yading@10: /*
yading@10:  * Copyright (c) 2012
yading@10:  *      MIPS Technologies, Inc., California.
yading@10:  *
yading@10:  * Redistribution and use in source and binary forms, with or without
yading@10:  * modification, are permitted provided that the following conditions
yading@10:  * are met:
yading@10:  * 1. Redistributions of source code must retain the above copyright
yading@10:  *    notice, this list of conditions and the following disclaimer.
yading@10:  * 2. Redistributions in binary form must reproduce the above copyright
yading@10:  *    notice, this list of conditions and the following disclaimer in the
yading@10:  *    documentation and/or other materials provided with the distribution.
yading@10:  * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
yading@10:  *    contributors may be used to endorse or promote products derived from
yading@10:  *    this software without specific prior written permission.
yading@10:  *
yading@10:  * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
yading@10:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
yading@10:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
yading@10:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
yading@10:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
yading@10:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
yading@10:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
yading@10:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
yading@10:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
yading@10:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
yading@10:  * SUCH DAMAGE.
yading@10:  *
yading@10:  * Author:  Stanislav Ocovaj (socovaj@mips.com)
yading@10:  * Author:  Zoran Lukic (zoranl@mips.com)
yading@10:  *
yading@10:  * Optimized MDCT/IMDCT and FFT transforms
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: #include "config.h"
yading@10: #include "libavcodec/fft.h"
yading@10: #include "fft_table.h"
yading@10: 
yading@10: /**
yading@10:  * FFT transform
yading@10:  */
yading@10: 
yading@10: #if HAVE_INLINE_ASM
yading@10: static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z)
yading@10: {
yading@10:     int nbits, i, n, num_transforms, offset, step;
yading@10:     int n4, n2, n34;
yading@10:     FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
yading@10:     FFTComplex *tmpz;
yading@10:     float w_re, w_im;
yading@10:     float *w_re_ptr, *w_im_ptr;
yading@10:     const int fft_size = (1 << s->nbits);
yading@10:     int s_n = s->nbits;
yading@10:     int tem1, tem2;
yading@10:     float pom,  pom1,  pom2,  pom3;
yading@10:     float temp, temp1, temp3, temp4;
yading@10:     FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4;
yading@10:     FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i;
yading@10: 
yading@10:     /**
yading@10:     *num_transforms = (0x2aab >> (16 - s->nbits)) | 1;
yading@10:     */
yading@10:     __asm__ volatile (
yading@10:         "li   %[tem1], 16                                      \n\t"
yading@10:         "sub  %[s_n],  %[tem1], %[s_n]                         \n\t"
yading@10:         "li   %[tem2], 10923                                   \n\t"
yading@10:         "srav %[tem2], %[tem2], %[s_n]                         \n\t"
yading@10:         "ori  %[num_t],%[tem2], 1                              \n\t"
yading@10:         : [num_t]"=r"(num_transforms), [s_n]"+r"(s_n),
yading@10:           [tem1]"=&r"(tem1), [tem2]"=&r"(tem2)
yading@10:     );
yading@10: 
yading@10: 
yading@10:     for (n=0; n<num_transforms; n++) {
yading@10:         offset = fft_offsets_lut[n] << 2;
yading@10:         tmpz = z + offset;
yading@10: 
yading@10:         tmp1 = tmpz[0].re + tmpz[1].re;
yading@10:         tmp5 = tmpz[2].re + tmpz[3].re;
yading@10:         tmp2 = tmpz[0].im + tmpz[1].im;
yading@10:         tmp6 = tmpz[2].im + tmpz[3].im;
yading@10:         tmp3 = tmpz[0].re - tmpz[1].re;
yading@10:         tmp8 = tmpz[2].im - tmpz[3].im;
yading@10:         tmp4 = tmpz[0].im - tmpz[1].im;
