FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
fft_mips.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012
3  * MIPS Technologies, Inc., California.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  * notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  * notice, this list of conditions and the following disclaimer in the
12  * documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14  * contributors may be used to endorse or promote products derived from
15  * this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * Author: Stanislav Ocovaj (socovaj@mips.com)
30  * Author: Zoran Lukic (zoranl@mips.com)
31  *
32  * Optimized MDCT/IMDCT and FFT transforms
33  *
34  * This file is part of FFmpeg.
35  *
36  * FFmpeg is free software; you can redistribute it and/or
37  * modify it under the terms of the GNU Lesser General Public
38  * License as published by the Free Software Foundation; either
39  * version 2.1 of the License, or (at your option) any later version.
40  *
41  * FFmpeg is distributed in the hope that it will be useful,
42  * but WITHOUT ANY WARRANTY; without even the implied warranty of
43  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
44  * Lesser General Public License for more details.
45  *
46  * You should have received a copy of the GNU Lesser General Public
47  * License along with FFmpeg; if not, write to the Free Software
48  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
49  */
50 #include "config.h"
51 #include "libavcodec/fft.h"
52 #include "libavcodec/fft_table.h"
53 #include "libavutil/mips/asmdefs.h"
54 
55 /**
56  * FFT transform
57  */
58 
59 #if HAVE_INLINE_ASM
60 static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
61 {
62  int nbits, i, n, num_transforms, offset, step;
63  int n4, n2, n34;
64  FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
65  FFTComplex *tmpz;
66  float w_re, w_im;
67  float *w_re_ptr, *w_im_ptr;
68  const int fft_size = (1 << s->nbits);
69  float pom, pom1, pom2, pom3;
70  float temp, temp1, temp3, temp4;
71  FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
72  FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
73 
74  num_transforms = (0x2aab >> (16 - s->nbits)) | 1;
75 
76  for (n=0; n<num_transforms; n++) {
77  offset = ff_fft_offsets_lut[n] << 2;
78  tmpz = z + offset;
79 
80  tmp1 = tmpz[0].re + tmpz[1].re;
81  tmp5 = tmpz[2].re + tmpz[3].re;
82  tmp2 = tmpz[0].im + tmpz[1].im;
83  tmp6 = tmpz[2].im + tmpz[3].im;
84  tmp3 = tmpz[0].re - tmpz[1].re;
85  tmp8 = tmpz[2].im - tmpz[3].im;
86  tmp4 = tmpz[0].im - tmpz[1].im;
87  tmp7 = tmpz[2].re - tmpz[3].re;
88 
89  tmpz[0].re = tmp1 + tmp5;
90  tmpz[2].re = tmp1 - tmp5;
91  tmpz[0].im = tmp2 + tmp6;
92  tmpz[2].im = tmp2 - tmp6;
93  tmpz[1].re = tmp3 + tmp8;
94  tmpz[3].re = tmp3 - tmp8;
95  tmpz[1].im = tmp4 - tmp7;
96  tmpz[3].im = tmp4 + tmp7;
97 
98  }
99 
100  if (fft_size < 8)
101  return;
102 
103  num_transforms = (num_transforms >> 1) | 1;
104 
105  for (n=0; n<num_transforms; n++) {
106  offset = ff_fft_offsets_lut[n] << 3;
107  tmpz = z + offset;
108 
109  __asm__ volatile (
110  "lwc1 %[tmp1], 32(%[tmpz]) \n\t"
111  "lwc1 %[pom], 40(%[tmpz]) \n\t"
112  "lwc1 %[tmp3], 48(%[tmpz]) \n\t"
113  "lwc1 %[pom1], 56(%[tmpz]) \n\t"
114  "lwc1 %[tmp2], 36(%[tmpz]) \n\t"
115  "lwc1 %[pom2], 44(%[tmpz]) \n\t"
