• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
ff_fft_calc_mips(FFTContext * s,FFTComplex * z)61 static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
62 {
63     int nbits, i, n, num_transforms, offset, step;
64     int n4, n2, n34;
65     FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
66     FFTComplex *tmpz;
67     float w_re, w_im;
68     float *w_re_ptr, *w_im_ptr;
69     const int fft_size = (1 << s->nbits);
70     float pom,  pom1,  pom2,  pom3;
71     float temp, temp1, temp3, temp4;
72     FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
73     FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
74     float f1 = 0.7071067812;
75 
76     num_transforms = (21845 >> (17 - s->nbits)) | 1;
77 
78     for (n=0; n<num_transforms; n++) {
79         offset = ff_fft_offsets_lut[n] << 2;
80         tmpz = z + offset;
81 
82         tmp1 = tmpz[0].re + tmpz[1].re;
83         tmp5 = tmpz[2].re + tmpz[3].re;
84         tmp2 = tmpz[0].im + tmpz[1].im;
85         tmp6 = tmpz[2].im + tmpz[3].im;
86         tmp3 = tmpz[0].re - tmpz[1].re;
87         tmp8 = tmpz[2].im - tmpz[3].im;
88         tmp4 = tmpz[0].im - tmpz[1].im;
89         tmp7 = tmpz[2].re - tmpz[3].re;
90 
91         tmpz[0].re = tmp1 + tmp5;
92         tmpz[2].re = tmp1 - tmp5;
93         tmpz[0].im = tmp2 + tmp6;
94         tmpz[2].im = tmp2 - tmp6;
95         tmpz[1].re = tmp3 + tmp8;
96         tmpz[3].re = tmp3 - tmp8;
97         tmpz[1].im = tmp4 - tmp7;
98         tmpz[3].im = tmp4 + tmp7;
99 
100     }
101 
102     if (fft_size < 8)
103         return;
104 
105     num_transforms = (num_transforms >> 1) | 1;
106 
107     for (n=0; n<num_transforms; n++) {
108         offset = ff_fft_offsets_lut[n] << 3;
109         tmpz = z + offset;
110 
111         __asm__ volatile (
112             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
113             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
114             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
115             "lwc1  %[pom1], 56(%[tmpz])                     \n\t"
116             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
117             "lwc1  %[pom2], 44(%[tmpz])                     \n\t"
118             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
119             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
120             "add.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re + tmpz[5].re;
121             "add.s %[tmp3], %[tmp3],    %[pom1]             \n\t"  // tmp3 = tmpz[6].re + tmpz[7].re;
122             "add.s %[tmp2], %[tmp2],    %[pom2]             \n\t"  // tmp2 = tmpz[4].im + tmpz[5].im;
123             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
124             "add.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im + tmpz[7].im;
125             "add.s %[tmp5], %[tmp1],    %[tmp3]             \n\t"  // tmp5 = tmp1 + tmp3;
126             "sub.s %[tmp7], %[tmp1],    %[tmp3]             \n\t"  // tmp7 = tmp1 - tmp3;
127             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
128             "lwc1  %[pom1], 44(%[tmpz])                     \n\t"
129             "add.s %[tmp6], %[tmp2],    %[tmp4]             \n\t"  // tmp6 = tmp2 + tmp4;
130             "sub.s %[tmp8], %[tmp2],    %[tmp4]             \n\t"  // tmp8 = tmp2 - tmp4;
131             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