yading@10:         tmp7 = tmpz[2].re - tmpz[3].re;
yading@10: 
yading@10:         tmpz[0].re = tmp1 + tmp5;
yading@10:         tmpz[2].re = tmp1 - tmp5;
yading@10:         tmpz[0].im = tmp2 + tmp6;
yading@10:         tmpz[2].im = tmp2 - tmp6;
yading@10:         tmpz[1].re = tmp3 + tmp8;
yading@10:         tmpz[3].re = tmp3 - tmp8;
yading@10:         tmpz[1].im = tmp4 - tmp7;
yading@10:         tmpz[3].im = tmp4 + tmp7;
yading@10: 
yading@10:     }
yading@10: 
yading@10:     if (fft_size < 8)
yading@10:         return;
yading@10: 
yading@10:     num_transforms = (num_transforms >> 1) | 1;
yading@10: 
yading@10:     for (n=0; n<num_transforms; n++) {
yading@10:         offset = fft_offsets_lut[n] << 3;
yading@10:         tmpz = z + offset;
yading@10: 
yading@10:         __asm__ volatile (
yading@10:             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom1], 56(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom2], 44(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
yading@10:             "add.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re + tmpz[5].re;
yading@10:             "add.s %[tmp3], %[tmp3],    %[pom1]             \n\t"  // tmp3 = tmpz[6].re + tmpz[7].re;
yading@10:             "add.s %[tmp2], %[tmp2],    %[pom2]             \n\t"  // tmp2 = tmpz[4].im + tmpz[5].im;
yading@10:             "lwc1  %[pom],  40(%[tmpz])                     \n\t"
yading@10:             "add.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im + tmpz[7].im;
yading@10:             "add.s %[tmp5], %[tmp1],    %[tmp3]             \n\t"  // tmp5 = tmp1 + tmp3;
yading@10:             "sub.s %[tmp7], %[tmp1],    %[tmp3]             \n\t"  // tmp7 = tmp1 - tmp3;
yading@10:             "lwc1  %[tmp1], 32(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom1], 44(%[tmpz])                     \n\t"
yading@10:             "add.s %[tmp6], %[tmp2],    %[tmp4]             \n\t"  // tmp6 = tmp2 + tmp4;
yading@10:             "sub.s %[tmp8], %[tmp2],    %[tmp4]             \n\t"  // tmp8 = tmp2 - tmp4;
yading@10:             "lwc1  %[tmp2], 36(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom2], 56(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom3], 60(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[tmp3], 48(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[tmp4], 52(%[tmpz])                     \n\t"
yading@10:             "sub.s %[tmp1], %[tmp1],    %[pom]              \n\t"  // tmp1 = tmpz[4].re - tmpz[5].re;
yading@10:             "lwc1  %[pom],  0(%[tmpz])                      \n\t"
yading@10:             "sub.s %[tmp2], %[tmp2],    %[pom1]             \n\t"  // tmp2 = tmpz[4].im - tmpz[5].im;
yading@10:             "sub.s %[tmp3], %[tmp3],    %[pom2]             \n\t"  // tmp3 = tmpz[6].re - tmpz[7].re;
yading@10:             "lwc1  %[pom2], 4(%[tmpz])                      \n\t"
yading@10:             "sub.s %[pom1], %[pom],     %[tmp5]             \n\t"
yading@10:             "sub.s %[tmp4], %[tmp4],    %[pom3]             \n\t"  // tmp4 = tmpz[6].im - tmpz[7].im;
yading@10:             "add.s %[pom3], %[pom],     %[tmp5]             \n\t"
yading@10:             "sub.s %[pom],  %[pom2],    %[tmp6]             \n\t"
yading@10:             "add.s %[pom2], %[pom2],    %[tmp6]             \n\t"
yading@10:             "swc1  %[pom1], 32(%[tmpz])                     \n\t"  // tmpz[4].re = tmpz[0].re - tmp5;
yading@10:             "swc1  %[pom3], 0(%[tmpz])                      \n\t"  // tmpz[0].re = tmpz[0].re + tmp5;
yading@10:             "swc1  %[pom],  36(%[tmpz])                     \n\t"  // tmpz[4].im = tmpz[0].im - tmp6;
yading@10:             "swc1  %[pom2], 4(%[tmpz])                      \n\t"  // tmpz[0].im = tmpz[0].im + tmp6;
yading@10:             "lwc1  %[pom1], 16(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom3], 20(%[tmpz])                     \n\t"
yading@10:             "li.s  %[pom],  0.7071067812                    \n\t"  // float pom = 0.7071067812f;