116  "lwc1 %[pom3], 60(%[tmpz]) \n\t"
117  "lwc1 %[tmp4], 52(%[tmpz]) \n\t"
118  "add.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re + tmpz[5].re;
119  "add.s %[tmp3], %[tmp3], %[pom1] \n\t" // tmp3 = tmpz[6].re + tmpz[7].re;
120  "add.s %[tmp2], %[tmp2], %[pom2] \n\t" // tmp2 = tmpz[4].im + tmpz[5].im;
121  "lwc1 %[pom], 40(%[tmpz]) \n\t"
122  "add.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im + tmpz[7].im;
123  "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3;
124  "sub.s %[tmp7], %[tmp1], %[tmp3] \n\t" // tmp7 = tmp1 - tmp3;
125  "lwc1 %[tmp1], 32(%[tmpz]) \n\t"
126  "lwc1 %[pom1], 44(%[tmpz]) \n\t"
127  "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4;
128  "sub.s %[tmp8], %[tmp2], %[tmp4] \n\t" // tmp8 = tmp2 - tmp4;
129  "lwc1 %[tmp2], 36(%[tmpz]) \n\t"
130  "lwc1 %[pom2], 56(%[tmpz]) \n\t"
131  "lwc1 %[pom3], 60(%[tmpz]) \n\t"
132  "lwc1 %[tmp3], 48(%[tmpz]) \n\t"
133  "lwc1 %[tmp4], 52(%[tmpz]) \n\t"
134  "sub.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re - tmpz[5].re;
135  "lwc1 %[pom], 0(%[tmpz]) \n\t"
136  "sub.s %[tmp2], %[tmp2], %[pom1] \n\t" // tmp2 = tmpz[4].im - tmpz[5].im;
137  "sub.s %[tmp3], %[tmp3], %[pom2] \n\t" // tmp3 = tmpz[6].re - tmpz[7].re;
138  "lwc1 %[pom2], 4(%[tmpz]) \n\t"
139  "sub.s %[pom1], %[pom], %[tmp5] \n\t"
140  "sub.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im - tmpz[7].im;
141  "add.s %[pom3], %[pom], %[tmp5] \n\t"
142  "sub.s %[pom], %[pom2], %[tmp6] \n\t"
143  "add.s %[pom2], %[pom2], %[tmp6] \n\t"
144  "swc1 %[pom1], 32(%[tmpz]) \n\t" // tmpz[4].re = tmpz[0].re - tmp5;
145  "swc1 %[pom3], 0(%[tmpz]) \n\t" // tmpz[0].re = tmpz[0].re + tmp5;
146  "swc1 %[pom], 36(%[tmpz]) \n\t" // tmpz[4].im = tmpz[0].im - tmp6;
147  "swc1 %[pom2], 4(%[tmpz]) \n\t" // tmpz[0].im = tmpz[0].im + tmp6;
148  "lwc1 %[pom1], 16(%[tmpz]) \n\t"
149  "lwc1 %[pom3], 20(%[tmpz]) \n\t"
150  "li.s %[pom], 0.7071067812 \n\t" // float pom = 0.7071067812f;
151  "add.s %[temp1],%[tmp1], %[tmp2] \n\t"
152  "sub.s %[temp], %[pom1], %[tmp8] \n\t"
153  "add.s %[pom2], %[pom3], %[tmp7] \n\t"
154  "sub.s %[temp3],%[tmp3], %[tmp4] \n\t"
155  "sub.s %[temp4],%[tmp2], %[tmp1] \n\t"
156  "swc1 %[temp], 48(%[tmpz]) \n\t" // tmpz[6].re = tmpz[2].re - tmp8;
157  "swc1 %[pom2], 52(%[tmpz]) \n\t" // tmpz[6].im = tmpz[2].im + tmp7;
158  "add.s %[pom1], %[pom1], %[tmp8] \n\t"
159  "sub.s %[pom3], %[pom3], %[tmp7] \n\t"
160  "add.s %[tmp3], %[tmp3], %[tmp4] \n\t"
161  "mul.s %[tmp5], %[pom], %[temp1] \n\t" // tmp5 = pom * (tmp1 + tmp2);
162  "mul.s %[tmp7], %[pom], %[temp3] \n\t" // tmp7 = pom * (tmp3 - tmp4);
163  "mul.s %[tmp6], %[pom], %[temp4] \n\t" // tmp6 = pom * (tmp2 - tmp1);
164  "mul.s %[tmp8], %[pom], %[tmp3] \n\t" // tmp8 = pom * (tmp3 + tmp4);
165  "swc1 %[pom1], 16(%[tmpz]) \n\t" // tmpz[2].re = tmpz[2].re + tmp8;
166  "swc1 %[pom3], 20(%[tmpz]) \n\t" // tmpz[2].im = tmpz[2].im - tmp7;
167  "add.s %[tmp1], %[tmp5], %[tmp7] \n\t" // tmp1 = tmp5 + tmp7;
168  "sub.s %[tmp3], %[tmp5], %[tmp7] \n\t" // tmp3 = tmp5 - tmp7;
169  "add.s %[tmp2], %[tmp6], %[tmp8] \n\t" // tmp2 = tmp6 + tmp8;
170  "sub.s %[tmp4], %[tmp6], %[tmp8] \n\t" // tmp4 = tmp6 - tmp8;
171  "lwc1 %[temp], 8(%[tmpz]) \n\t"
172  "lwc1 %[temp1],12(%[tmpz]) \n\t"
173  "lwc1 %[pom], 24(%[tmpz]) \n\t"
174  "lwc1 %[pom2], 28(%[tmpz]) \n\t"
175  "sub.s %[temp4],%[temp], %[tmp1] \n\t"