132             "lwc1  %[pom2], 56(%[tmpz])                     \n\t"
133             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
134             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
135             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
136             "sub.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re - tmpz[5].re;
137             "lwc1  %[pom],  0(%[tmpz])                      \n\t"
138             "sub.s %[tmp2], %[tmp2],    %[pom1]             \n\t"  // tmp2 = tmpz[4].im - tmpz[5].im;
139             "sub.s %[tmp3], %[tmp3],    %[pom2]             \n\t"  // tmp3 = tmpz[6].re - tmpz[7].re;
140             "lwc1  %[pom2], 4(%[tmpz])                      \n\t"
141             "sub.s %[pom1], %[pom],     %[tmp5]             \n\t"
142             "sub.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im - tmpz[7].im;
143             "add.s %[pom3], %[pom],     %[tmp5]             \n\t"
144             "sub.s %[pom],  %[pom2],    %[tmp6]             \n\t"
145             "add.s %[pom2], %[pom2],    %[tmp6]             \n\t"
146             "swc1  %[pom1], 32(%[tmpz])                     \n\t"  // tmpz[4].re = tmpz[0].re - tmp5;
147             "swc1  %[pom3], 0(%[tmpz])                      \n\t"  // tmpz[0].re = tmpz[0].re + tmp5;
148             "swc1  %[pom],  36(%[tmpz])                     \n\t"  // tmpz[4].im = tmpz[0].im - tmp6;
149             "swc1  %[pom2], 4(%[tmpz])                      \n\t"  // tmpz[0].im = tmpz[0].im + tmp6;
150             "lwc1  %[pom1], 16(%[tmpz])                     \n\t"
151             "lwc1  %[pom3], 20(%[tmpz])                     \n\t"
152             "add.s %[temp1],%[tmp1],    %[tmp2]             \n\t"
153             "sub.s %[temp], %[pom1],    %[tmp8]             \n\t"
154             "add.s %[pom2], %[pom3],    %[tmp7]             \n\t"
155             "sub.s %[temp3],%[tmp3],    %[tmp4]             \n\t"
156             "sub.s %[temp4],%[tmp2],    %[tmp1]             \n\t"
157             "swc1  %[temp], 48(%[tmpz])                     \n\t"  // tmpz[6].re = tmpz[2].re - tmp8;
158             "swc1  %[pom2], 52(%[tmpz])                     \n\t"  // tmpz[6].im = tmpz[2].im + tmp7;
159             "add.s %[pom1], %[pom1],    %[tmp8]             \n\t"
160             "sub.s %[pom3], %[pom3],    %[tmp7]             \n\t"
161             "add.s %[tmp3], %[tmp3],    %[tmp4]             \n\t"
162             "mul.s %[tmp5], %[f1],      %[temp1]            \n\t"  // tmp5 = pom * (tmp1 + tmp2);
163             "mul.s %[tmp7], %[f1],      %[temp3]            \n\t"  // tmp7 = pom * (tmp3 - tmp4);
164             "mul.s %[tmp6], %[f1],      %[temp4]            \n\t"  // tmp6 = pom * (tmp2 - tmp1);
165             "mul.s %[tmp8], %[f1],      %[tmp3]             \n\t"  // tmp8 = pom * (tmp3 + tmp4);
166             "swc1  %[pom1], 16(%[tmpz])                     \n\t"  // tmpz[2].re = tmpz[2].re + tmp8;
167             "swc1  %[pom3], 20(%[tmpz])                     \n\t"  // tmpz[2].im = tmpz[2].im - tmp7;
168             "add.s %[tmp1], %[tmp5],    %[tmp7]             \n\t"  // tmp1 = tmp5 + tmp7;
169             "sub.s %[tmp3], %[tmp5],    %[tmp7]             \n\t"  // tmp3 = tmp5 - tmp7;
170             "add.s %[tmp2], %[tmp6],    %[tmp8]             \n\t"  // tmp2 = tmp6 + tmp8;
171             "sub.s %[tmp4], %[tmp6],    %[tmp8]             \n\t"  // tmp4 = tmp6 - tmp8;