yading@10:             "add.s %[temp1],%[tmp1],    %[tmp2]             \n\t"
yading@10:             "sub.s %[temp], %[pom1],    %[tmp8]             \n\t"
yading@10:             "add.s %[pom2], %[pom3],    %[tmp7]             \n\t"
yading@10:             "sub.s %[temp3],%[tmp3],    %[tmp4]             \n\t"
yading@10:             "sub.s %[temp4],%[tmp2],    %[tmp1]             \n\t"
yading@10:             "swc1  %[temp], 48(%[tmpz])                     \n\t"  // tmpz[6].re = tmpz[2].re - tmp8;
yading@10:             "swc1  %[pom2], 52(%[tmpz])                     \n\t"  // tmpz[6].im = tmpz[2].im + tmp7;
yading@10:             "add.s %[pom1], %[pom1],    %[tmp8]             \n\t"
yading@10:             "sub.s %[pom3], %[pom3],    %[tmp7]             \n\t"
yading@10:             "add.s %[tmp3], %[tmp3],    %[tmp4]             \n\t"
yading@10:             "mul.s %[tmp5], %[pom],     %[temp1]            \n\t"  // tmp5 = pom * (tmp1 + tmp2);
yading@10:             "mul.s %[tmp7], %[pom],     %[temp3]            \n\t"  // tmp7 = pom * (tmp3 - tmp4);
yading@10:             "mul.s %[tmp6], %[pom],     %[temp4]            \n\t"  // tmp6 = pom * (tmp2 - tmp1);
yading@10:             "mul.s %[tmp8], %[pom],     %[tmp3]             \n\t"  // tmp8 = pom * (tmp3 + tmp4);
yading@10:             "swc1  %[pom1], 16(%[tmpz])                     \n\t"  // tmpz[2].re = tmpz[2].re + tmp8;
yading@10:             "swc1  %[pom3], 20(%[tmpz])                     \n\t"  // tmpz[2].im = tmpz[2].im - tmp7;
yading@10:             "add.s %[tmp1], %[tmp5],    %[tmp7]             \n\t"  // tmp1 = tmp5 + tmp7;
yading@10:             "sub.s %[tmp3], %[tmp5],    %[tmp7]             \n\t"  // tmp3 = tmp5 - tmp7;
yading@10:             "add.s %[tmp2], %[tmp6],    %[tmp8]             \n\t"  // tmp2 = tmp6 + tmp8;
yading@10:             "sub.s %[tmp4], %[tmp6],    %[tmp8]             \n\t"  // tmp4 = tmp6 - tmp8;
yading@10:             "lwc1  %[temp], 8(%[tmpz])                      \n\t"
yading@10:             "lwc1  %[temp1],12(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom],  24(%[tmpz])                     \n\t"
yading@10:             "lwc1  %[pom2], 28(%[tmpz])                     \n\t"
yading@10:             "sub.s %[temp4],%[temp],    %[tmp1]             \n\t"
yading@10:             "sub.s %[temp3],%[temp1],   %[tmp2]             \n\t"
yading@10:             "add.s %[temp], %[temp],    %[tmp1]             \n\t"
yading@10:             "add.s %[temp1],%[temp1],   %[tmp2]             \n\t"
yading@10:             "sub.s %[pom1], %[pom],     %[tmp4]             \n\t"
yading@10:             "add.s %[pom3], %[pom2],    %[tmp3]             \n\t"
yading@10:             "add.s %[pom],  %[pom],     %[tmp4]             \n\t"
yading@10:             "sub.s %[pom2], %[pom2],    %[tmp3]             \n\t"
yading@10:             "swc1  %[temp4],40(%[tmpz])                     \n\t"  // tmpz[5].re = tmpz[1].re - tmp1;
yading@10:             "swc1  %[temp3],44(%[tmpz])                     \n\t"  // tmpz[5].im = tmpz[1].im - tmp2;
yading@10:             "swc1  %[temp], 8(%[tmpz])                      \n\t"  // tmpz[1].re = tmpz[1].re + tmp1;
yading@10:             "swc1  %[temp1],12(%[tmpz])                     \n\t"  // tmpz[1].im = tmpz[1].im + tmp2;
yading@10:             "swc1  %[pom1], 56(%[tmpz])                     \n\t"  // tmpz[7].re = tmpz[3].re - tmp4;
yading@10:             "swc1  %[pom3], 60(%[tmpz])                     \n\t"  // tmpz[7].im = tmpz[3].im + tmp3;
yading@10:             "swc1  %[pom],  24(%[tmpz])                     \n\t"  // tmpz[3].re = tmpz[3].re + tmp4;
yading@10:             "swc1  %[pom2], 28(%[tmpz])                     \n\t"  // tmpz[3].im = tmpz[3].im - tmp3;
yading@10:             : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),   [pom1]"=&f"(pom1), [pom2]"=&f"(pom2),
yading@10:               [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5),  [tmp7]"=&f"(tmp7),
yading@10:               [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1),
