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 "libavutil/attributes.h"
52#include "libavcodec/fft.h"
53#include "libavcodec/fft_table.h"
54#include "libavutil/mips/asmdefs.h"
55
56/**
57 * FFT transform
58 */
59
60#if HAVE_INLINE_ASM
61#if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
62static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
63{
64    int nbits, i, n, num_transforms, offset, step;
65    int n4, n2, n34;
66    FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
67    FFTComplex *tmpz;
68    float w_re, w_im;
69    float *w_re_ptr, *w_im_ptr;
70    const int fft_size = (1 << s->nbits);
71    float pom,  pom1,  pom2,  pom3;
72    float temp, temp1, temp3, temp4;
73    FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
74    FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
75    float f1 = 0.7071067812;
76
77    num_transforms = (21845 >> (17 - s->nbits)) | 1;
78
79    for (n=0; n<num_transforms; n++) {
80        offset = ff_fft_offsets_lut[n] << 2;
81        tmpz = z + offset;
82
83        tmp1 = tmpz[0].re + tmpz[1].re;
84        tmp5 = tmpz[2].re + tmpz[3].re;
85        tmp2 = tmpz[0].im + tmpz[1].im;
86        tmp6 = tmpz[2].im + tmpz[3].im;
87        tmp3 = tmpz[0].re - tmpz[1].re;
88        tmp8 = tmpz[2].im - tmpz[3].im;
89        tmp4 = tmpz[0].im - tmpz[1].im;
90        tmp7 = tmpz[2].re - tmpz[3].re;
91
92        tmpz[0].re = tmp1 + tmp5;
93        tmpz[2].re = tmp1 - tmp5;
94        tmpz[0].im = tmp2 + tmp6;
95        tmpz[2].im = tmp2 - tmp6;
96        tmpz[1].re = tmp3 + tmp8;
97        tmpz[3].re = tmp3 - tmp8;
98        tmpz[1].im = tmp4 - tmp7;
99        tmpz[3].im = tmp4 + tmp7;
100
101    }
102
103    if (fft_size < 8)
104        return;
105
106    num_transforms = (num_transforms >> 1) | 1;
107
108    for (n=0; n<num_transforms; n++) {
109        offset = ff_fft_offsets_lut[n] << 3;
110        tmpz = z + offset;
111
112        __asm__ volatile (
113            "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
114            "lwc1  %[pom],  40(%[tmpz])                     \n\t"
115            "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
116            "lwc1  %[pom1], 56(%[tmpz])                     \n\t"
117            "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
118            "lwc1  %[pom2], 44(%[tmpz])                     \n\t"
119            "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
120            "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
121            "add.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re + tmpz[5].re;
122            "add.s %[tmp3], %[tmp3],    %[pom1]             \n\t"  // tmp3 = tmpz[6].re + tmpz[7].re;
123            "add.s %[tmp2], %[tmp2],    %[pom2]             \n\t"  // tmp2 = tmpz[4].im + tmpz[5].im;
124            "lwc1  %[pom],  40(%[tmpz])                     \n\t"
125            "add.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im + tmpz[7].im;
126            "add.s %[tmp5], %[tmp1],    %[tmp3]             \n\t"  // tmp5 = tmp1 + tmp3;
127            "sub.s %[tmp7], %[tmp1],    %[tmp3]             \n\t"  // tmp7 = tmp1 - tmp3;
128            "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
129            "lwc1  %[pom1], 44(%[tmpz])                     \n\t"
130            "add.s %[tmp6], %[tmp2],    %[tmp4]             \n\t"  // tmp6 = tmp2 + tmp4;
