FFmpeg: libavcodec/mips/fft_mips.c Source File

00001 /*
00002  * Copyright (c) 2012
00003  *      MIPS Technologies, Inc., California.
00004  *
00005  * Redistribution and use in source and binary forms, with or without
00006  * modification, are permitted provided that the following conditions
00007  * are met:
00008  * 1. Redistributions of source code must retain the above copyright
00009  *    notice, this list of conditions and the following disclaimer.
00010  * 2. Redistributions in binary form must reproduce the above copyright
00011  *    notice, this list of conditions and the following disclaimer in the
00012  *    documentation and/or other materials provided with the distribution.
00013  * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
00014  *    contributors may be used to endorse or promote products derived from
00015  *    this software without specific prior written permission.
00016  *
00017  * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
00018  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
00019  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
00020  * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
00021  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
00022  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
00023  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
00024  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00025  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
00026  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
00027  * SUCH DAMAGE.
00028  *
00029  * Author:  Stanislav Ocovaj (socovaj@mips.com)
00030  * Author:  Zoran Lukic (zoranl@mips.com)
00031  *
00032  * Optimized MDCT/IMDCT and FFT transforms
00033  *
00034  * This file is part of FFmpeg.
00035  *
00036  * FFmpeg is free software; you can redistribute it and/or
00037  * modify it under the terms of the GNU Lesser General Public
00038  * License as published by the Free Software Foundation; either
00039  * version 2.1 of the License, or (at your option) any later version.
00040  *
00041  * FFmpeg is distributed in the hope that it will be useful,
00042  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00043  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00044  * Lesser General Public License for more details.
00045  *
00046  * You should have received a copy of the GNU Lesser General Public
00047  * License along with FFmpeg; if not, write to the Free Software
00048  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00049  */
00050 #include "config.h"
00051 #include "libavcodec/fft.h"
00052 #include "fft_table.h"
00053 
00058 #if HAVE_INLINE_ASM
00059 static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
00060 {
00061     int nbits, i, n, num_transforms, offset, step;
00062     int n4, n2, n34;
00063     FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
00064     FFTComplex *tmpz;
00065     float w_re, w_im;
00066     float *w_re_ptr, *w_im_ptr;
00067     const int fft_size = (1 << s->nbits);
00068     int s_n = s->nbits;
00069     int tem1, tem2;
00070     float pom,  pom1,  pom2,  pom3;
00071     float temp, temp1, temp3, temp4;
00072     FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
00073     FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
00074 
00078     __asm__ volatile (
00079         "li   %[tem1], 16                                      \n\t"
00080         "sub  %[s_n],  %[tem1], %[s_n]                         \n\t"
00081         "li   %[tem2], 10923                                   \n\t"
00082         "srav %[tem2], %[tem2], %[s_n]                         \n\t"
00083         "ori  %[num_t],%[tem2], 1                              \n\t"
00084         : [num_t]"=r"(num_transforms), [s_n]"+r"(s_n),
00085           [tem1]"=&r"(tem1), [tem2]"=&r"(tem2)
00086     );
00087 
00088 
00089     for (n=0; n<num_transforms; n++) {
00090         offset = fft_offsets_lut[n] << 2;
00091         tmpz = z + offset;
00092 
00093         tmp1 = tmpz[0].re + tmpz[1].re;
00094         tmp5 = tmpz[2].re + tmpz[3].re;
00095         tmp2 = tmpz[0].im + tmpz[1].im;
00096         tmp6 = tmpz[2].im + tmpz[3].im;
00097         tmp3 = tmpz[0].re - tmpz[1].re;
00098         tmp8 = tmpz[2].im - tmpz[3].im;
00099         tmp4 = tmpz[0].im - tmpz[1].im;
00100         tmp7 = tmpz[2].re - tmpz[3].re;