yading@10:               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4)
yading@10:             : [tmpz]"r"(tmpz)
yading@10:             : "memory"
yading@10:         );
yading@10:     }
yading@10: 
yading@10:     step = 1 << (MAX_LOG2_NFFT - 4);
yading@10:     n4 = 4;
yading@10: 
yading@10:     for (nbits=4; nbits<=s->nbits; nbits++) {
yading@10:         /*
yading@10:         * num_transforms = (num_transforms >> 1) | 1;
yading@10:         */
yading@10:         __asm__ volatile (
yading@10:             "sra %[num_t], %[num_t], 1               \n\t"
yading@10:             "ori %[num_t], %[num_t], 1               \n\t"
yading@10: 
yading@10:             : [num_t] "+r" (num_transforms)
yading@10:         );
yading@10:         n2  = 2 * n4;
yading@10:         n34 = 3 * n4;
yading@10: 
yading@10:         for (n=0; n<num_transforms; n++) {
yading@10:             offset = fft_offsets_lut[n] << nbits;
yading@10:             tmpz = z + offset;
yading@10: 
yading@10:             tmpz_n2  = tmpz +  n2;
yading@10:             tmpz_n4  = tmpz +  n4;
yading@10:             tmpz_n34 = tmpz +  n34;
yading@10: 
yading@10:             __asm__ volatile (
yading@10:                 "lwc1  %[pom1], 0(%[tmpz_n2])            \n\t"
yading@10:                 "lwc1  %[pom],  0(%[tmpz_n34])           \n\t"
yading@10:                 "lwc1  %[pom2], 4(%[tmpz_n2])            \n\t"
yading@10:                 "lwc1  %[pom3], 4(%[tmpz_n34])           \n\t"
yading@10:                 "lwc1  %[temp1],0(%[tmpz])               \n\t"
yading@10:                 "lwc1  %[temp3],4(%[tmpz])               \n\t"
yading@10:                 "add.s %[tmp5], %[pom1],      %[pom]     \n\t"   //  tmp5 = tmpz[ n2].re + tmpz[n34].re;
yading@10:                 "sub.s %[tmp1], %[pom1],      %[pom]     \n\t"   //  tmp1 = tmpz[ n2].re - tmpz[n34].re;
yading@10:                 "add.s %[tmp6], %[pom2],      %[pom3]    \n\t"   //  tmp6 = tmpz[ n2].im + tmpz[n34].im;
yading@10:                 "sub.s %[tmp2], %[pom2],      %[pom3]    \n\t"   //  tmp2 = tmpz[ n2].im - tmpz[n34].im;
yading@10:                 "sub.s %[temp], %[temp1],     %[tmp5]    \n\t"
yading@10:                 "add.s %[temp1],%[temp1],     %[tmp5]    \n\t"
yading@10:                 "sub.s %[temp4],%[temp3],     %[tmp6]    \n\t"
yading@10:                 "add.s %[temp3],%[temp3],     %[tmp6]    \n\t"
yading@10:                 "swc1  %[temp], 0(%[tmpz_n2])            \n\t"   //  tmpz[ n2].re = tmpz[ 0].re - tmp5;
yading@10:                 "swc1  %[temp1],0(%[tmpz])               \n\t"   //  tmpz[  0].re = tmpz[ 0].re + tmp5;
yading@10:                 "lwc1  %[pom1], 0(%[tmpz_n4])            \n\t"
yading@10:                 "swc1  %[temp4],4(%[tmpz_n2])            \n\t"   //  tmpz[ n2].im = tmpz[ 0].im - tmp6;
yading@10:                 "lwc1  %[temp], 4(%[tmpz_n4])            \n\t"
yading@10:                 "swc1  %[temp3],4(%[tmpz])               \n\t"   //  tmpz[  0].im = tmpz[ 0].im + tmp6;
yading@10:                 "sub.s %[pom],  %[pom1],      %[tmp2]    \n\t"
yading@10:                 "add.s %[pom1], %[pom1],      %[tmp2]    \n\t"
yading@10:                 "add.s %[temp1],%[temp],      %[tmp1]    \n\t"
yading@10:                 "sub.s %[temp], %[temp],      %[tmp1]    \n\t"
yading@10:                 "swc1  %[pom],  0(%[tmpz_n34])           \n\t"   //  tmpz[n34].re = tmpz[n4].re - tmp2;
yading@10:                 "swc1  %[pom1], 0(%[tmpz_n4])            \n\t"   //  tmpz[ n4].re = tmpz[n4].re + tmp2;
yading@10:                 "swc1  %[temp1],4(%[tmpz_n34])           \n\t"   //  tmpz[n34].im = tmpz[n4].im + tmp1;
yading@10:                 "swc1  %[temp], 4(%[tmpz_n4])            \n\t"   //  tmpz[ n4].im = tmpz[n4].im - tmp1;
yading@10:                 : [tmp5]"=&f"(tmp5),
yading@10:                   [tmp1]"=&f"(tmp1), [pom]"=&f"(pom),        [pom1]"=&f"(pom1),        [pom2]"=&f"(pom2),
yading@10:                   [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6),          [pom3]"=&f"(pom3),
