FFmpeg: libavcodec/mips/celp_filters

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:  Nedeljko Babic (nbabic@mips.com)
00030  *
00031  * various filters for CELP-based codecs optimized for MIPS
00032  *
00033  * This file is part of FFmpeg.
00034  *
00035  * FFmpeg is free software; you can redistribute it and/or
00036  * modify it under the terms of the GNU Lesser General Public
00037  * License as published by the Free Software Foundation; either
00038  * version 2.1 of the License, or (at your option) any later version.
00039  *
00040  * FFmpeg is distributed in the hope that it will be useful,
00041  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00042  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00043  * Lesser General Public License for more details.
00044  *
00045  * You should have received a copy of the GNU Lesser General Public
00046  * License along with FFmpeg; if not, write to the Free Software
00047  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00048  */
00049 
00054 #include "libavutil/attributes.h"
00055 #include "libavutil/common.h"
00056 #include "libavcodec/celp_filters.h"
00057 
00058 static void ff_celp_lp_synthesis_filterf_mips(float *out,
00059                                   const float *filter_coeffs,
00060                                   const float* in, int buffer_length,
00061                                   int filter_length)
00062 {
00063     int i,n;
00064 
00065     float out0, out1, out2, out3;
00066     float old_out0, old_out1, old_out2, old_out3;
00067     float a,b,c;
00068     const float *p_filter_coeffs;
00069     float *p_out;
00070 
00071     a = filter_coeffs[0];
00072     b = filter_coeffs[1];
00073     c = filter_coeffs[2];
00074     b -= filter_coeffs[0] * filter_coeffs[0];
00075     c -= filter_coeffs[1] * filter_coeffs[0];
00076     c -= filter_coeffs[0] * b;
00077 
00078     old_out0 = out[-4];
00079     old_out1 = out[-3];
00080     old_out2 = out[-2];
00081     old_out3 = out[-1];
00082     for (n = 0; n <= buffer_length - 4; n+=4) {
00083         p_filter_coeffs = filter_coeffs;
00084         p_out = out;
00085 
00086         out0 = in[0];
00087         out1 = in[1];
00088         out2 = in[2];
00089         out3 = in[3];
00090 
00091         __asm__ volatile(
00092             "lwc1       $f2,     8(%[filter_coeffs])                        \n\t"
00093             "lwc1       $f1,     4(%[filter_coeffs])                        \n\t"
00094             "lwc1       $f0,     0(%[filter_coeffs])                        \n\t"
00095             "nmsub.s    %[out0], %[out0],             $f2, %[old_out1]      \n\t"
00096             "nmsub.s    %[out1], %[out1],             $f2, %[old_out2]      \n\t"
00097             "nmsub.s    %[out2], %[out2],             $f2, %[old_out3]      \n\t"
00098             "lwc1       $f3,     12(%[filter_coeffs])                       \n\t"
00099             "nmsub.s    %[out0], %[out0],             $f1, %[old_out2]      \n\t"
00100             "nmsub.s    %[out1], %[out1],             $f1, %[old_out3]      \n\t"
00101             "nmsub.s    %[out2], %[out2],             $f3, %[old_out2]      \n\t"
00102             "nmsub.s    %[out0], %[out0],             $f0, %[old_out3]      \n\t"
00103             "nmsub.s    %[out3], %[out3],             $f3, %[old_out3]      \n\t"
00104             "nmsub.s    %[out1], %[out1],             $f3, %[old_out1]      \n\t"
00105             "nmsub.s    %[out0], %[out0],             $f3, %[old_out0]      \n\t"
00106 
00107             : [out0]"+f"(out0), [out1]"+f"(out1),
00108               [out2]"+f"(out2), [out3]"+f"(out3)
00109             : [old_out0]"f"(old_out0), [old_out1]"f"(old_out1),
00110               [old_out2]"f"(old_out2), [old_out3]"f"(old_out3),
00111               [filter_coeffs]"r"(filter_coeffs)
00112             : "$f0", "$f1", "$f2", "$f3", "$f4"
00113         );
00114 
00115         for (i = 5; i <= filter_length; i += 2) {
00116             __asm__ volatile(
00117                 "lwc1    %[old_out3], -20(%[p_out])                         \n\t"
00118                 "lwc1    $f5,         16(%[p_filter_coeffs])                \n\t"
00119                 "addiu   %[p_out],    -8                                    \n\t"
00120                 "addiu   %[p_filter_coeffs], 8                              \n\t"