00101 
00102         tmpz[0].re = tmp1 + tmp5;
00103         tmpz[2].re = tmp1 - tmp5;
00104         tmpz[0].im = tmp2 + tmp6;
00105         tmpz[2].im = tmp2 - tmp6;
00106         tmpz[1].re = tmp3 + tmp8;
00107         tmpz[3].re = tmp3 - tmp8;
00108         tmpz[1].im = tmp4 - tmp7;
00109         tmpz[3].im = tmp4 + tmp7;
00110 
00111     }
00112 
00113     if (fft_size < 8)
00114         return;
00115 
00116     num_transforms = (num_transforms >> 1) | 1;
00117 
00118     for (n=0; n<num_transforms; n++) {
00119         offset = fft_offsets_lut[n] << 3;
00120         tmpz = z + offset;
00121 
00122         __asm__ volatile (
00123             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
00124             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
00125             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
00126             "lwc1  %[pom1], 56(%[tmpz])                     \n\t"
00127             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
00128             "lwc1  %[pom2], 44(%[tmpz])                     \n\t"
00129             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
00130             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
00131             "add.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re + tmpz[5].re;
00132             "add.s %[tmp3], %[tmp3],    %[pom1]             \n\t"  // tmp3 = tmpz[6].re + tmpz[7].re;
00133             "add.s %[tmp2], %[tmp2],    %[pom2]             \n\t"  // tmp2 = tmpz[4].im + tmpz[5].im;
00134             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
00135             "add.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im + tmpz[7].im;
00136             "add.s %[tmp5], %[tmp1],    %[tmp3]             \n\t"  // tmp5 = tmp1 + tmp3;
00137             "sub.s %[tmp7], %[tmp1],    %[tmp3]             \n\t"  // tmp7 = tmp1 - tmp3;
00138             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
00139             "lwc1  %[pom1], 44(%[tmpz])                     \n\t"
00140             "add.s %[tmp6], %[tmp2],    %[tmp4]             \n\t"  // tmp6 = tmp2 + tmp4;
00141             "sub.s %[tmp8], %[tmp2],    %[tmp4]             \n\t"  // tmp8 = tmp2 - tmp4;
00142             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
00143             "lwc1  %[pom2], 56(%[tmpz])                     \n\t"
00144             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
00145             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
00146             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
00147             "sub.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re - tmpz[5].re;
00148             "lwc1  %[pom],  0(%[tmpz])                      \n\t"
00149             "sub.s %[tmp2], %[tmp2],    %[pom1]             \n\t"  // tmp2 = tmpz[4].im - tmpz[5].im;
00150             "sub.s %[tmp3], %[tmp3],    %[pom2]             \n\t"  // tmp3 = tmpz[6].re - tmpz[7].re;
00151             "lwc1  %[pom2], 4(%[tmpz])                      \n\t"
00152             "sub.s %[pom1], %[pom],     %[tmp5]             \n\t"
00153             "sub.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im - tmpz[7].im;
00154             "add.s %[pom3], %[pom],     %[tmp5]             \n\t"
00155             "sub.s %[pom],  %[pom2],    %[tmp6]             \n\t"
00156             "add.s %[pom2], %[pom2],    %[tmp6]             \n\t"
00157             "swc1  %[pom1], 32(%[tmpz])                     \n\t"  // tmpz[4].re = tmpz[0].re - tmp5;
00158             "swc1  %[pom3], 0(%[tmpz])                      \n\t"  // tmpz[0].re = tmpz[0].re + tmp5;
00159             "swc1  %[pom],  36(%[tmpz])                     \n\t"  // tmpz[4].im = tmpz[0].im - tmp6;
00160             "swc1  %[pom2], 4(%[tmpz])                      \n\t"  // tmpz[0].im = tmpz[0].im + tmp6;
00161             "lwc1  %[pom1], 16(%[tmpz])                     \n\t"
00162             "lwc1  %[pom3], 20(%[tmpz])                     \n\t"
00163             "li.s  %[pom],  0.7071067812                    \n\t"  // float pom = 0.7071067812f;
00164             "add.s %[temp1],%[tmp1],    %[tmp2]             \n\t"
00165             "sub.s %[temp], %[pom1],    %[tmp8]             \n\t"
00166             "add.s %[pom2], %[pom3],    %[tmp7]             \n\t"