yading@10:                   [temp]"=&f"(temp), [temp1]"=&f"(temp1),     [temp3]"=&f"(temp3),       [temp4]"=&f"(temp4)
yading@10:                 : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4)
yading@10:                 : "memory"
yading@10:             );
yading@10: 
yading@10:             w_re_ptr = (float*)(ff_cos_65536 + step);
yading@10:             w_im_ptr = (float*)(ff_cos_65536 + MAX_FFT_SIZE/4 - step);
yading@10: 
yading@10:             for (i=1; i<n4; i++) {
yading@10:                 w_re = w_re_ptr[0];
yading@10:                 w_im = w_im_ptr[0];
yading@10:                 tmpz_n2_i = tmpz_n2  + i;
yading@10:                 tmpz_n4_i = tmpz_n4  + i;
yading@10:                 tmpz_n34_i= tmpz_n34 + i;
yading@10:                 tmpz_i    = tmpz     + i;
yading@10: 
yading@10:                 __asm__ volatile (
yading@10:                     "lwc1     %[temp],  0(%[tmpz_n2_i])               \n\t"
yading@10:                     "lwc1     %[temp1], 4(%[tmpz_n2_i])               \n\t"
yading@10:                     "lwc1     %[pom],   0(%[tmpz_n34_i])              \n\t"
yading@10:                     "lwc1     %[pom1],  4(%[tmpz_n34_i])              \n\t"
yading@10:                     "mul.s    %[temp3], %[w_im],    %[temp]           \n\t"
yading@10:                     "mul.s    %[temp4], %[w_im],    %[temp1]          \n\t"
yading@10:                     "mul.s    %[pom2],  %[w_im],    %[pom1]           \n\t"
yading@10:                     "mul.s    %[pom3],  %[w_im],    %[pom]            \n\t"
yading@10:                     "msub.s   %[tmp2],  %[temp3],   %[w_re], %[temp1] \n\t"  // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re;
yading@10:                     "madd.s   %[tmp1],  %[temp4],   %[w_re], %[temp]  \n\t"  // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im;
yading@10:                     "msub.s   %[tmp3],  %[pom2],    %[w_re], %[pom]   \n\t"  // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im;
yading@10:                     "madd.s   %[tmp4],  %[pom3],    %[w_re], %[pom1]  \n\t"  // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re;
yading@10:                     "lwc1     %[temp],  0(%[tmpz_i])                  \n\t"
yading@10:                     "lwc1     %[pom],   4(%[tmpz_i])                  \n\t"
yading@10:                     "add.s    %[tmp5],  %[tmp1],    %[tmp3]           \n\t"  // tmp5 = tmp1 + tmp3;
yading@10:                     "sub.s    %[tmp1],  %[tmp1],    %[tmp3]           \n\t"  // tmp1 = tmp1 - tmp3;
yading@10:                     "add.s    %[tmp6],  %[tmp2],    %[tmp4]           \n\t"  // tmp6 = tmp2 + tmp4;
yading@10:                     "sub.s    %[tmp2],  %[tmp2],    %[tmp4]           \n\t"  // tmp2 = tmp2 - tmp4;
yading@10:                     "sub.s    %[temp1], %[temp],    %[tmp5]           \n\t"
yading@10:                     "add.s    %[temp],  %[temp],    %[tmp5]           \n\t"
yading@10:                     "sub.s    %[pom1],  %[pom],     %[tmp6]           \n\t"
yading@10:                     "add.s    %[pom],   %[pom],     %[tmp6]           \n\t"
yading@10:                     "lwc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"
yading@10:                     "lwc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"
yading@10:                     "swc1     %[temp1], 0(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].re = tmpz[   i].re - tmp5;
yading@10:                     "swc1     %[temp],  0(%[tmpz_i])                  \n\t"  // tmpz[    i].re = tmpz[   i].re + tmp5;
yading@10:                     "swc1     %[pom1],  4(%[tmpz_n2_i])               \n\t"  // tmpz[ n2+i].im = tmpz[   i].im - tmp6;
yading@10:                     "swc1     %[pom] ,  4(%[tmpz_i])                  \n\t"  // tmpz[    i].im = tmpz[   i].im + tmp6;
yading@10:                     "sub.s    %[temp4], %[temp3],   %[tmp2]           \n\t"
yading@10:                     "add.s    %[pom3],  %[pom2],    %[tmp1]           \n\t"