131            "sub.s %[tmp8], %[tmp2],    %[tmp4]             \n\t"  // tmp8 = tmp2 - tmp4;
132            "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
133            "lwc1  %[pom2], 56(%[tmpz])                     \n\t"
134            "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
135            "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
136            "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
137            "sub.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re - tmpz[5].re;
138            "lwc1  %[pom],  0(%[tmpz])                      \n\t"
139            "sub.s %[tmp2], %[tmp2],    %[pom1]             \n\t"  // tmp2 = tmpz[4].im - tmpz[5].im;
140            "sub.s %[tmp3], %[tmp3],    %[pom2]             \n\t"  // tmp3 = tmpz[6].re - tmpz[7].re;
141            "lwc1  %[pom2], 4(%[tmpz])                      \n\t"
142            "sub.s %[pom1], %[pom],     %[tmp5]             \n\t"
143            "sub.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im - tmpz[7].im;
144            "add.s %[pom3], %[pom],     %[tmp5]             \n\t"
145            "sub.s %[pom],  %[pom2],    %[tmp6]             \n\t"
146            "add.s %[pom2], %[pom2],    %[tmp6]             \n\t"
147            "swc1  %[pom1], 32(%[tmpz])                     \n\t"  // tmpz[4].re = tmpz[0].re - tmp5;
148            "swc1  %[pom3], 0(%[tmpz])                      \n\t"  // tmpz[0].re = tmpz[0].re + tmp5;
149            "swc1  %[pom],  36(%[tmpz])                     \n\t"  // tmpz[4].im = tmpz[0].im - tmp6;
150            "swc1  %[pom2], 4(%[tmpz])                      \n\t"  // tmpz[0].im = tmpz[0].im + tmp6;
151            "lwc1  %[pom1], 16(%[tmpz])                     \n\t"
152            "lwc1  %[pom3], 20(%[tmpz])                     \n\t"
153            "add.s %[temp1],%[tmp1],    %[tmp2]             \n\t"
154            "sub.s %[temp], %[pom1],    %[tmp8]             \n\t"
155            "add.s %[pom2], %[pom3],    %[tmp7]             \n\t"
156            "sub.s %[temp3],%[tmp3],    %[tmp4]             \n\t"
157            "sub.s %[temp4],%[tmp2],    %[tmp1]             \n\t"
158            "swc1  %[temp], 48(%[tmpz])                     \n\t"  // tmpz[6].re = tmpz[2].re - tmp8;
159            "swc1  %[pom2], 52(%[tmpz])                     \n\t"  // tmpz[6].im = tmpz[2].im + tmp7;
160            "add.s %[pom1], %[pom1],    %[tmp8]             \n\t"
161            "sub.s %[pom3], %[pom3],    %[tmp7]             \n\t"
162            "add.s %[tmp3], %[tmp3],    %[tmp4]             \n\t"
163            "mul.s %[tmp5], %[f1],      %[temp1]            \n\t"  // tmp5 = pom * (tmp1 + tmp2);
164            "mul.s %[tmp7], %[f1],      %[temp3]            \n\t"  // tmp7 = pom * (tmp3 - tmp4);
165            "mul.s %[tmp6], %[f1],      %[temp4]            \n\t"  // tmp6 = pom * (tmp2 - tmp1);
166            "mul.s %[tmp8], %[f1],      %[tmp3]             \n\t"  // tmp8 = pom * (tmp3 + tmp4);
167            "swc1  %[pom1], 16(%[tmpz])                     \n\t"  // tmpz[2].re = tmpz[2].re + tmp8;
168            "swc1  %[pom3], 20(%[tmpz])                     \n\t"  // tmpz[2].im = tmpz[2].im - tmp7;
169            "add.s %[tmp1], %[tmp5],    %[tmp7]             \n\t"  // tmp1 = tmp5 + tmp7;
170            "sub.s %[tmp3], %[tmp5],    %[tmp7]             \n\t"  // tmp3 = tmp5 - tmp7;
171            "add.s %[tmp2], %[tmp6],    %[tmp8]             \n\t"  // tmp2 = tmp6 + tmp8;