00167             "sub.s %[temp3],%[tmp3],    %[tmp4]             \n\t"
00168             "sub.s %[temp4],%[tmp2],    %[tmp1]             \n\t"
00169             "swc1  %[temp], 48(%[tmpz])                     \n\t"  // tmpz[6].re = tmpz[2].re - tmp8;
00170             "swc1  %[pom2], 52(%[tmpz])                     \n\t"  // tmpz[6].im = tmpz[2].im + tmp7;
00171             "add.s %[pom1], %[pom1],    %[tmp8]             \n\t"
00172             "sub.s %[pom3], %[pom3],    %[tmp7]             \n\t"
00173             "add.s %[tmp3], %[tmp3],    %[tmp4]             \n\t"
00174             "mul.s %[tmp5], %[pom],     %[temp1]            \n\t"  // tmp5 = pom * (tmp1 + tmp2);
00175             "mul.s %[tmp7], %[pom],     %[temp3]            \n\t"  // tmp7 = pom * (tmp3 - tmp4);
00176             "mul.s %[tmp6], %[pom],     %[temp4]            \n\t"  // tmp6 = pom * (tmp2 - tmp1);
00177             "mul.s %[tmp8], %[pom],     %[tmp3]             \n\t"  // tmp8 = pom * (tmp3 + tmp4);
00178             "swc1  %[pom1], 16(%[tmpz])                     \n\t"  // tmpz[2].re = tmpz[2].re + tmp8;
00179             "swc1  %[pom3], 20(%[tmpz])                     \n\t"  // tmpz[2].im = tmpz[2].im - tmp7;
00180             "add.s %[tmp1], %[tmp5],    %[tmp7]             \n\t"  // tmp1 = tmp5 + tmp7;
00181             "sub.s %[tmp3], %[tmp5],    %[tmp7]             \n\t"  // tmp3 = tmp5 - tmp7;
00182             "add.s %[tmp2], %[tmp6],    %[tmp8]             \n\t"  // tmp2 = tmp6 + tmp8;
00183             "sub.s %[tmp4], %[tmp6],    %[tmp8]             \n\t"  // tmp4 = tmp6 - tmp8;
00184             "lwc1  %[temp], 8(%[tmpz])                      \n\t"
00185             "lwc1  %[temp1],12(%[tmpz])                     \n\t"
00186             "lwc1  %[pom],  24(%[tmpz])                     \n\t"
00187             "lwc1  %[pom2], 28(%[tmpz])                     \n\t"
00188             "sub.s %[temp4],%[temp],    %[tmp1]             \n\t"
00189             "sub.s %[temp3],%[temp1],   %[tmp2]             \n\t"
00190             "add.s %[temp], %[temp],    %[tmp1]             \n\t"
00191             "add.s %[temp1],%[temp1],   %[tmp2]             \n\t"
00192             "sub.s %[pom1], %[pom],     %[tmp4]             \n\t"
00193             "add.s %[pom3], %[pom2],    %[tmp3]             \n\t"
00194             "add.s %[pom],  %[pom],     %[tmp4]             \n\t"
00195             "sub.s %[pom2], %[pom2],    %[tmp3]             \n\t"
00196             "swc1  %[temp4],40(%[tmpz])                     \n\t"  // tmpz[5].re = tmpz[1].re - tmp1;
00197             "swc1  %[temp3],44(%[tmpz])                     \n\t"  // tmpz[5].im = tmpz[1].im - tmp2;
00198             "swc1  %[temp], 8(%[tmpz])                      \n\t"  // tmpz[1].re = tmpz[1].re + tmp1;
00199             "swc1  %[temp1],12(%[tmpz])                     \n\t"  // tmpz[1].im = tmpz[1].im + tmp2;
00200             "swc1  %[pom1], 56(%[tmpz])                     \n\t"  // tmpz[7].re = tmpz[3].re - tmp4;
00201             "swc1  %[pom3], 60(%[tmpz])                     \n\t"  // tmpz[7].im = tmpz[3].im + tmp3;
00202             "swc1  %[pom],  24(%[tmpz])                     \n\t"  // tmpz[3].re = tmpz[3].re + tmp4;
00203             "swc1  %[pom2], 28(%[tmpz])                     \n\t"  // tmpz[3].im = tmpz[3].im - tmp3;
00204             : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),   [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
00205               [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5),  [tmp7]"=&f"(tmp7),
00206               [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
00207               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
00208             : [tmpz]"r"(tmpz)
00209             : "memory"
00210         );
00211     }
00212 
00213     step = 1 << (MAX_LOG2_NFFT - 4);
00214     n4 = 4;
00215 
00216     for (nbits=4; nbits<=s->nbits; nbits++) {
00217         /*
00218         * num_transforms = (num_transforms >> 1) | 1;
00219         */
00220         __asm__ volatile (
00221             "sra %[num_t], %[num_t], 1               \n\t"
00222             "ori %[num_t], %[num_t], 1               \n\t"
00223 
00224             : [num_t] "+r" (num_transforms)
00225         );
00226         n2  = 2 * n4;
00227         n34 = 3 * n4;
00228 