yading@10:                     "add.s    %[temp3], %[temp3],   %[tmp2]           \n\t"
yading@10:                     "sub.s    %[pom2],  %[pom2],    %[tmp1]           \n\t"
yading@10:                     "swc1     %[temp4], 0(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
yading@10:                     "swc1     %[pom3],  4(%[tmpz_n34_i])              \n\t"  // tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
yading@10:                     "swc1     %[temp3], 0(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
yading@10:                     "swc1     %[pom2],  4(%[tmpz_n4_i])               \n\t"  // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
yading@10:                     : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3),
yading@10:                       [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6),
yading@10:                       [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
yading@10:                       [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3)
yading@10:                     : [w_re]"f"(w_re), [w_im]"f"(w_im),
yading@10:                       [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i),
yading@10:                       [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i)
yading@10:                     : "memory"
yading@10:                 );
yading@10:                 w_re_ptr += step;
yading@10:                 w_im_ptr -= step;
yading@10:             }
yading@10:         }
yading@10:         step >>= 1;
yading@10:         n4   <<= 1;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * MDCT/IMDCT transforms.
yading@10:  */
yading@10: 
yading@10: static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
yading@10: {
yading@10:     int k, n8, n4, n2, n, j;
yading@10:     const uint16_t *revtab = s->revtab;
yading@10:     const FFTSample *tcos = s->tcos;
yading@10:     const FFTSample *tsin = s->tsin;
yading@10:     const FFTSample *in1, *in2, *in3, *in4;
yading@10:     FFTComplex *z = (FFTComplex *)output;
yading@10: 
yading@10:     int j1;
yading@10:     const float *tcos1, *tsin1, *tcos2, *tsin2;
yading@10:     float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8,
yading@10:         temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16;
yading@10:     FFTComplex *z1, *z2;
yading@10: 
yading@10:     n = 1 << s->mdct_bits;
yading@10:     n2 = n >> 1;
yading@10:     n4 = n >> 2;
yading@10:     n8 = n >> 3;
yading@10: 
yading@10:     /* pre rotation */
yading@10:     in1 = input;
yading@10:     in2 = input + n2 - 1;
yading@10:     in3 = input + 2;
yading@10:     in4 = input + n2 - 3;
yading@10: 
yading@10:     tcos1 = tcos;
yading@10:     tsin1 = tsin;
yading@10: 
yading@10:     /* n4 = 64 or 128 */
yading@10:     for(k = 0; k < n4; k += 2) {
yading@10:         j  = revtab[k    ];
yading@10:         j1 = revtab[k + 1];
yading@10: 
yading@10:         __asm__ volatile (
yading@10:             "lwc1           %[temp1],       0(%[in2])                           \t\n"
yading@10:             "lwc1           %[temp2],       0(%[tcos1])                         \t\n"
yading@10:             "lwc1           %[temp3],       0(%[tsin1])                         \t\n"
yading@10:             "lwc1           %[temp4],       0(%[in1])                           \t\n"
yading@10:             "lwc1           %[temp5],       0(%[in4])                           \t\n"
yading@10:             "mul.s          %[temp9],       %[temp1],   %[temp2]                \t\n"
yading@10:             "mul.s          %[temp10],      %[temp1],   %[temp3]                \t\n"
yading@10:             "lwc1           %[temp6],       4(%[tcos1])                         \t\n"
yading@10:             "lwc1           %[temp7],       4(%[tsin1])                         \t\n"
yading@10:             "nmsub.s        %[temp9],       %[temp9],   %[temp4],   %[temp3]    \t\n"
yading@10:             "madd.s         %[temp10],      %[temp10],  %[temp4],   %[temp2]    \t\n"