176  "sub.s %[temp3],%[temp1], %[tmp2] \n\t"
177  "add.s %[temp], %[temp], %[tmp1] \n\t"
178  "add.s %[temp1],%[temp1], %[tmp2] \n\t"
179  "sub.s %[pom1], %[pom], %[tmp4] \n\t"
180  "add.s %[pom3], %[pom2], %[tmp3] \n\t"
181  "add.s %[pom], %[pom], %[tmp4] \n\t"
182  "sub.s %[pom2], %[pom2], %[tmp3] \n\t"
183  "swc1 %[temp4],40(%[tmpz]) \n\t" // tmpz[5].re = tmpz[1].re - tmp1;
184  "swc1 %[temp3],44(%[tmpz]) \n\t" // tmpz[5].im = tmpz[1].im - tmp2;
185  "swc1 %[temp], 8(%[tmpz]) \n\t" // tmpz[1].re = tmpz[1].re + tmp1;
186  "swc1 %[temp1],12(%[tmpz]) \n\t" // tmpz[1].im = tmpz[1].im + tmp2;
187  "swc1 %[pom1], 56(%[tmpz]) \n\t" // tmpz[7].re = tmpz[3].re - tmp4;
188  "swc1 %[pom3], 60(%[tmpz]) \n\t" // tmpz[7].im = tmpz[3].im + tmp3;
189  "swc1 %[pom], 24(%[tmpz]) \n\t" // tmpz[3].re = tmpz[3].re + tmp4;
190  "swc1 %[pom2], 28(%[tmpz]) \n\t" // tmpz[3].im = tmpz[3].im - tmp3;
191  : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
192  [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp7]"=&f"(tmp7),
193  [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
194  [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
195  : [tmpz]"r"(tmpz)
196  : "memory"
197  );
198  }
199 
200  step = 1 << (MAX_LOG2_NFFT - 4);
201  n4 = 4;
202 
203  for (nbits=4; nbits<=s->nbits; nbits++) {
204  num_transforms = (num_transforms >> 1) | 1;
205  n2 = 2 * n4;
206  n34 = 3 * n4;
207 
208  for (n=0; n<num_transforms; n++) {
209  offset = ff_fft_offsets_lut[n] << nbits;
210  tmpz = z + offset;
211 
212  tmpz_n2 = tmpz + n2;
213  tmpz_n4 = tmpz + n4;
214  tmpz_n34 = tmpz + n34;
215 
216  __asm__ volatile (
217  "lwc1 %[pom1], 0(%[tmpz_n2]) \n\t"
218  "lwc1 %[pom], 0(%[tmpz_n34]) \n\t"
219  "lwc1 %[pom2], 4(%[tmpz_n2]) \n\t"
220  "lwc1 %[pom3], 4(%[tmpz_n34]) \n\t"
221  "lwc1 %[temp1],0(%[tmpz]) \n\t"
222  "lwc1 %[temp3],4(%[tmpz]) \n\t"
223  "add.s %[tmp5], %[pom1], %[pom] \n\t" // tmp5 = tmpz[ n2].re + tmpz[n34].re;
224  "sub.s %[tmp1], %[pom1], %[pom] \n\t" // tmp1 = tmpz[ n2].re - tmpz[n34].re;
225  "add.s %[tmp6], %[pom2], %[pom3] \n\t" // tmp6 = tmpz[ n2].im + tmpz[n34].im;
226  "sub.s %[tmp2], %[pom2], %[pom3] \n\t" // tmp2 = tmpz[ n2].im - tmpz[n34].im;
227  "sub.s %[temp], %[temp1], %[tmp5] \n\t"
228  "add.s %[temp1],%[temp1], %[tmp5] \n\t"
229  "sub.s %[temp4],%[temp3], %[tmp6] \n\t"
230  "add.s %[temp3],%[temp3], %[tmp6] \n\t"
231  "swc1 %[temp], 0(%[tmpz_n2]) \n\t" // tmpz[ n2].re = tmpz[ 0].re - tmp5;
232  "swc1 %[temp1],0(%[tmpz]) \n\t" // tmpz[ 0].re = tmpz[ 0].re + tmp5;
233  "lwc1 %[pom1], 0(%[tmpz_n4]) \n\t"
234  "swc1 %[temp4],4(%[tmpz_n2]) \n\t" // tmpz[ n2].im = tmpz[ 0].im - tmp6;
235  "lwc1 %[temp], 4(%[tmpz_n4]) \n\t"
236  "swc1 %[temp3],4(%[tmpz]) \n\t" // tmpz[ 0].im = tmpz[ 0].im + tmp6;
237  "sub.s %[pom], %[pom1], %[tmp2] \n\t"
238  "add.s %[pom1], %[pom1], %[tmp2] \n\t"
239  "add.s %[temp1],%[temp], %[tmp1] \n\t"
240  "sub.s %[temp], %[temp], %[tmp1] \n\t"
241  "swc1 %[pom], 0(%[tmpz_n34]) \n\t" // tmpz[n34].re = tmpz[n4].re - tmp2;
242  "swc1 %[pom1], 0(%[tmpz_n4]) \n\t" // tmpz[ n4].re = tmpz[n4].re + tmp2;
243  "swc1 %[temp1],4(%[tmpz_n34]) \n\t" // tmpz[n34].im = tmpz[n4].im + tmp1;
244  "swc1 %[temp], 4(%[tmpz_n4]) \n\t" // tmpz[ n4].im = tmpz[n4].im - tmp1;
245  : [tmp5]"=&f"(tmp5),
246  [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