00229         for (n=0; n<num_transforms; n++) {
00230             offset = fft_offsets_lut[n] << nbits;
00231             tmpz = z + offset;
00232 
00233             tmpz_n2  = tmpz +  n2;
00234             tmpz_n4  = tmpz +  n4;
00235             tmpz_n34 = tmpz +  n34;
00236 
00237             __asm__ volatile (
00238                 "lwc1  %[pom1], 0(%[tmpz_n2])            \n\t"
00239                 "lwc1  %[pom],  0(%[tmpz_n34])           \n\t"
00240                 "lwc1  %[pom2], 4(%[tmpz_n2])            \n\t"
00241                 "lwc1  %[pom3], 4(%[tmpz_n34])           \n\t"
00242                 "lwc1  %[temp1],0(%[tmpz])               \n\t"
00243                 "lwc1  %[temp3],4(%[tmpz])               \n\t"
00244                 "add.s %[tmp5], %[pom1],      %[pom]     \n\t"   //  tmp5 = tmpz[ n2].re + tmpz[n34].re;
00245                 "sub.s %[tmp1], %[pom1],      %[pom]     \n\t"   //  tmp1 = tmpz[ n2].re - tmpz[n34].re;
00246                 "add.s %[tmp6], %[pom2],      %[pom3]    \n\t"   //  tmp6 = tmpz[ n2].im + tmpz[n34].im;
00247                 "sub.s %[tmp2], %[pom2],      %[pom3]    \n\t"   //  tmp2 = tmpz[ n2].im - tmpz[n34].im;
00248                 "sub.s %[temp], %[temp1],     %[tmp5]    \n\t"
00249                 "add.s %[temp1],%[temp1],     %[tmp5]    \n\t"
00250                 "sub.s %[temp4],%[temp3],     %[tmp6]    \n\t"
00251                 "add.s %[temp3],%[temp3],     %[tmp6]    \n\t"
00252                 "swc1  %[temp], 0(%[tmpz_n2])            \n\t"   //  tmpz[ n2].re = tmpz[ 0].re - tmp5;
00253                 "swc1  %[temp1],0(%[tmpz])               \n\t"   //  tmpz[  0].re = tmpz[ 0].re + tmp5;
00254                 "lwc1  %[pom1], 0(%[tmpz_n4])            \n\t"
00255                 "swc1  %[temp4],4(%[tmpz_n2])            \n\t"   //  tmpz[ n2].im = tmpz[ 0].im - tmp6;
00256                 "lwc1  %[temp], 4(%[tmpz_n4])            \n\t"
00257                 "swc1  %[temp3],4(%[tmpz])               \n\t"   //  tmpz[  0].im = tmpz[ 0].im + tmp6;
00258                 "sub.s %[pom],  %[pom1],      %[tmp2]    \n\t"
00259                 "add.s %[pom1], %[pom1],      %[tmp2]    \n\t"
00260                 "add.s %[temp1],%[temp],      %[tmp1]    \n\t"
00261                 "sub.s %[temp], %[temp],      %[tmp1]    \n\t"
00262                 "swc1  %[pom],  0(%[tmpz_n34])           \n\t"   //  tmpz[n34].re = tmpz[n4].re - tmp2;
00263                 "swc1  %[pom1], 0(%[tmpz_n4])            \n\t"   //  tmpz[ n4].re = tmpz[n4].re + tmp2;
00264                 "swc1  %[temp1],4(%[tmpz_n34])           \n\t"   //  tmpz[n34].im = tmpz[n4].im + tmp1;
00265                 "swc1  %[temp], 4(%[tmpz_n4])            \n\t"   //  tmpz[ n4].im = tmpz[n4].im - tmp1;
00266                 : [tmp5]"=&f"(tmp5),
00267                   [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),        [pom1]"=&f"(pom1),        [pom2]"=&f"(pom2),
00268                   [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6),          [pom3]"=&f"(pom3),
00269                   [temp]"=&f"(temp), [temp1]"=&f"(temp1),     [temp3]"=&f"(temp3),       [temp4]"=&f"(temp4)
00270                 : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
00271                 : "memory"
00272             );
00273 
00274             w_re_ptr = (float*)(ff_cos_65536 + step);
00275             w_im_ptr = (float*)(ff_cos_65536 + MAX_FFT_SIZE/4 - step);
00276 
00277             for (i=1; i<n4; i++) {
00278                 w_re = w_re_ptr[0];
00279                 w_im = w_im_ptr[0];
00280                 tmpz_n2_i = tmpz_n2  + i;
00281                 tmpz_n4_i = tmpz_n4  + i;
00282                 tmpz_n34_i= tmpz_n34 + i;
00283                 tmpz_i    = tmpz     + i;
00284 
00285                 __asm__ volatile (
00286                     "lwc1     %[temp],  0(%[tmpz_n2_i])               \n\t"
00287                     "lwc1     %[temp1], 4(%[tmpz_n2_i])               \n\t"
00288                     "lwc1     %[pom],   0(%[tmpz_n34_i])              \n\t"
00289                     "lwc1     %[pom1],  4(%[tmpz_n34_i])              \n\t"
00290                     "mul.s    %[temp3], %[w_im],    %[temp]           \n\t"
00291                     "mul.s    %[temp4], %[w_im],    %[temp1]          \n\t"
00292                     "mul.s    %[pom2],  %[w_im],    %[pom1]           \n\t"