172             "lwc1  %[temp], 8(%[tmpz])                      \n\t"
173             "lwc1  %[temp1],12(%[tmpz])                     \n\t"
174             "lwc1  %[pom],  24(%[tmpz])                     \n\t"
175             "lwc1  %[pom2], 28(%[tmpz])                     \n\t"
176             "sub.s %[temp4],%[temp],    %[tmp1]             \n\t"
177             "sub.s %[temp3],%[temp1],   %[tmp2]             \n\t"
178             "add.s %[temp], %[temp],    %[tmp1]             \n\t"
179             "add.s %[temp1],%[temp1],   %[tmp2]             \n\t"
180             "sub.s %[pom1], %[pom],     %[tmp4]             \n\t"
181             "add.s %[pom3], %[pom2],    %[tmp3]             \n\t"
182             "add.s %[pom],  %[pom],     %[tmp4]             \n\t"
183             "sub.s %[pom2], %[pom2],    %[tmp3]             \n\t"
184             "swc1  %[temp4],40(%[tmpz])                     \n\t"  // tmpz[5].re = tmpz[1].re - tmp1;
185             "swc1  %[temp3],44(%[tmpz])                     \n\t"  // tmpz[5].im = tmpz[1].im - tmp2;
186             "swc1  %[temp], 8(%[tmpz])                      \n\t"  // tmpz[1].re = tmpz[1].re + tmp1;
187             "swc1  %[temp1],12(%[tmpz])                     \n\t"  // tmpz[1].im = tmpz[1].im + tmp2;
188             "swc1  %[pom1], 56(%[tmpz])                     \n\t"  // tmpz[7].re = tmpz[3].re - tmp4;
189             "swc1  %[pom3], 60(%[tmpz])                     \n\t"  // tmpz[7].im = tmpz[3].im + tmp3;
190             "swc1  %[pom],  24(%[tmpz])                     \n\t"  // tmpz[3].re = tmpz[3].re + tmp4;
191             "swc1  %[pom2], 28(%[tmpz])                     \n\t"  // tmpz[3].im = tmpz[3].im - tmp3;
192             : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),   [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
193               [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5),  [tmp7]"=&f"(tmp7),
194               [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
195               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
196             : [tmpz]"r"(tmpz), [f1]"f"(f1)
197             : "memory"
198         );
199     }
200 
201     step = 1 << (MAX_LOG2_NFFT - 4);
202     n4 = 4;
203 
204     for (nbits=4; nbits<=s->nbits; nbits++) {
205         num_transforms = (num_transforms >> 1) | 1;
206         n2  = 2 * n4;
207         n34 = 3 * n4;
208 
209         for (n=0; n<num_transforms; n++) {
210             offset = ff_fft_offsets_lut[n] << nbits;
211             tmpz = z + offset;
212 
213             tmpz_n2  = tmpz +  n2;
214             tmpz_n4  = tmpz +  n4;
215             tmpz_n34 = tmpz +  n34;
216 
217             __asm__ volatile (
218                 "lwc1  %[pom1], 0(%[tmpz_n2])            \n\t"
219                 "lwc1  %[pom],  0(%[tmpz_n34])           \n\t"
220                 "lwc1  %[pom2], 4(%[tmpz_n2])            \n\t"
221                 "lwc1  %[pom3], 4(%[tmpz_n34])           \n\t"
222                 "lwc1  %[temp1],0(%[tmpz])               \n\t"
223                 "lwc1  %[temp3],4(%[tmpz])               \n\t"
224                 "add.s %[tmp5], %[pom1],      %[pom]     \n\t"   //  tmp5 = tmpz[ n2].re + tmpz[n34].re;
225                 "sub.s %[tmp1], %[pom1],      %[pom]     \n\t"   //  tmp1 = tmpz[ n2].re - tmpz[n34].re;
226                 "add.s %[tmp6], %[pom2],      %[pom3]    \n\t"   //  tmp6 = tmpz[ n2].im + tmpz[n34].im;
