix(even_table);
201
202         ofs1 = sizeof(float)*2*8;
203         ofs2 = sizeof(float)*4*8;
204         ofs3 = sizeof(float)*6*8;
205
206         i = 8;
207
208 #define        OA(fblock,ofs)   *(float*)((char*)fblock + ofs)
209
210         do {
211                 fr0 = OA(fblock,   0);
212                 fr1 = OA(fblock,ofs1);
213                 fr2 = OA(fblock,ofs2);
214                 fr3 = OA(fblock,ofs3);
215                 ftrv();
216                 OA(fblock,0   ) = fr0;
217                 OA(fblock,ofs1) = fr1;
218                 OA(fblock,ofs2) = fr2;
219                 OA(fblock,ofs3) = fr3;
220                 fblock++;
221         } while(--i);
222         fblock-=8;
223
224         load_matrix(odd_table);
225
226         i=8;
227         do {
228                 float t0,t1,t2,t3;
229                 t0 = OA(fblock,   0); /* [8*0] */
230                 t1 = OA(fblock,ofs1); /* [8*2] */
231                 t2 = OA(fblock,ofs2); /* [8*4] */
232                 t3 = OA(fblock,ofs3); /* [8*6] */
233                 fblock+=8;
234                 fr0 = OA(fblock,   0); /* [8*1] */
235                 fr1 = OA(fblock,ofs1); /* [8*3] */
236                 fr2 = OA(fblock,ofs2); /* [8*5] */
237                 fr3 = OA(fblock,ofs3); /* [8*7] */
238                 fblock+=-8+1;
239                 ftrv();
240                 block[8*0] = DESCALE(t0 + fr0,3);
241                 block[8*7] = DESCALE(t0 - fr0,3);
242                 block[8*1] = DESCALE(t1 + fr1,3);
243                 block[8*6] = DESCALE(t1 - fr1,3);
244                 block[8*2] = DESCALE(t2 + fr2,3);
245                 block[8*5] = DESCALE(t2 - fr2,3);
246                 block[8*3] = DESCALE(t3 + fr3,3);
247                 block[8*4] = DESCALE(t3 - fr3,3);
248                 block++;
249         } while(--i);
250
251 #if defined(__SH4__)
252         __asm__ ("fschg");
253 #endif
254 }
255 #else
256 void idct_sh4(DCTELEM *block)
257 {
258         DEFREG;
259
260         int i;
261         float   tblock[8*8],*fblock;
262
263         /* row */
264
265         /* even part */
266         load_matrix(even_table);
267
268         fblock = tblock;
269         i = 8;
270         do {
271                 fr0 = block[0];
272                 fr1 = block[2];
273                 fr2 = block[4];
274                 fr3 = block[6];
275                 block+=8;
276                 ftrv();
277                 fblock[0] = fr0;
278                 fblock[2] = fr1;
279                 fblock[4] = fr2;
280                 fblock[6] = fr3;
281                 fblock+=8;
282         } while(--i);
283         block-=8*8;
284         fblock-=8*8;
285
286         load_matrix(odd_table);
287
288         i = 8;
289
290         do {
291                 float t0,t1,t2,t3;
292                 fr0 = block[1];
293                 fr1 = block[3];
294                 fr2 = block[5];
295                 fr3 = block[7];
296                 block+=8;
297                 ftrv();
298                 t0 = fblock[0];
299                 t1 = fblock[2];
300                 t2 = fblock[4];
301                 t3 = fblock[6];
302                 fblock[0] = t0 + fr0;
303                 fblock[7] = t0 - fr0;
304                 fblock[1] = t1 + fr1;
305                 fblock[6] = t1 - fr1;
306                 fblock[2] = t2 + fr2;
307                 fblock[5] = t2 - fr2;
308                 fblock[3] = t3 + fr3;
309                 fblock[4] = t3 - fr3;
310                 fblock+=8;
311         } while(--i);
312         block-=8*8;
313         fblock-=8*8;
314
315         /* col */
316
317         /* even part */
318         load_matrix(even_table);
319
320         i = 8;
321
322         do {
323                 fr0 = fblock[8*0];
324                 fr1 = fblock[8*2];
325                 fr2 = fblock[8*4];
326                 fr3 = fblock[8*6];
327                 ftrv();
328                 fblock[8*0] = fr0;
329                 fblock[8*2] = fr1;
330                 fblock[8*4] = fr2;
331                 fblock[8*6] = fr3;
332                 fblock++;
333         } while(--i);
334         fblock-=8;
335
336         load_matrix(odd_table);
337
338         i=8;
339         do {
340                 float t0,t1,t2,t3;
341                 fr0 = fblock[8*1];
342                 fr1 = fblock[8*3];
343                 fr2 = fblock[8*5];
344                 fr3 = fblock[8*7];
345                 ftrv();
346                 t0 = fblock[8*0];
347                 t1 = fblock[8*2];
348                 t2 = fblock[8*4];
349                 t3 = fblock[8*6];
350                 fblock++;
351                 block[8*0] = DESCALE(t0 + fr0,3);
352                 block[8*7] = DESCALE(t0 - fr0,3);
353                 block[8*1] = DESCALE(t1 + fr1,3);
354                 block[8*6] = DESCALE(t1 - fr1,3);
355                 block[8*2] = DESCALE(t2 + fr2,3);
356                 block[8*5] = DESCALE(t2 - fr2,3);
357                 block[8*3] = DESCALE(t3 + fr3,3);
358                 block[8*4] = DESCALE(t3 - fr3,3);
359                 block++;
360         } while(--i);
361 }
362 #endif