00293                     "mul.s    %[pom3],  %[w_im],    %[pom]            \n\t"
00294                     "msub.s   %[tmp2],  %[temp3],   %[w_re], %[temp1] \n\t"  // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
00295                     "madd.s   %[tmp1],  %[temp4],   %[w_re], %[temp]  \n\t"  // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
00296                     "msub.s   %[tmp3],  %[pom2],    %[w_re], %[pom]   \n\t"  // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
00297                     "madd.s   %[tmp4],  %[pom3],    %[w_re], %[pom1]  \n\t"  // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
00298                     "lwc1     %[temp],  0(%[tmpz_i])                  \n\t"
00299                     "lwc1     %[pom],   4(%[tmpz_i])                  \n\t"
00300                     "add.s    %[tmp5],  %[tmp1],    %[tmp3]           \n\t"  // tmp5 = tmp1 + tmp3;
00301                     "sub.s    %[tmp1],  %[tmp1],    %[tmp3]           \n\t"  // tmp1 = tmp1 - tmp3;
00302                     "add.s    %[tmp6],  %[tmp2],    %[tmp4]           \n\t"  // tmp6 = tmp2 + tmp4;
00303                     "sub.s    %[tmp2],  %[tmp2],    %[tmp4]           \n\t"  // tmp2 = tmp2 - tmp4;
00304                     "sub.s    %[temp1], %[temp],    %[tmp5]           \n\t"
00305                     "add.s    %[temp],  %[temp],    %[tmp5]           \n\t"
00306                     "sub.s    %[pom1],  %[pom],     %[tmp6]           \n\t"
00307                     "add.s    %[pom],   %[pom],     %[tmp6]           \n\t"
00308                     "lwc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"
00309                     "lwc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"
00310                     "swc1     %[temp1], 0(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].re = tmpz[   i].re - tmp5;
00311                     "swc1     %[temp],  0(%[tmpz_i])                  \n\t"  // tmpz[    i].re = tmpz[   i].re + tmp5;
00312                     "swc1     %[pom1],  4(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].im = tmpz[   i].im - tmp6;
00313                     "swc1     %[pom] ,  4(%[tmpz_i])                  \n\t"  // tmpz[    i].im = tmpz[   i].im + tmp6;
00314                     "sub.s    %[temp4], %[temp3],   %[tmp2]           \n\t"
00315                     "add.s    %[pom3],  %[pom2],    %[tmp1]           \n\t"
00316                     "add.s    %[temp3], %[temp3],   %[tmp2]           \n\t"
00317                     "sub.s    %[pom2],  %[pom2],    %[tmp1]           \n\t"
00318                     "swc1     %[temp4], 0(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
00319                     "swc1     %[pom3],  4(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
00320                     "swc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
00321                     "swc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
00322                     : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
00323                       [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
00324                       [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
00325                       [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
00326                     : [w_re]"f"(w_re), [w_im]"f"(w_im),
00327                       [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
00328                       [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
00329                     : "memory"
00330                 );
00331                 w_re_ptr += step;
00332                 w_im_ptr -= step;
00333             }
00334         }
00335         step >>= 1;
00336         n4   <<= 1;
00337     }
00338 }
00339 
00344 static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
00345 {
00346     int k, n8, n4, n2, n, j;
00347     const uint16_t *revtab = s->revtab;
00348     const FFTSample *tcos = s->tcos;
00349     const FFTSample *tsin = s->tsin;
00350     const FFTSample *in1, *in2, *in3, *in4;
00351     FFTComplex *z = (FFTComplex *)output;
00352 
00353     int j1;
00354     const float *tcos1, *tsin1, *tcos2, *tsin2;
00355     float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
00356         temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