172            "sub.s %[tmp4], %[tmp6],    %[tmp8]             \n\t"  // tmp4 = tmp6 - tmp8;
173            "lwc1  %[temp], 8(%[tmpz])                      \n\t"
174            "lwc1  %[temp1],12(%[tmpz])                     \n\t"
175            "lwc1  %[pom],  24(%[tmpz])                     \n\t"
176            "lwc1  %[pom2], 28(%[tmpz])                     \n\t"
177            "sub.s %[temp4],%[temp],    %[tmp1]             \n\t"
178            "sub.s %[temp3],%[temp1],   %[tmp2]             \n\t"
179            "add.s %[temp], %[temp],    %[tmp1]             \n\t"
180            "add.s %[temp1],%[temp1],   %[tmp2]             \n\t"
181            "sub.s %[pom1], %[pom],     %[tmp4]             \n\t"
182            "add.s %[pom3], %[pom2],    %[tmp3]             \n\t"
183            "add.s %[pom],  %[pom],     %[tmp4]             \n\t"
184            "sub.s %[pom2], %[pom2],    %[tmp3]             \n\t"
185            "swc1  %[temp4],40(%[tmpz])                     \n\t"  // tmpz[5].re = tmpz[1].re - tmp1;
186            "swc1  %[temp3],44(%[tmpz])                     \n\t"  // tmpz[5].im = tmpz[1].im - tmp2;
187            "swc1  %[temp], 8(%[tmpz])                      \n\t"  // tmpz[1].re = tmpz[1].re + tmp1;
188            "swc1  %[temp1],12(%[tmpz])                     \n\t"  // tmpz[1].im = tmpz[1].im + tmp2;
189            "swc1  %[pom1], 56(%[tmpz])                     \n\t"  // tmpz[7].re = tmpz[3].re - tmp4;
190            "swc1  %[pom3], 60(%[tmpz])                     \n\t"  // tmpz[7].im = tmpz[3].im + tmp3;
191            "swc1  %[pom],  24(%[tmpz])                     \n\t"  // tmpz[3].re = tmpz[3].re + tmp4;
192            "swc1  %[pom2], 28(%[tmpz])                     \n\t"  // tmpz[3].im = tmpz[3].im - tmp3;
193            : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),   [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
194              [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5),  [tmp7]"=&f"(tmp7),
195              [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
196              [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
197            : [tmpz]"r"(tmpz), [f1]"f"(f1)
198            : "memory"
199        );
200    }
201
202    step = 1 << (MAX_LOG2_NFFT - 4);
203    n4 = 4;
204
205    for (nbits=4; nbits<=s->nbits; nbits++) {
206        num_transforms = (num_transforms >> 1) | 1;
207        n2  = 2 * n4;
208        n34 = 3 * n4;
209
210        for (n=0; n<num_transforms; n++) {
211            offset = ff_fft_offsets_lut[n] << nbits;
212            tmpz = z + offset;
213
214            tmpz_n2  = tmpz +  n2;
215            tmpz_n4  = tmpz +  n4;
216            tmpz_n34 = tmpz +  n34;
217
218            __asm__ volatile (
219                "lwc1  %[pom1], 0(%[tmpz_n2])            \n\t"
220                "lwc1  %[pom],  0(%[tmpz_n34])           \n\t"
221                "lwc1  %[pom2], 4(%[tmpz_n2])            \n\t"
222                "lwc1  %[pom3], 4(%[tmpz_n34])           \n\t"
223                "lwc1  %[temp1],0(%[tmpz])               \n\t"
224                "lwc1  %[temp3],4(%[tmpz])               \n\t"
225                "add.s %[tmp5], %[pom1],      %[pom]     \n\t"   //  tmp5 = tmpz[ n2].re + tmpz[n34].re;
226                "sub.s %[tmp1], %[pom1],      %[pom]     \n\t"   //  tmp1 = tmpz[ n2].re - tmpz[n34].re;
227                "add.s %[tmp6], %[pom2],      %[pom3]    \n\t"   //  tmp6 = tmpz[ n2].im + tmpz[n34].im;
