git.sesse.net Git - ffmpeg/blob - libavcodec/mips/acelp_filters_mips.c

   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:  Nedeljko Babic (nbabic@mips.com)
  30  *
  31  * various filters for ACELP-based codecs optimized for MIPS
  32  *
  33  * This file is part of FFmpeg.
  34  *
  35  * FFmpeg is free software; you can redistribute it and/or
  36  * modify it under the terms of the GNU Lesser General Public
  37  * License as published by the Free Software Foundation; either
  38  * version 2.1 of the License, or (at your option) any later version.
  39  *
  40  * FFmpeg is distributed in the hope that it will be useful,
  41  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  42  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  43  * Lesser General Public License for more details.
  44  *
  45  * You should have received a copy of the GNU Lesser General Public
  46  * License along with FFmpeg; if not, write to the Free Software
  47  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  48  */
  49
  50 /**
  51  * @file
  52  * Reference: libavcodec/acelp_filters.c
  53  */
  54 #include "config.h"
  55 #include "libavutil/attributes.h"
  56 #include "libavcodec/acelp_filters.h"
  57 #include "libavutil/mips/asmdefs.h"
  58
  59 #if HAVE_INLINE_ASM
  60 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
  61 static void ff_acelp_interpolatef_mips(float *out, const float *in,
  62                            const float *filter_coeffs, int precision,
  63                            int frac_pos, int filter_length, int length)
  64 {
  65     int n, i;
  66     int prec = precision * 4;
  67     int fc_offset = precision - frac_pos;
  68     float in_val_p, in_val_m, fc_val_p, fc_val_m;
  69
  70     for (n = 0; n < length; n++) {
  71         /**
  72         * four pointers are defined in order to minimize number of
  73         * computations done in inner loop
  74         */
  75         const float *p_in_p = &in[n];
  76         const float *p_in_m = &in[n-1];
  77         const float *p_filter_coeffs_p = &filter_coeffs[frac_pos];
  78         const float *p_filter_coeffs_m = filter_coeffs + fc_offset;
  79         float v = 0;
  80
  81         for (i = 0; i < filter_length;i++) {
  82             __asm__ volatile (
  83                 "lwc1   %[in_val_p],           0(%[p_in_p])                    \n\t"
  84                 "lwc1   %[fc_val_p],           0(%[p_filter_coeffs_p])         \n\t"
  85                 "lwc1   %[in_val_m],           0(%[p_in_m])                    \n\t"
  86                 "lwc1   %[fc_val_m],           0(%[p_filter_coeffs_m])         \n\t"
  87                 PTR_ADDIU "%[p_in_p],          %[p_in_p],              4       \n\t"
  88                 "madd.s %[v],%[v],             %[in_val_p],%[fc_val_p]         \n\t"
  89                 PTR_ADDIU "%[p_in_m],          %[p_in_m],              -4      \n\t"
  90                 PTR_ADDU "%[p_filter_coeffs_p],%[p_filter_coeffs_p],   %[prec] \n\t"
  91                 PTR_ADDU "%[p_filter_coeffs_m],%[p_filter_coeffs_m],   %[prec] \n\t"
  92                 "madd.s %[v],%[v],%[in_val_m], %[fc_val_m]                     \n\t"
  93
  94                 : [v] "+&f" (v),[p_in_p] "+r" (p_in_p), [p_in_m] "+r" (p_in_m),
  95                   [p_filter_coeffs_p] "+r" (p_filter_coeffs_p),
  96                   [in_val_p] "=&f" (in_val_p), [in_val_m] "=&f" (in_val_m),
  97                   [fc_val_p] "=&f" (fc_val_p), [fc_val_m] "=&f" (fc_val_m),
  98                   [p_filter_coeffs_m] "+r" (p_filter_coeffs_m)
  99                 : [prec] "r" (prec)
 100                 : "memory"
 101             );
 102         }
 103         out[n] = v;
 104     }
 105 }
 106
 107 static void ff_acelp_apply_order_2_transfer_function_mips(float *out, const float *in,
 108                                               const float zero_coeffs[2],
 109                                               const float pole_coeffs[2],