00357     FFTComplex *z1, *z2;
00358 
00359     n = 1 << s->mdct_bits;
00360     n2 = n >> 1;
00361     n4 = n >> 2;
00362     n8 = n >> 3;
00363 
00364     /* pre rotation */
00365     in1 = input;
00366     in2 = input + n2 - 1;
00367     in3 = input + 2;
00368     in4 = input + n2 - 3;
00369 
00370     tcos1 = tcos;
00371     tsin1 = tsin;
00372 
00373     /* n4 = 64 or 128 */
00374     for(k = 0; k < n4; k += 2) {
00375         j  = revtab[k    ];
00376         j1 = revtab[k + 1];
00377 
00378         __asm__ volatile (
00379             "lwc1           %[temp1],       0(%[in2])                           \t\n"
00380             "lwc1           %[temp2],       0(%[tcos1])                         \t\n"
00381             "lwc1           %[temp3],       0(%[tsin1])                         \t\n"
00382             "lwc1           %[temp4],       0(%[in1])                           \t\n"
00383             "lwc1           %[temp5],       0(%[in4])                           \t\n"
00384             "mul.s          %[temp9],       %[temp1],   %[temp2]                \t\n"
00385             "mul.s          %[temp10],      %[temp1],   %[temp3]                \t\n"
00386             "lwc1           %[temp6],       4(%[tcos1])                         \t\n"
00387             "lwc1           %[temp7],       4(%[tsin1])                         \t\n"
00388             "nmsub.s        %[temp9],       %[temp9],   %[temp4],   %[temp3]    \t\n"
00389             "madd.s         %[temp10],      %[temp10],  %[temp4],   %[temp2]    \t\n"
00390             "mul.s          %[temp11],      %[temp5],   %[temp6]                \t\n"
00391             "mul.s          %[temp12],      %[temp5],   %[temp7]                \t\n"
00392             "lwc1           %[temp8],       0(%[in3])                           \t\n"
00393             "addiu          %[tcos1],       %[tcos1],   8                       \t\n"
00394             "addiu          %[tsin1],       %[tsin1],   8                       \t\n"
00395             "addiu          %[in1],         %[in1],     16                      \t\n"
00396             "nmsub.s        %[temp11],      %[temp11],  %[temp8],   %[temp7]    \t\n"
00397             "madd.s         %[temp12],      %[temp12],  %[temp8],   %[temp6]    \t\n"
00398             "addiu          %[in2],         %[in2],     -16                     \t\n"
00399             "addiu          %[in3],         %[in3],     16                      \t\n"
00400             "addiu          %[in4],         %[in4],     -16                     \t\n"
00401 
00402             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
00403               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
00404               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
00405               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
00406               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
00407               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
00408               [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
00409               [in1]"+r"(in1), [in2]"+r"(in2),
00410               [in3]"+r"(in3), [in4]"+r"(in4)
00411         );
00412 
00413         z[j ].re = temp9;
00414         z[j ].im = temp10;
00415         z[j1].re = temp11;
00416         z[j1].im = temp12;
00417     }
00418 
00419     s->fft_calc(s, z);
00420 
00421     /* post rotation + reordering */
00422     /* n8 = 32 or 64 */
00423     for(k = 0; k < n8; k += 2) {
00424         tcos1 = &tcos[n8 - k - 2];
00425         tsin1 = &tsin[n8 - k - 2];
00426         tcos2 = &tcos[n8 + k];
00427         tsin2 = &tsin[n8 + k];
00428         z1 = &z[n8 - k - 2];
00429         z2 = &z[n8 + k    ];
00430 
00431         __asm__ volatile (
00432             "lwc1       %[temp1],   12(%[z1])                           \t\n"
00433             "lwc1       %[temp2],   4(%[tsin1])                         \t\n"
00434             "lwc1       %[temp3],   4(%[tcos1])                         \t\n"
00435             "lwc1       %[temp4],   8(%[z1])                            \t\n"
00436             "lwc1       %[temp5],   4(%[z1])                            \t\n"
00437             "mul.s      %[temp9],   %[temp1],   %[temp2]                \t\n"
00438             "mul.s      %[temp10],  %[temp1],   %[temp3]                \t\n"