228                "sub.s %[tmp2], %[pom2],      %[pom3]    \n\t"   //  tmp2 = tmpz[ n2].im - tmpz[n34].im;
229                "sub.s %[temp], %[temp1],     %[tmp5]    \n\t"
230                "add.s %[temp1],%[temp1],     %[tmp5]    \n\t"
231                "sub.s %[temp4],%[temp3],     %[tmp6]    \n\t"
232                "add.s %[temp3],%[temp3],     %[tmp6]    \n\t"
233                "swc1  %[temp], 0(%[tmpz_n2])            \n\t"   //  tmpz[ n2].re = tmpz[ 0].re - tmp5;
234                "swc1  %[temp1],0(%[tmpz])               \n\t"   //  tmpz[  0].re = tmpz[ 0].re + tmp5;
235                "lwc1  %[pom1], 0(%[tmpz_n4])            \n\t"
236                "swc1  %[temp4],4(%[tmpz_n2])            \n\t"   //  tmpz[ n2].im = tmpz[ 0].im - tmp6;
237                "lwc1  %[temp], 4(%[tmpz_n4])            \n\t"
238                "swc1  %[temp3],4(%[tmpz])               \n\t"   //  tmpz[  0].im = tmpz[ 0].im + tmp6;
239                "sub.s %[pom],  %[pom1],      %[tmp2]    \n\t"
240                "add.s %[pom1], %[pom1],      %[tmp2]    \n\t"
241                "add.s %[temp1],%[temp],      %[tmp1]    \n\t"
242                "sub.s %[temp], %[temp],      %[tmp1]    \n\t"
243                "swc1  %[pom],  0(%[tmpz_n34])           \n\t"   //  tmpz[n34].re = tmpz[n4].re - tmp2;
244                "swc1  %[pom1], 0(%[tmpz_n4])            \n\t"   //  tmpz[ n4].re = tmpz[n4].re + tmp2;
245                "swc1  %[temp1],4(%[tmpz_n34])           \n\t"   //  tmpz[n34].im = tmpz[n4].im + tmp1;
246                "swc1  %[temp], 4(%[tmpz_n4])            \n\t"   //  tmpz[ n4].im = tmpz[n4].im - tmp1;
247                : [tmp5]"=&f"(tmp5),
248                  [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),        [pom1]"=&f"(pom1),        [pom2]"=&f"(pom2),
249                  [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6),          [pom3]"=&f"(pom3),
250                  [temp]"=&f"(temp), [temp1]"=&f"(temp1),     [temp3]"=&f"(temp3),       [temp4]"=&f"(temp4)
251                : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
252                : "memory"
253            );
254
255            w_re_ptr = (float*)(ff_cos_131072 + step);
256            w_im_ptr = (float*)(ff_cos_131072 + MAX_FFT_SIZE/4 - step);
257
258            for (i=1; i<n4; i++) {
259                w_re = w_re_ptr[0];
260                w_im = w_im_ptr[0];
261                tmpz_n2_i = tmpz_n2  + i;
262                tmpz_n4_i = tmpz_n4  + i;
263                tmpz_n34_i= tmpz_n34 + i;
264                tmpz_i    = tmpz     + i;
265
266                __asm__ volatile (
267                    "lwc1     %[temp],  0(%[tmpz_n2_i])               \n\t"
268                    "lwc1     %[temp1], 4(%[tmpz_n2_i])               \n\t"
269                    "lwc1     %[pom],   0(%[tmpz_n34_i])              \n\t"
270                    "lwc1     %[pom1],  4(%[tmpz_n34_i])              \n\t"
271                    "mul.s    %[temp3], %[w_im],    %[temp]           \n\t"
272                    "mul.s    %[temp4], %[w_im],    %[temp1]          \n\t"
273                    "mul.s    %[pom2],  %[w_im],    %[pom1]           \n\t"
274                    "mul.s    %[pom3],  %[w_im],    %[pom]            \n\t"
275                    "msub.s   %[tmp2],  %[temp3],   %[w_re], %[temp1] \n\t"  // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