00121                 "nmsub.s %[out1],     %[out1],      $f5, %[old_out0]        \n\t"
00122                 "nmsub.s %[out3],     %[out3],      $f5, %[old_out2]        \n\t"
00123                 "lwc1    $f4,         12(%[p_filter_coeffs])                \n\t"
00124                 "lwc1    %[old_out2], -16(%[p_out])                         \n\t"
00125                 "nmsub.s %[out0],     %[out0],      $f5, %[old_out3]        \n\t"
00126                 "nmsub.s %[out2],     %[out2],      $f5, %[old_out1]        \n\t"
00127                 "nmsub.s %[out1],     %[out1],      $f4, %[old_out3]        \n\t"
00128                 "nmsub.s %[out3],     %[out3],      $f4, %[old_out1]        \n\t"
00129                 "mov.s   %[old_out1], %[old_out3]                           \n\t"
00130                 "nmsub.s %[out0],     %[out0],      $f4, %[old_out2]        \n\t"
00131                 "nmsub.s %[out2],     %[out2],      $f4, %[old_out0]        \n\t"
00132 
00133                 : [out0]"+f"(out0), [out1]"+f"(out1),
00134                   [out2]"+f"(out2), [out3]"+f"(out3), [old_out0]"+f"(old_out0),
00135                   [old_out1]"+f"(old_out1), [old_out2]"+f"(old_out2),
00136                   [old_out3]"+f"(old_out3),[p_filter_coeffs]"+r"(p_filter_coeffs),
00137                   [p_out]"+r"(p_out)
00138                 :
00139                 : "$f4", "$f5"
00140             );
00141             FFSWAP(float, old_out0, old_out2);
00142         }
00143 
00144         __asm__ volatile(
00145             "nmsub.s    %[out3], %[out3], %[a], %[out2]                     \n\t"
00146             "nmsub.s    %[out2], %[out2], %[a], %[out1]                     \n\t"
00147             "nmsub.s    %[out3], %[out3], %[b], %[out1]                     \n\t"
00148             "nmsub.s    %[out1], %[out1], %[a], %[out0]                     \n\t"
00149             "nmsub.s    %[out2], %[out2], %[b], %[out0]                     \n\t"
00150             "nmsub.s    %[out3], %[out3], %[c], %[out0]                     \n\t"
00151 
00152             : [out0]"+f"(out0), [out1]"+f"(out1),
00153               [out2]"+f"(out2), [out3]"+f"(out3)
00154             : [a]"f"(a), [b]"f"(b), [c]"f"(c)
00155         );
00156 
00157         out[0] = out0;
00158         out[1] = out1;
00159         out[2] = out2;
00160         out[3] = out3;
00161 
00162         old_out0 = out0;
00163         old_out1 = out1;
00164         old_out2 = out2;
00165         old_out3 = out3;
00166 
00167         out += 4;
00168         in  += 4;
00169     }
00170 
00171     out -= n;
00172     in -= n;
00173     for (; n < buffer_length; n++) {
00174         float out_val, out_val_i, fc_val;
00175         p_filter_coeffs = filter_coeffs;
00176         p_out = &out[n];
00177         out_val = in[n];
00178         for (i = 1; i <= filter_length; i++) {
00179             __asm__ volatile(
00180                 "lwc1    %[fc_val],          0(%[p_filter_coeffs])                        \n\t"
00181                 "lwc1    %[out_val_i],       -4(%[p_out])                                 \n\t"
00182                 "addiu   %[p_filter_coeffs], 4                                            \n\t"
00183                 "addiu   %[p_out],           -4                                           \n\t"
00184                 "nmsub.s %[out_val],         %[out_val],          %[fc_val], %[out_val_i] \n\t"
00185 
00186                 : [fc_val]"=&f"(fc_val), [out_val]"+f"(out_val),
00187                   [out_val_i]"=&f"(out_val_i), [p_out]"+r"(p_out),
00188                   [p_filter_coeffs]"+r"(p_filter_coeffs)
00189             );
00190         }
00191         out[n] = out_val;
00192     }
00193 }
00194 
00195 static void ff_celp_lp_zero_synthesis_filterf_mips(float *out,
00196                                        const float *filter_coeffs,
00197                                        const float *in, int buffer_length,
00198                                        int filter_length)
00199 {
00200     int i,n;
00201     float sum_out8, sum_out7, sum_out6, sum_out5, sum_out4, fc_val;
00202     float sum_out3, sum_out2, sum_out1;
00203     const float *p_filter_coeffs, *p_in;
00204 
00205     for (n = 0; n < buffer_length; n+=8) {
00206         p_in = &in[n];
00207         p_filter_coeffs = filter_coeffs;
00208         sum_out8 = in[n+7];
00209         sum_out7 = in[n+6];
00210         sum_out6 = in[n+5];