00439             "lwc1       %[temp6],   0(%[tsin1])                         \t\n"
00440             "lwc1       %[temp7],   0(%[tcos1])                         \t\n"
00441             "nmsub.s    %[temp9],   %[temp9],   %[temp4],   %[temp3]    \t\n"
00442             "madd.s     %[temp10],  %[temp10],  %[temp4],   %[temp2]    \t\n"
00443             "mul.s      %[temp11],  %[temp5],   %[temp6]                \t\n"
00444             "mul.s      %[temp12],  %[temp5],   %[temp7]                \t\n"
00445             "lwc1       %[temp8],   0(%[z1])                            \t\n"
00446             "lwc1       %[temp1],   4(%[z2])                            \t\n"
00447             "lwc1       %[temp2],   0(%[tsin2])                         \t\n"
00448             "lwc1       %[temp3],   0(%[tcos2])                         \t\n"
00449             "nmsub.s    %[temp11],  %[temp11],  %[temp8],   %[temp7]    \t\n"
00450             "madd.s     %[temp12],  %[temp12],  %[temp8],   %[temp6]    \t\n"
00451             "mul.s      %[temp13],  %[temp1],   %[temp2]                \t\n"
00452             "mul.s      %[temp14],  %[temp1],   %[temp3]                \t\n"
00453             "lwc1       %[temp4],   0(%[z2])                            \t\n"
00454             "lwc1       %[temp5],   12(%[z2])                           \t\n"
00455             "lwc1       %[temp6],   4(%[tsin2])                         \t\n"
00456             "lwc1       %[temp7],   4(%[tcos2])                         \t\n"
00457             "nmsub.s    %[temp13],  %[temp13],  %[temp4],   %[temp3]    \t\n"
00458             "madd.s     %[temp14],  %[temp14],  %[temp4],   %[temp2]    \t\n"
00459             "mul.s      %[temp15],  %[temp5],   %[temp6]                \t\n"
00460             "mul.s      %[temp16],  %[temp5],   %[temp7]                \t\n"
00461             "lwc1       %[temp8],   8(%[z2])                            \t\n"
00462             "nmsub.s    %[temp15],  %[temp15],  %[temp8],   %[temp7]    \t\n"
00463             "madd.s     %[temp16],  %[temp16],  %[temp8],   %[temp6]    \t\n"
00464             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
00465               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
00466               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
00467               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
00468               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
00469               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
00470               [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
00471               [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
00472             : [z1]"r"(z1), [z2]"r"(z2),
00473               [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
00474               [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
00475         );
00476 
00477         z1[1].re = temp9;
00478         z1[1].im = temp14;
00479         z2[0].re = temp13;
00480         z2[0].im = temp10;
00481 
00482         z1[0].re = temp11;
00483         z1[0].im = temp16;
00484         z2[1].re = temp15;
00485         z2[1].im = temp12;
00486     }
00487 }
00488 #endif /* HAVE_INLINE_ASM */
00489 
00495 static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
00496 {
00497     int k;
00498     int n = 1 << s->mdct_bits;
00499     int n2 = n >> 1;
00500     int n4 = n >> 2;
00501 
00502     ff_imdct_half_mips(s, output+n4, input);
00503 
00504     for(k = 0; k < n4; k+=4) {
00505         output[k] = -output[n2-k-1];
00506         output[k+1] = -output[n2-k-2];
00507         output[k+2] = -output[n2-k-3];
00508         output[k+3] = -output[n2-k-4];
00509 
00510         output[n-k-1] = output[n2+k];
00511         output[n-k-2] = output[n2+k+1];
00512         output[n-k-3] = output[n2+k+2];
00513         output[n-k-4] = output[n2+k+3];
00514     }
00515 }
00516 
00517 av_cold void ff_fft_init_mips(FFTContext *s)
00518 {
00519     int n=0;
00520 
00521     ff_fft_lut_init(fft_offsets_lut, 0, 1 << 16, &n);
00522 
00523 #if HAVE_INLINE_ASM
00524     s->fft_calc     = ff_fft_calc_mips;
00525 #endif
00526 #if CONFIG_MDCT
00527     s->imdct_calc   = ff_imdct_calc_mips;
00528     s->imdct_half   = ff_imdct_half_mips;
00529 #endif
00530 }