276                    "madd.s   %[tmp1],  %[temp4],   %[w_re], %[temp]  \n\t"  // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
277                    "msub.s   %[tmp3],  %[pom2],    %[w_re], %[pom]   \n\t"  // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
278                    "madd.s   %[tmp4],  %[pom3],    %[w_re], %[pom1]  \n\t"  // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
279                    "lwc1     %[temp],  0(%[tmpz_i])                  \n\t"
280                    "lwc1     %[pom],   4(%[tmpz_i])                  \n\t"
281                    "add.s    %[tmp5],  %[tmp1],    %[tmp3]           \n\t"  // tmp5 = tmp1 + tmp3;
282                    "sub.s    %[tmp1],  %[tmp1],    %[tmp3]           \n\t"  // tmp1 = tmp1 - tmp3;
283                    "add.s    %[tmp6],  %[tmp2],    %[tmp4]           \n\t"  // tmp6 = tmp2 + tmp4;
284                    "sub.s    %[tmp2],  %[tmp2],    %[tmp4]           \n\t"  // tmp2 = tmp2 - tmp4;
285                    "sub.s    %[temp1], %[temp],    %[tmp5]           \n\t"
286                    "add.s    %[temp],  %[temp],    %[tmp5]           \n\t"
287                    "sub.s    %[pom1],  %[pom],     %[tmp6]           \n\t"
288                    "add.s    %[pom],   %[pom],     %[tmp6]           \n\t"
289                    "lwc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"
290                    "lwc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"
291                    "swc1     %[temp1], 0(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].re = tmpz[   i].re - tmp5;
292                    "swc1     %[temp],  0(%[tmpz_i])                  \n\t"  // tmpz[    i].re = tmpz[   i].re + tmp5;
293                    "swc1     %[pom1],  4(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].im = tmpz[   i].im - tmp6;
294                    "swc1     %[pom] ,  4(%[tmpz_i])                  \n\t"  // tmpz[    i].im = tmpz[   i].im + tmp6;
295                    "sub.s    %[temp4], %[temp3],   %[tmp2]           \n\t"
296                    "add.s    %[pom3],  %[pom2],    %[tmp1]           \n\t"
297                    "add.s    %[temp3], %[temp3],   %[tmp2]           \n\t"
298                    "sub.s    %[pom2],  %[pom2],    %[tmp1]           \n\t"
299                    "swc1     %[temp4], 0(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
300                    "swc1     %[pom3],  4(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
301                    "swc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
302                    "swc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
303                    : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
304                      [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
305                      [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
306                      [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
307                    : [w_re]"f"(w_re), [w_im]"f"(w_im),
308                      [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
309                      [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
310                    : "memory"
311                );
312                w_re_ptr += step;
313                w_im_ptr -= step;
314            }
315        }
316        step >>= 1;
317        n4   <<= 1;
318    }
319}
320
321/**
322 * MDCT/IMDCT transforms.