247  [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6), [pom3]"=&f"(pom3),
248  [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
249  : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
250  : "memory"
251  );
252 
253  w_re_ptr = (float*)(ff_cos_65536 + step);
254  w_im_ptr = (float*)(ff_cos_65536 + MAX_FFT_SIZE/4 - step);
255 
256  for (i=1; i<n4; i++) {
257  w_re = w_re_ptr[0];
258  w_im = w_im_ptr[0];
259  tmpz_n2_i = tmpz_n2 + i;
260  tmpz_n4_i = tmpz_n4 + i;
261  tmpz_n34_i= tmpz_n34 + i;
262  tmpz_i = tmpz + i;
263 
264  __asm__ volatile (
265  "lwc1 %[temp], 0(%[tmpz_n2_i]) \n\t"
266  "lwc1 %[temp1], 4(%[tmpz_n2_i]) \n\t"
267  "lwc1 %[pom], 0(%[tmpz_n34_i]) \n\t"
268  "lwc1 %[pom1], 4(%[tmpz_n34_i]) \n\t"
269  "mul.s %[temp3], %[w_im], %[temp] \n\t"
270  "mul.s %[temp4], %[w_im], %[temp1] \n\t"
271  "mul.s %[pom2], %[w_im], %[pom1] \n\t"
272  "mul.s %[pom3], %[w_im], %[pom] \n\t"
273  "msub.s %[tmp2], %[temp3], %[w_re], %[temp1] \n\t" // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
274  "madd.s %[tmp1], %[temp4], %[w_re], %[temp] \n\t" // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
275  "msub.s %[tmp3], %[pom2], %[w_re], %[pom] \n\t" // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
276  "madd.s %[tmp4], %[pom3], %[w_re], %[pom1] \n\t" // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
277  "lwc1 %[temp], 0(%[tmpz_i]) \n\t"
278  "lwc1 %[pom], 4(%[tmpz_i]) \n\t"
279  "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3;
280  "sub.s %[tmp1], %[tmp1], %[tmp3] \n\t" // tmp1 = tmp1 - tmp3;
281  "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4;
282  "sub.s %[tmp2], %[tmp2], %[tmp4] \n\t" // tmp2 = tmp2 - tmp4;
283  "sub.s %[temp1], %[temp], %[tmp5] \n\t"
284  "add.s %[temp], %[temp], %[tmp5] \n\t"
285  "sub.s %[pom1], %[pom], %[tmp6] \n\t"
286  "add.s %[pom], %[pom], %[tmp6] \n\t"
287  "lwc1 %[temp3], 0(%[tmpz_n4_i]) \n\t"
288  "lwc1 %[pom2], 4(%[tmpz_n4_i]) \n\t"
289  "swc1 %[temp1], 0(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].re = tmpz[ i].re - tmp5;
290  "swc1 %[temp], 0(%[tmpz_i]) \n\t" // tmpz[ i].re = tmpz[ i].re + tmp5;
291  "swc1 %[pom1], 4(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].im = tmpz[ i].im - tmp6;
292  "swc1 %[pom] , 4(%[tmpz_i]) \n\t" // tmpz[ i].im = tmpz[ i].im + tmp6;
293  "sub.s %[temp4], %[temp3], %[tmp2] \n\t"
294  "add.s %[pom3], %[pom2], %[tmp1] \n\t"
295  "add.s %[temp3], %[temp3], %[tmp2] \n\t"
296  "sub.s %[pom2], %[pom2], %[tmp1] \n\t"
297  "swc1 %[temp4], 0(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
298  "swc1 %[pom3], 4(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
299  "swc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
300  "swc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
301  : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
302  [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
303  [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
304  [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
305  : [w_re]"f"(w_re), [w_im]"f"(w_im),
306  [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
307  [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
308  : "memory"
309  );
310  w_re_ptr += step;
311  w_im_ptr -= step;
312  }
313  }
314  step >>= 1;
315  n4 <<= 1;
316  }
317 }
318 
319 /**
320  * MDCT/IMDCT transforms.