00211         sum_out5 = in[n+4];
00212         sum_out4 = in[n+3];
00213         sum_out3 = in[n+2];
00214         sum_out2 = in[n+1];
00215         sum_out1 = in[n];
00216         i = filter_length;
00217 
00218         /* i is always greater than 0
00219         * outer loop is unrolled eight times so there is less memory access
00220         * inner loop is unrolled two times
00221         */
00222         __asm__ volatile(
00223             "filt_lp_inner%=:                                               \n\t"
00224             "lwc1   %[fc_val],   0(%[p_filter_coeffs])                      \n\t"
00225             "lwc1   $f7,         6*4(%[p_in])                               \n\t"
00226             "lwc1   $f6,         5*4(%[p_in])                               \n\t"
00227             "lwc1   $f5,         4*4(%[p_in])                               \n\t"
00228             "lwc1   $f4,         3*4(%[p_in])                               \n\t"
00229             "lwc1   $f3,         2*4(%[p_in])                               \n\t"
00230             "lwc1   $f2,         4(%[p_in])                                 \n\t"
00231             "lwc1   $f1,         0(%[p_in])                                 \n\t"
00232             "lwc1   $f0,         -4(%[p_in])                                \n\t"
00233             "addiu  %[i],        -2                                         \n\t"
00234             "madd.s %[sum_out8], %[sum_out8],          %[fc_val], $f7       \n\t"
00235             "madd.s %[sum_out7], %[sum_out7],          %[fc_val], $f6       \n\t"
00236             "madd.s %[sum_out6], %[sum_out6],          %[fc_val], $f5       \n\t"
00237             "madd.s %[sum_out5], %[sum_out5],          %[fc_val], $f4       \n\t"
00238             "madd.s %[sum_out4], %[sum_out4],          %[fc_val], $f3       \n\t"
00239             "madd.s %[sum_out3], %[sum_out3],          %[fc_val], $f2       \n\t"
00240             "madd.s %[sum_out2], %[sum_out2],          %[fc_val], $f1       \n\t"
00241             "madd.s %[sum_out1], %[sum_out1],          %[fc_val], $f0       \n\t"
00242             "lwc1   %[fc_val],   4(%[p_filter_coeffs])                      \n\t"
00243             "lwc1   $f7,         -8(%[p_in])                                \n\t"
00244             "addiu  %[p_filter_coeffs], 8                                   \n\t"
00245             "addiu  %[p_in],     -8                                         \n\t"
00246             "madd.s %[sum_out8], %[sum_out8],          %[fc_val], $f6       \n\t"
00247             "madd.s %[sum_out7], %[sum_out7],          %[fc_val], $f5       \n\t"
00248             "madd.s %[sum_out6], %[sum_out6],          %[fc_val], $f4       \n\t"
00249             "madd.s %[sum_out5], %[sum_out5],          %[fc_val], $f3       \n\t"
00250             "madd.s %[sum_out4], %[sum_out4],          %[fc_val], $f2       \n\t"
00251             "madd.s %[sum_out3], %[sum_out3],          %[fc_val], $f1       \n\t"
00252             "madd.s %[sum_out2], %[sum_out2],          %[fc_val], $f0       \n\t"
00253             "madd.s %[sum_out1], %[sum_out1],          %[fc_val], $f7       \n\t"
00254             "bgtz   %[i],        filt_lp_inner%=                            \n\t"
00255 
00256             : [sum_out8]"+f"(sum_out8), [sum_out7]"+f"(sum_out7),
00257               [sum_out6]"+f"(sum_out6), [sum_out5]"+f"(sum_out5),
00258               [sum_out4]"+f"(sum_out4), [sum_out3]"+f"(sum_out3),
00259               [sum_out2]"+f"(sum_out2), [sum_out1]"+f"(sum_out1),
00260               [fc_val]"=&f"(fc_val), [p_filter_coeffs]"+r"(p_filter_coeffs),
00261               [p_in]"+r"(p_in), [i]"+r"(i)
00262             :
00263             : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5", "$f6", "$f7"
00264         );
00265 
00266         out[n+7] = sum_out8;
00267         out[n+6] = sum_out7;
00268         out[n+5] = sum_out6;
00269         out[n+4] = sum_out5;
00270         out[n+3] = sum_out4;
00271         out[n+2] = sum_out3;
00272         out[n+1] = sum_out2;
00273         out[n] = sum_out1;
00274     }
00275 }
00276 
00277 void ff_celp_filter_init_mips(CELPFContext *c)
00278 {
00279     c->celp_lp_synthesis_filterf        = ff_celp_lp_synthesis_filterf_mips;
00280     c->celp_lp_zero_synthesis_filterf   = ff_celp_lp_zero_synthesis_filterf_mips;
00281 }
libavcodec/mips/celp_filters_mips.c