323 */
324
325static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
326{
327    int k, n8, n4, n2, n, j;
328    const uint16_t *revtab = s->revtab;
329    const FFTSample *tcos = s->tcos;
330    const FFTSample *tsin = s->tsin;
331    const FFTSample *in1, *in2, *in3, *in4;
332    FFTComplex *z = (FFTComplex *)output;
333
334    int j1;
335    const float *tcos1, *tsin1, *tcos2, *tsin2;
336    float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
337        temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
338    FFTComplex *z1, *z2;
339
340    n = 1 << s->mdct_bits;
341    n2 = n >> 1;
342    n4 = n >> 2;
343    n8 = n >> 3;
344
345    /* pre rotation */
346    in1 = input;
347    in2 = input + n2 - 1;
348    in3 = input + 2;
349    in4 = input + n2 - 3;
350
351    tcos1 = tcos;
352    tsin1 = tsin;
353
354    /* n4 = 64 or 128 */
355    for(k = 0; k < n4; k += 2) {
356        j  = revtab[k    ];
357        j1 = revtab[k + 1];
358
359        __asm__ volatile (
360            "lwc1           %[temp1],       0(%[in2])                           \t\n"
361            "lwc1           %[temp2],       0(%[tcos1])                         \t\n"
362            "lwc1           %[temp3],       0(%[tsin1])                         \t\n"
363            "lwc1           %[temp4],       0(%[in1])                           \t\n"
364            "lwc1           %[temp5],       0(%[in4])                           \t\n"
365            "mul.s          %[temp9],       %[temp1],   %[temp2]                \t\n"
366            "mul.s          %[temp10],      %[temp1],   %[temp3]                \t\n"
367            "lwc1           %[temp6],       4(%[tcos1])                         \t\n"
368            "lwc1           %[temp7],       4(%[tsin1])                         \t\n"
369            "nmsub.s        %[temp9],       %[temp9],   %[temp4],   %[temp3]    \t\n"
370            "madd.s         %[temp10],      %[temp10],  %[temp4],   %[temp2]    \t\n"
371            "mul.s          %[temp11],      %[temp5],   %[temp6]                \t\n"
372            "mul.s          %[temp12],      %[temp5],   %[temp7]                \t\n"
373            "lwc1           %[temp8],       0(%[in3])                           \t\n"
374            PTR_ADDIU "     %[tcos1],       %[tcos1],   8                       \t\n"
375            PTR_ADDIU "     %[tsin1],       %[tsin1],   8                       \t\n"
376            PTR_ADDIU "     %[in1],         %[in1],     16                      \t\n"
377            "nmsub.s        %[temp11],      %[temp11],  %[temp8],   %[temp7]    \t\n"
378            "madd.s         %[temp12],      %[temp12],  %[temp8],   %[temp6]    \t\n"
379            PTR_ADDIU "     %[in2],         %[in2],     -16                     \t\n"
380            PTR_ADDIU "     %[in3],         %[in3],     16                      \t\n"
381            PTR_ADDIU "     %[in4],         %[in4],     -16                     \t\n"
382
383            : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
384              [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
385              [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
386              [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
387              [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
388              [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
389              [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
390              [in1]"+r"(in1), [in2]"+r"(in2),
391              [in3]"+r"(in3), [in4]"+r"(in4)
392            :
393            : "memory"
394        );
395
396        z[j ].re = temp9;
397        z[j ].im = temp10;
398        z[j1].re = temp11;
399        z[j1].im = temp12;
400    }
401
402    s->fft_calc(s, z);
403
404    /* post rotation + reordering */
405    /* n8 = 32 or 64 */
406    for(k = 0; k < n8; k += 2) {
407        tcos1 = &tcos[n8 - k - 2];
408        tsin1 = &tsin[n8 - k - 2];
409        tcos2 = &tcos[n8 + k];
410        tsin2 = &tsin[n8 + k];
411        z1 = &z[n8 - k - 2];
412        z2 = &z[n8 + k    ];
413
414        __asm__ volatile (
415            "lwc1       %[temp1],   12(%[z1])                           \t\n"
416            "lwc1       %[temp2],   4(%[tsin1])                         \t\n"
417            "lwc1       %[temp3],   4(%[tcos1])                         \t\n"
418            "lwc1       %[temp4],   8(%[z1])                            \t\n"
419            "lwc1       %[temp5],   4(%[z1])                            \t\n"