321  */
322 
323 static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
324 {
325  int k, n8, n4, n2, n, j;
326  const uint16_t *revtab = s->revtab;
327  const FFTSample *tcos = s->tcos;
328  const FFTSample *tsin = s->tsin;
329  const FFTSample *in1, *in2, *in3, *in4;
330  FFTComplex *z = (FFTComplex *)output;
331 
332  int j1;
333  const float *tcos1, *tsin1, *tcos2, *tsin2;
334  float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
335  temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
336  FFTComplex *z1, *z2;
337 
338  n = 1 << s->mdct_bits;
339  n2 = n >> 1;
340  n4 = n >> 2;
341  n8 = n >> 3;
342 
343  /* pre rotation */
344  in1 = input;
345  in2 = input + n2 - 1;
346  in3 = input + 2;
347  in4 = input + n2 - 3;
348 
349  tcos1 = tcos;
350  tsin1 = tsin;
351 
352  /* n4 = 64 or 128 */
353  for(k = 0; k < n4; k += 2) {
354  j = revtab[k ];
355  j1 = revtab[k + 1];
356 
357  __asm__ volatile (
358  "lwc1 %[temp1], 0(%[in2]) \t\n"
359  "lwc1 %[temp2], 0(%[tcos1]) \t\n"
360  "lwc1 %[temp3], 0(%[tsin1]) \t\n"
361  "lwc1 %[temp4], 0(%[in1]) \t\n"
362  "lwc1 %[temp5], 0(%[in4]) \t\n"
363  "mul.s %[temp9], %[temp1], %[temp2] \t\n"
364  "mul.s %[temp10], %[temp1], %[temp3] \t\n"
365  "lwc1 %[temp6], 4(%[tcos1]) \t\n"
366  "lwc1 %[temp7], 4(%[tsin1]) \t\n"
367  "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n"
368  "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n"
369  "mul.s %[temp11], %[temp5], %[temp6] \t\n"
370  "mul.s %[temp12], %[temp5], %[temp7] \t\n"
371  "lwc1 %[temp8], 0(%[in3]) \t\n"
372  PTR_ADDIU " %[tcos1], %[tcos1], 8 \t\n"
373  PTR_ADDIU " %[tsin1], %[tsin1], 8 \t\n"
374  PTR_ADDIU " %[in1], %[in1], 16 \t\n"
375  "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n"
376  "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n"
377  PTR_ADDIU " %[in2], %[in2], -16 \t\n"
378  PTR_ADDIU " %[in3], %[in3], 16 \t\n"
379  PTR_ADDIU " %[in4], %[in4], -16 \t\n"
380 
381  : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
382  [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
383  [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
384  [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
385  [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
386  [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
387  [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
388  [in1]"+r"(in1), [in2]"+r"(in2),
389  [in3]"+r"(in3), [in4]"+r"(in4)
390  :
391  : "memory"
392  );
393 
394  z[j ].re = temp9;
395  z[j ].im = temp10;
396  z[j1].re = temp11;
397  z[j1].im = temp12;
398  }
399 
400  s->fft_calc(s, z);
401 
402  /* post rotation + reordering */
403  /* n8 = 32 or 64 */
404  for(k = 0; k < n8; k += 2) {
405  tcos1 = &tcos[n8 - k - 2];
406  tsin1 = &tsin[n8 - k - 2];
407  tcos2 = &tcos[n8 + k];
408  tsin2 = &tsin[n8 + k];
409  z1 = &z[n8 - k - 2];
410  z2 = &z[n8 + k ];
411 
412  __asm__ volatile (
413  "lwc1 %[temp1], 12(%[z1]) \t\n"
414  "lwc1 %[temp2], 4(%[tsin1]) \t\n"
415  "lwc1 %[temp3], 4(%[tcos1]) \t\n"
416  "lwc1 %[temp4], 8(%[z1]) \t\n"
417  "lwc1 %[temp5], 4(%[z1]) \t\n"
418  "mul.s %[temp9], %[temp1], %[temp2] \t\n"