227                 "sub.s %[tmp2], %[pom2],      %[pom3]    \n\t"   //  tmp2 = tmpz[ n2].im - tmpz[n34].im;
228                 "sub.s %[temp], %[temp1],     %[tmp5]    \n\t"
229                 "add.s %[temp1],%[temp1],     %[tmp5]    \n\t"
230                 "sub.s %[temp4],%[temp3],     %[tmp6]    \n\t"
231                 "add.s %[temp3],%[temp3],     %[tmp6]    \n\t"
232                 "swc1  %[temp], 0(%[tmpz_n2])            \n\t"   //  tmpz[ n2].re = tmpz[ 0].re - tmp5;
233                 "swc1  %[temp1],0(%[tmpz])               \n\t"   //  tmpz[  0].re = tmpz[ 0].re + tmp5;
234                 "lwc1  %[pom1], 0(%[tmpz_n4])            \n\t"
235                 "swc1  %[temp4],4(%[tmpz_n2])            \n\t"   //  tmpz[ n2].im = tmpz[ 0].im - tmp6;
236                 "lwc1  %[temp], 4(%[tmpz_n4])            \n\t"
237                 "swc1  %[temp3],4(%[tmpz])               \n\t"   //  tmpz[  0].im = tmpz[ 0].im + tmp6;
238                 "sub.s %[pom],  %[pom1],      %[tmp2]    \n\t"
239                 "add.s %[pom1], %[pom1],      %[tmp2]    \n\t"
240                 "add.s %[temp1],%[temp],      %[tmp1]    \n\t"
241                 "sub.s %[temp], %[temp],      %[tmp1]    \n\t"
242                 "swc1  %[pom],  0(%[tmpz_n34])           \n\t"   //  tmpz[n34].re = tmpz[n4].re - tmp2;
243                 "swc1  %[pom1], 0(%[tmpz_n4])            \n\t"   //  tmpz[ n4].re = tmpz[n4].re + tmp2;
244                 "swc1  %[temp1],4(%[tmpz_n34])           \n\t"   //  tmpz[n34].im = tmpz[n4].im + tmp1;
245                 "swc1  %[temp], 4(%[tmpz_n4])            \n\t"   //  tmpz[ n4].im = tmpz[n4].im - tmp1;
246                 : [tmp5]"=&f"(tmp5),
247                   [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),        [pom1]"=&f"(pom1),        [pom2]"=&f"(pom2),
248                   [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6),          [pom3]"=&f"(pom3),
249                   [temp]"=&f"(temp), [temp1]"=&f"(temp1),     [temp3]"=&f"(temp3),       [temp4]"=&f"(temp4)
250                 : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
251                 : "memory"
252             );
253 
254             w_re_ptr = (float*)(ff_cos_131072 + step);
255             w_im_ptr = (float*)(ff_cos_131072 + MAX_FFT_SIZE/4 - step);
256 
257             for (i=1; i<n4; i++) {
258                 w_re = w_re_ptr[0];
259                 w_im = w_im_ptr[0];
260                 tmpz_n2_i = tmpz_n2  + i;
261                 tmpz_n4_i = tmpz_n4  + i;
262                 tmpz_n34_i= tmpz_n34 + i;
263                 tmpz_i    = tmpz     + i;
264 
265                 __asm__ volatile (
266                     "lwc1     %[temp],  0(%[tmpz_n2_i])               \n\t"
267                     "lwc1     %[temp1], 4(%[tmpz_n2_i])               \n\t"
268                     "lwc1     %[pom],   0(%[tmpz_n34_i])              \n\t"
269                     "lwc1     %[pom1],  4(%[tmpz_n34_i])              \n\t"
270                     "mul.s    %[temp3], %[w_im],    %[temp]           \n\t"
271                     "mul.s    %[temp4], %[w_im],    %[temp1]          \n\t"
272                     "mul.s    %[pom2],  %[w_im],    %[pom1]           \n\t"
273                     "mul.s    %[pom3],  %[w_im],    %[pom]            \n\t"
274                     "msub.s   %[tmp2],  %[temp3],   %[w_re], %[temp1] \n\t"  // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