420            "mul.s      %[temp9],   %[temp1],   %[temp2]                \t\n"
421            "mul.s      %[temp10],  %[temp1],   %[temp3]                \t\n"
422            "lwc1       %[temp6],   0(%[tsin1])                         \t\n"
423            "lwc1       %[temp7],   0(%[tcos1])                         \t\n"
424            "nmsub.s    %[temp9],   %[temp9],   %[temp4],   %[temp3]    \t\n"
425            "madd.s     %[temp10],  %[temp10],  %[temp4],   %[temp2]    \t\n"
426            "mul.s      %[temp11],  %[temp5],   %[temp6]                \t\n"
427            "mul.s      %[temp12],  %[temp5],   %[temp7]                \t\n"
428            "lwc1       %[temp8],   0(%[z1])                            \t\n"
429            "lwc1       %[temp1],   4(%[z2])                            \t\n"
430            "lwc1       %[temp2],   0(%[tsin2])                         \t\n"
431            "lwc1       %[temp3],   0(%[tcos2])                         \t\n"
432            "nmsub.s    %[temp11],  %[temp11],  %[temp8],   %[temp7]    \t\n"
433            "madd.s     %[temp12],  %[temp12],  %[temp8],   %[temp6]    \t\n"
434            "mul.s      %[temp13],  %[temp1],   %[temp2]                \t\n"
435            "mul.s      %[temp14],  %[temp1],   %[temp3]                \t\n"
436            "lwc1       %[temp4],   0(%[z2])                            \t\n"
437            "lwc1       %[temp5],   12(%[z2])                           \t\n"
438            "lwc1       %[temp6],   4(%[tsin2])                         \t\n"
439            "lwc1       %[temp7],   4(%[tcos2])                         \t\n"
440            "nmsub.s    %[temp13],  %[temp13],  %[temp4],   %[temp3]    \t\n"
441            "madd.s     %[temp14],  %[temp14],  %[temp4],   %[temp2]    \t\n"
442            "mul.s      %[temp15],  %[temp5],   %[temp6]                \t\n"
443            "mul.s      %[temp16],  %[temp5],   %[temp7]                \t\n"
444            "lwc1       %[temp8],   8(%[z2])                            \t\n"
445            "nmsub.s    %[temp15],  %[temp15],  %[temp8],   %[temp7]    \t\n"
446            "madd.s     %[temp16],  %[temp16],  %[temp8],   %[temp6]    \t\n"
447            : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
448              [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
449              [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
450              [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
451              [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
452              [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
453              [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
454              [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
455            : [z1]"r"(z1), [z2]"r"(z2),
456              [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
457              [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
458            : "memory"
459        );
460
461        z1[1].re = temp9;
462        z1[1].im = temp14;
463        z2[0].re = temp13;
464        z2[0].im = temp10;
465
466        z1[0].re = temp11;
467        z1[0].im = temp16;
468        z2[1].re = temp15;
469        z2[1].im = temp12;
470    }
471}
472
473/**
474 * Compute inverse MDCT of size N = 2^nbits
475 * @param output N samples
476 * @param input N/2 samples
477 */
478static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
479{
480    int k;
481    int n = 1 << s->mdct_bits;
482    int n2 = n >> 1;
483    int n4 = n >> 2;
484
485    ff_imdct_half_mips(s, output+n4, input);
486
487    for(k = 0; k < n4; k+=4) {
488        output[k] = -output[n2-k-1];
489        output[k+1] = -output[n2-k-2];
490        output[k+2] = -output[n2-k-3];
491        output[k+3] = -output[n2-k-4];
492
493        output[n-k-1] = output[n2+k];
494        output[n-k-2] = output[n2+k+1];
495        output[n-k-3] = output[n2+k+2];
496        output[n-k-4] = output[n2+k+3];
497    }
498}
499#endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
500#endif /* HAVE_INLINE_ASM */
501
502av_cold void ff_fft_init_mips(FFTContext *s)
503{
504    ff_fft_lut_init();
505    ff_init_ff_cos_tabs(17);
506
507#if HAVE_INLINE_ASM
508#if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
509    s->fft_calc     = ff_fft_calc_mips;
510#if CONFIG_MDCT
511    s->imdct_calc   = ff_imdct_calc_mips;
512    s->imdct_half   = ff_imdct_half_mips;
513#endif
514#endif
515#endif
516}
517