419  "mul.s %[temp10], %[temp1], %[temp3] \t\n"
420  "lwc1 %[temp6], 0(%[tsin1]) \t\n"
421  "lwc1 %[temp7], 0(%[tcos1]) \t\n"
422  "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n"
423  "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n"
424  "mul.s %[temp11], %[temp5], %[temp6] \t\n"
425  "mul.s %[temp12], %[temp5], %[temp7] \t\n"
426  "lwc1 %[temp8], 0(%[z1]) \t\n"
427  "lwc1 %[temp1], 4(%[z2]) \t\n"
428  "lwc1 %[temp2], 0(%[tsin2]) \t\n"
429  "lwc1 %[temp3], 0(%[tcos2]) \t\n"
430  "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n"
431  "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n"
432  "mul.s %[temp13], %[temp1], %[temp2] \t\n"
433  "mul.s %[temp14], %[temp1], %[temp3] \t\n"
434  "lwc1 %[temp4], 0(%[z2]) \t\n"
435  "lwc1 %[temp5], 12(%[z2]) \t\n"
436  "lwc1 %[temp6], 4(%[tsin2]) \t\n"
437  "lwc1 %[temp7], 4(%[tcos2]) \t\n"
438  "nmsub.s %[temp13], %[temp13], %[temp4], %[temp3] \t\n"
439  "madd.s %[temp14], %[temp14], %[temp4], %[temp2] \t\n"
440  "mul.s %[temp15], %[temp5], %[temp6] \t\n"
441  "mul.s %[temp16], %[temp5], %[temp7] \t\n"
442  "lwc1 %[temp8], 8(%[z2]) \t\n"
443  "nmsub.s %[temp15], %[temp15], %[temp8], %[temp7] \t\n"
444  "madd.s %[temp16], %[temp16], %[temp8], %[temp6] \t\n"
445  : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
446  [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
447  [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
448  [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
449  [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
450  [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
451  [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
452  [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
453  : [z1]"r"(z1), [z2]"r"(z2),
454  [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
455  [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
456  : "memory"
457  );
458 
459  z1[1].re = temp9;
460  z1[1].im = temp14;
461  z2[0].re = temp13;
462  z2[0].im = temp10;
463 
464  z1[0].re = temp11;
465  z1[0].im = temp16;
466  z2[1].re = temp15;
467  z2[1].im = temp12;
468  }
469 }
470 
471 /**
472  * Compute inverse MDCT of size N = 2^nbits
473  * @param output N samples
474  * @param input N/2 samples
475  */
476 static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
477 {
478  int k;
479  int n = 1 << s->mdct_bits;
480  int n2 = n >> 1;
481  int n4 = n >> 2;
482 
483  ff_imdct_half_mips(s, output+n4, input);
484 
485  for(k = 0; k < n4; k+=4) {
486  output[k] = -output[n2-k-1];
487  output[k+1] = -output[n2-k-2];
488  output[k+2] = -output[n2-k-3];
489  output[k+3] = -output[n2-k-4];
490 
491  output[n-k-1] = output[n2+k];
492  output[n-k-2] = output[n2+k+1];
493  output[n-k-3] = output[n2+k+2];
494  output[n-k-4] = output[n2+k+3];
495  }
496 }
497 #endif /* HAVE_INLINE_ASM */
498 
500 {
501  int n=0;
502 
503  ff_fft_lut_init(ff_fft_offsets_lut, 0, 1 << 16, &n);
505 
506 #if HAVE_INLINE_ASM
507  s->fft_calc = ff_fft_calc_mips;
508 #if CONFIG_MDCT
509  s->imdct_calc = ff_imdct_calc_mips;
510  s->imdct_half = ff_imdct_half_mips;
511 #endif
512 #endif
513 }