yading@10:             "mul.s          %[temp11],      %[temp5],   %[temp6]                \t\n"
yading@10:             "mul.s          %[temp12],      %[temp5],   %[temp7]                \t\n"
yading@10:             "lwc1           %[temp8],       0(%[in3])                           \t\n"
yading@10:             "addiu          %[tcos1],       %[tcos1],   8                       \t\n"
yading@10:             "addiu          %[tsin1],       %[tsin1],   8                       \t\n"
yading@10:             "addiu          %[in1],         %[in1],     16                      \t\n"
yading@10:             "nmsub.s        %[temp11],      %[temp11],  %[temp8],   %[temp7]    \t\n"
yading@10:             "madd.s         %[temp12],      %[temp12],  %[temp8],   %[temp6]    \t\n"
yading@10:             "addiu          %[in2],         %[in2],     -16                     \t\n"
yading@10:             "addiu          %[in3],         %[in3],     16                      \t\n"
yading@10:             "addiu          %[in4],         %[in4],     -16                     \t\n"
yading@10: 
yading@10:             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
yading@10:               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
yading@10:               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
yading@10:               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
yading@10:               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
yading@10:               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
yading@10:               [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1),
yading@10:               [in1]"+r"(in1), [in2]"+r"(in2),
yading@10:               [in3]"+r"(in3), [in4]"+r"(in4)
yading@10:             :
yading@10:             : "memory"
yading@10:         );
yading@10: 
yading@10:         z[j ].re = temp9;
yading@10:         z[j ].im = temp10;
yading@10:         z[j1].re = temp11;
yading@10:         z[j1].im = temp12;
yading@10:     }
yading@10: 
yading@10:     s->fft_calc(s, z);
yading@10: 
yading@10:     /* post rotation + reordering */
yading@10:     /* n8 = 32 or 64 */
yading@10:     for(k = 0; k < n8; k += 2) {
yading@10:         tcos1 = &tcos[n8 - k - 2];
yading@10:         tsin1 = &tsin[n8 - k - 2];
yading@10:         tcos2 = &tcos[n8 + k];
yading@10:         tsin2 = &tsin[n8 + k];
yading@10:         z1 = &z[n8 - k - 2];
yading@10:         z2 = &z[n8 + k    ];
yading@10: 
yading@10:         __asm__ volatile (
yading@10:             "lwc1       %[temp1],   12(%[z1])                           \t\n"
yading@10:             "lwc1       %[temp2],   4(%[tsin1])                         \t\n"
yading@10:             "lwc1       %[temp3],   4(%[tcos1])                         \t\n"
yading@10:             "lwc1       %[temp4],   8(%[z1])                            \t\n"
yading@10:             "lwc1       %[temp5],   4(%[z1])                            \t\n"
yading@10:             "mul.s      %[temp9],   %[temp1],   %[temp2]                \t\n"
yading@10:             "mul.s      %[temp10],  %[temp1],   %[temp3]                \t\n"
yading@10:             "lwc1       %[temp6],   0(%[tsin1])                         \t\n"
yading@10:             "lwc1       %[temp7],   0(%[tcos1])                         \t\n"
yading@10:             "nmsub.s    %[temp9],   %[temp9],   %[temp4],   %[temp3]    \t\n"
yading@10:             "madd.s     %[temp10],  %[temp10],  %[temp4],   %[temp2]    \t\n"
yading@10:             "mul.s      %[temp11],  %[temp5],   %[temp6]                \t\n"
yading@10:             "mul.s      %[temp12],  %[temp5],   %[temp7]                \t\n"
yading@10:             "lwc1       %[temp8],   0(%[z1])                            \t\n"
yading@10:             "lwc1       %[temp1],   4(%[z2])                            \t\n"
yading@10:             "lwc1       %[temp2],   0(%[tsin2])                         \t\n"
yading@10:             "lwc1       %[temp3],   0(%[tcos2])                         \t\n"