275                     "madd.s   %[tmp1],  %[temp4],   %[w_re], %[temp]  \n\t"  // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
276                     "msub.s   %[tmp3],  %[pom2],    %[w_re], %[pom]   \n\t"  // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
277                     "madd.s   %[tmp4],  %[pom3],    %[w_re], %[pom1]  \n\t"  // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
278                     "lwc1     %[temp],  0(%[tmpz_i])                  \n\t"
279                     "lwc1     %[pom],   4(%[tmpz_i])                  \n\t"
280                     "add.s    %[tmp5],  %[tmp1],    %[tmp3]           \n\t"  // tmp5 = tmp1 + tmp3;
281                     "sub.s    %[tmp1],  %[tmp1],    %[tmp3]           \n\t"  // tmp1 = tmp1 - tmp3;
282                     "add.s    %[tmp6],  %[tmp2],    %[tmp4]           \n\t"  // tmp6 = tmp2 + tmp4;
283                     "sub.s    %[tmp2],  %[tmp2],    %[tmp4]           \n\t"  // tmp2 = tmp2 - tmp4;
284                     "sub.s    %[temp1], %[temp],    %[tmp5]           \n\t"
285                     "add.s    %[temp],  %[temp],    %[tmp5]           \n\t"
286                     "sub.s    %[pom1],  %[pom],     %[tmp6]           \n\t"
287                     "add.s    %[pom],   %[pom],     %[tmp6]           \n\t"
288                     "lwc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"
289                     "lwc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"
290                     "swc1     %[temp1], 0(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].re = tmpz[   i].re - tmp5;
291                     "swc1     %[temp],  0(%[tmpz_i])                  \n\t"  // tmpz[    i].re = tmpz[   i].re + tmp5;
292                     "swc1     %[pom1],  4(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].im = tmpz[   i].im - tmp6;
293                     "swc1     %[pom] ,  4(%[tmpz_i])                  \n\t"  // tmpz[    i].im = tmpz[   i].im + tmp6;
294                     "sub.s    %[temp4], %[temp3],   %[tmp2]           \n\t"
295                     "add.s    %[pom3],  %[pom2],    %[tmp1]           \n\t"
296                     "add.s    %[temp3], %[temp3],   %[tmp2]           \n\t"
297                     "sub.s    %[pom2],  %[pom2],    %[tmp1]           \n\t"
298                     "swc1     %[temp4], 0(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
299                     "swc1     %[pom3],  4(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
300                     "swc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
301                     "swc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
302                     : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
303                       [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
304                       [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
305                       [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
306                     : [w_re]"f"(w_re), [w_im]"f"(w_im),
307                       [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
308                       [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
309                     : "memory"
310                 );
311                 w_re_ptr += step;
312                 w_im_ptr -= step;
313             }
314         }
315         step >>= 1;
316         n4   <<= 1;
317     }
318 }
319 
320 /**
321  * MDCT/IMDCT transforms.
322  */
323 
ff_imdct_half_mips(FFTContext * s,FFTSample * output,const FFTSample * input)324 static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
325 {
326     int k, n8, n4, n2, n, j;
327     const uint16_t *revtab = s->revtab;
328     const FFTSample *tcos = s->tcos;
329     const FFTSample *tsin = s->tsin;
330     const FFTSample *in1, *in2, *in3, *in4;
331     FFTComplex *z = (FFTComplex *)output;
332 
333     int j1;
334     const float *tcos1, *tsin1, *tcos2, *tsin2;
335     float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
336         temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
337     FFTComplex *z1, *z2;
338 
339     n = 1 << s->mdct_bits;
340     n2 = n >> 1;
341     n4 = n >> 2;
342     n8 = n >> 3;
343 
344     /* pre rotation */
345     in1 = input;
346     in2 = input + n2 - 1;
347     in3 = input + 2;
348     in4 = input + n2 - 3;
349 
350     tcos1 = tcos;
351     tsin1 = tsin;
352 
353     /* n4 = 64 or 128 */
354     for(k = 0; k < n4; k += 2) {
355         j  = revtab[k    ];
356         j1 = revtab[k + 1];
357 
358         __asm__ volatile (
359             "lwc1           %[temp1],       0(%[in2])                           \t\n"
360             "lwc1           %[temp2],       0(%[tcos1])                         \t\n"
361             "lwc1           %[temp3],       0(%[tsin1])                         \t\n"
362             "lwc1           %[temp4],       0(%[in1])                           \t\n"
363             "lwc1           %[temp5],       0(%[in4])                           \t\n"
364             "mul.s          %[temp9],       %[temp1],   %[temp2]                \t\n"
365             "mul.s          %[temp10],      %[temp1],   %[temp3]                \t\n"
366             "lwc1           %[temp6],       4(%[tcos1])                         \t\n"
367             "lwc1           %[temp7],       4(%[tsin1])                         \t\n"
368             "nmsub.s        %[temp9],       %[temp9],   %[temp4],   %[temp3]    \t\n"
369             "madd.s         %[temp10],      %[temp10],  %[temp4],   %[temp2]    \t\n"
370             "mul.s          %[temp11],      %[temp5],   %[temp6]                \t\n"
371             "mul.s          %[temp12],      %[temp5],   %[temp7]                \t\n"
372             "lwc1           %[temp8],       0(%[in3])                           \t\n"
373             PTR_ADDIU "     %[tcos1],       %[tcos1],   8                       \t\n"
374             PTR_ADDIU "     %[tsin1],       %[tsin1],   8                       \t\n"
375             PTR_ADDIU "     %[in1],         %[in1],     16                      \t\n"
376             "nmsub.s        %[temp11],      %[temp11],  %[temp8],   %[temp7]    \t\n"
377             "madd.s         %[temp12],      %[temp12],  %[temp8],   %[temp6]    \t\n"
378             PTR_ADDIU "     %[in2],         %[in2],     -16                     \t\n"
379             PTR_ADDIU "     %[in3],         %[in3],     16                      \t\n"
380             PTR_ADDIU "     %[in4],         %[in4],     -16                     \t\n"
381 
382             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
383               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
384               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
385               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
386               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
387               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
388               [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
389               [in1]"+r"(in1), [in2]"+r"(in2),
390               [in3]"+r"(in3), [in4]"+r"(in4)
391             :
392             : "memory"
393         );
394 
395         z[j ].re = temp9;
396         z[j ].im = temp10;
397         z[j1].re = temp11;
398         z[j1].im = temp12;
399     }
400 
401     s->fft_calc(s, z);
402 
403     /* post rotation + reordering */
404     /* n8 = 32 or 64 */
405     for(k = 0; k < n8; k += 2) {
406         tcos1 = &tcos[n8 - k - 2];
407         tsin1 = &tsin[n8 - k - 2];
408         tcos2 = &tcos[n8 + k];
409         tsin2 = &tsin[n8 + k];
410         z1 = &z[n8 - k - 2];
411         z2 = &z[n8 + k    ];
412 
413         __asm__ volatile (
414             "lwc1       %[temp1],   12(%[z1])                           \t\n"
415             "lwc1       %[temp2],   4(%[tsin1])                         \t\n"
416             "lwc1       %[temp3],   4(%[tcos1])                         \t\n"
417             "lwc1       %[temp4],   8(%[z1])                            \t\n"
418             "lwc1       %[temp5],   4(%[z1])                            \t\n"