yading@10:             "nmsub.s    %[temp11],  %[temp11],  %[temp8],   %[temp7]    \t\n"
yading@10:             "madd.s     %[temp12],  %[temp12],  %[temp8],   %[temp6]    \t\n"
yading@10:             "mul.s      %[temp13],  %[temp1],   %[temp2]                \t\n"
yading@10:             "mul.s      %[temp14],  %[temp1],   %[temp3]                \t\n"
yading@10:             "lwc1       %[temp4],   0(%[z2])                            \t\n"
yading@10:             "lwc1       %[temp5],   12(%[z2])                           \t\n"
yading@10:             "lwc1       %[temp6],   4(%[tsin2])                         \t\n"
yading@10:             "lwc1       %[temp7],   4(%[tcos2])                         \t\n"
yading@10:             "nmsub.s    %[temp13],  %[temp13],  %[temp4],   %[temp3]    \t\n"
yading@10:             "madd.s     %[temp14],  %[temp14],  %[temp4],   %[temp2]    \t\n"
yading@10:             "mul.s      %[temp15],  %[temp5],   %[temp6]                \t\n"
yading@10:             "mul.s      %[temp16],  %[temp5],   %[temp7]                \t\n"
yading@10:             "lwc1       %[temp8],   8(%[z2])                            \t\n"
yading@10:             "nmsub.s    %[temp15],  %[temp15],  %[temp8],   %[temp7]    \t\n"
yading@10:             "madd.s     %[temp16],  %[temp16],  %[temp8],   %[temp6]    \t\n"
yading@10:             : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
yading@10:               [temp3]"=&f"(temp3), [temp4]"=&f"(temp4),
yading@10:               [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
yading@10:               [temp7]"=&f"(temp7), [temp8]"=&f"(temp8),
yading@10:               [temp9]"=&f"(temp9), [temp10]"=&f"(temp10),
yading@10:               [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
yading@10:               [temp13]"=&f"(temp13), [temp14]"=&f"(temp14),
yading@10:               [temp15]"=&f"(temp15), [temp16]"=&f"(temp16)
yading@10:             : [z1]"r"(z1), [z2]"r"(z2),
yading@10:               [tsin1]"r"(tsin1), [tcos1]"r"(tcos1),
yading@10:               [tsin2]"r"(tsin2), [tcos2]"r"(tcos2)
yading@10:             : "memory"
yading@10:         );
yading@10: 
yading@10:         z1[1].re = temp9;
yading@10:         z1[1].im = temp14;
yading@10:         z2[0].re = temp13;
yading@10:         z2[0].im = temp10;
yading@10: 
yading@10:         z1[0].re = temp11;
yading@10:         z1[0].im = temp16;
yading@10:         z2[1].re = temp15;
yading@10:         z2[1].im = temp12;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Compute inverse MDCT of size N = 2^nbits
yading@10:  * @param output N samples
yading@10:  * @param input N/2 samples
yading@10:  */
yading@10: static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input)
yading@10: {
yading@10:     int k;
yading@10:     int n = 1 << s->mdct_bits;
yading@10:     int n2 = n >> 1;
yading@10:     int n4 = n >> 2;
yading@10: 
yading@10:     ff_imdct_half_mips(s, output+n4, input);
yading@10: 
yading@10:     for(k = 0; k < n4; k+=4) {
yading@10:         output[k] = -output[n2-k-1];
yading@10:         output[k+1] = -output[n2-k-2];
yading@10:         output[k+2] = -output[n2-k-3];
yading@10:         output[k+3] = -output[n2-k-4];
yading@10: 
yading@10:         output[n-k-1] = output[n2+k];
yading@10:         output[n-k-2] = output[n2+k+1];
yading@10:         output[n-k-3] = output[n2+k+2];
yading@10:         output[n-k-4] = output[n2+k+3];
yading@10:     }
yading@10: }
yading@10: #endif /* HAVE_INLINE_ASM */
yading@10: 
yading@10: av_cold void ff_fft_init_mips(FFTContext *s)
yading@10: {
yading@10:     int n=0;
yading@10: 
yading@10:     ff_fft_lut_init(fft_offsets_lut, 0, 1 << 16, &n);
yading@10:     ff_init_ff_cos_tabs(16);
yading@10: 
yading@10: #if HAVE_INLINE_ASM
yading@10:     s->fft_calc     = ff_fft_calc_mips;
yading@10: #if CONFIG_MDCT
yading@10:     s->imdct_calc   = ff_imdct_calc_mips;
yading@10:     s->imdct_half   = ff_imdct_half_mips;
yading@10: #endif
yading@10: #endif
yading@10: }