419             "mul.s      %[temp9],   %[temp1],   %[temp2]                \t\n"
420             "mul.s      %[temp10],  %[temp1],   %[temp3]                \t\n"
421             "lwc1       %[temp6],   0(%[tsin1])                         \t\n"
422             "lwc1       %[temp7],   0(%[tcos1])                         \t\n"
423             "nmsub.s    %[temp9],   %[temp9],   %[temp4],   %[temp3]    \t\n"
424             "madd.s     %[temp10],  %[temp10],  %[temp4],   %[temp2]    \t\n"
425             "mul.s      %[temp11],  %[temp5],   %[temp6]                \t\n"
426             "mul.s      %[temp12],  %[temp5],   %[temp7]                \t\n"
427             "lwc1       %[temp8],   0(%[z1])                            \t\n"
428             "lwc1       %[temp1],   4(%[z2])                            \t\n"
429             "lwc1       %[temp2],   0(%[tsin2])                         \t\n"
430             "lwc1       %[temp3],   0(%[tcos2])                         \t\n"
431             "nmsub.s    %[temp11],  %[temp11],  %[temp8],   %[temp7]    \t\n"
432             "madd.s     %[temp12],  %[temp12],  %[temp8],   %[temp6]    \t\n"
433             "mul.s      %[temp13],  %[temp1],   %[temp2]                \t\n"
434             "mul.s      %[temp14],  %[temp1],   %[temp3]                \t\n"
435             "lwc1       %[temp4],   0(%[z2])                            \t\n"
436             "lwc1       %[temp5],   12(%[z2])                           \t\n"
437             "lwc1       %[temp6],   4(%[tsin2])                         \t\n"
438             "lwc1       %[temp7],   4(%[tcos2])                         \t\n"
439             "nmsub.s    %[temp13],  %[temp13],  %[temp4],   %[temp3]    \t\n"
440             "madd.s     %[temp14],  %[temp14],  %[temp4],   %[temp2]    \t\n"
441             "mul.s      %[temp15],  %[temp5],   %[temp6]                \t\n"
442             "mul.s      %[temp16],  %[temp5],   %[temp7]                \t\n"
443             "lwc1       %[temp8],   8(%[z2])                            \t\n"
444             "nmsub.s    %[temp15],  %[temp15],  %[temp8],   %[temp7]    \t\n"
445             "madd.s     %[temp16],  %[temp16],  %[temp8],   %[temp6]    \t\n"
446             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
447               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
448               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
449               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
450               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
451               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
452               [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
453               [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
454             : [z1]"r"(z1), [z2]"r"(z2),
455               [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
456               [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
457             : "memory"
458         );
459 
460         z1[1].re = temp9;
461         z1[1].im = temp14;
462         z2[0].re = temp13;
463         z2[0].im = temp10;
464 
465         z1[0].re = temp11;
466         z1[0].im = temp16;
467         z2[1].re = temp15;
468         z2[1].im = temp12;
469     }
470 }
471 
472 /**
473  * Compute inverse MDCT of size N = 2^nbits
474  * @param output N samples
475  * @param input N/2 samples
476  */
ff_imdct_calc_mips(FFTContext * s,FFTSample * output,const FFTSample * input)477 static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
478 {
479     int k;
480     int n = 1 << s->mdct_bits;
481     int n2 = n >> 1;
482     int n4 = n >> 2;
483 
484     ff_imdct_half_mips(s, output+n4, input);
485 
486     for(k = 0; k < n4; k+=4) {
487         output[k] = -output[n2-k-1];
488         output[k+1] = -output[n2-k-2];
489         output[k+2] = -output[n2-k-3];
490         output[k+3] = -output[n2-k-4];
491 
492         output[n-k-1] = output[n2+k];
493         output[n-k-2] = output[n2+k+1];
494         output[n-k-3] = output[n2+k+2];
495         output[n-k-4] = output[n2+k+3];
496     }
497 }
498 #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
499 #endif /* HAVE_INLINE_ASM */
500 
ff_fft_init_mips(FFTContext * s)501 av_cold void ff_fft_init_mips(FFTContext *s)
502 {
503     ff_fft_lut_init();
504     ff_init_ff_cos_tabs(17);
505 
506 #if HAVE_INLINE_ASM
507 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
508     s->fft_calc     = ff_fft_calc_mips;
509 #if CONFIG_MDCT
510     s->imdct_calc   = ff_imdct_calc_mips;
511     s->imdct_half   = ff_imdct_half_mips;
512 #endif
513 #endif
514 #endif
515 }
516