 110                                               float gain, float mem[2], int n)
 111 {
 112     /**
 113     * loop is unrolled eight times
 114     */
 115
 116     __asm__ volatile (
 117         "lwc1   $f0,    0(%[mem])                                              \n\t"
 118         "blez   %[n],   ff_acelp_apply_order_2_transfer_function_end%=         \n\t"
 119         "lwc1   $f1,    4(%[mem])                                              \n\t"
 120         "lwc1   $f2,    0(%[pole_coeffs])                                      \n\t"
 121         "lwc1   $f3,    4(%[pole_coeffs])                                      \n\t"
 122         "lwc1   $f4,    0(%[zero_coeffs])                                      \n\t"
 123         "lwc1   $f5,    4(%[zero_coeffs])                                      \n\t"
 124
 125         "ff_acelp_apply_order_2_transfer_function_madd%=:                      \n\t"
 126
 127         "lwc1   $f6,    0(%[in])                                               \n\t"
 128         "mul.s  $f9,    $f3,      $f1                                          \n\t"
 129         "mul.s  $f7,    $f2,      $f0                                          \n\t"
 130         "msub.s $f7,    $f7,      %[gain], $f6                                 \n\t"
 131         "sub.s  $f7,    $f7,      $f9                                          \n\t"
 132         "madd.s $f8,    $f7,      $f4,     $f0                                 \n\t"
 133         "madd.s $f8,    $f8,      $f5,     $f1                                 \n\t"
 134         "lwc1   $f11,   4(%[in])                                               \n\t"
 135         "mul.s  $f12,   $f3,      $f0                                          \n\t"
 136         "mul.s  $f13,   $f2,      $f7                                          \n\t"
 137         "msub.s $f13,   $f13,     %[gain], $f11                                \n\t"
 138         "sub.s  $f13,   $f13,     $f12                                         \n\t"
 139         "madd.s $f14,   $f13,     $f4,     $f7                                 \n\t"
 140         "madd.s $f14,   $f14,     $f5,     $f0                                 \n\t"
 141         "swc1   $f8,    0(%[out])                                              \n\t"
 142         "lwc1   $f6,    8(%[in])                                               \n\t"
 143         "mul.s  $f9,    $f3,      $f7                                          \n\t"
 144         "mul.s  $f15,   $f2,      $f13                                         \n\t"
 145         "msub.s $f15,   $f15,     %[gain], $f6                                 \n\t"
 146         "sub.s  $f15,   $f15,     $f9                                          \n\t"
 147         "madd.s $f8,    $f15,     $f4,     $f13                                \n\t"
 148         "madd.s $f8,    $f8,      $f5,     $f7                                 \n\t"
 149         "swc1   $f14,   4(%[out])                                              \n\t"
 150         "lwc1   $f11,   12(%[in])                                              \n\t"
 151         "mul.s  $f12,   $f3,      $f13                                         \n\t"
 152         "mul.s  $f16,   $f2,      $f15                                         \n\t"
 153         "msub.s $f16,   $f16,     %[gain], $f11                                \n\t"
 154         "sub.s  $f16,   $f16,     $f12                                         \n\t"
 155         "madd.s $f14,   $f16,     $f4,     $f15                                \n\t"
 156         "madd.s $f14,   $f14,     $f5,     $f13                                \n\t"
 157         "swc1   $f8,    8(%[out])                                              \n\t"
 158         "lwc1   $f6,    16(%[in])                                              \n\t"
 159         "mul.s  $f9,    $f3,      $f15                                         \n\t"
 160         "mul.s  $f7,    $f2,      $f16                                         \n\t"
 161         "msub.s $f7,    $f7,      %[gain], $f6                                 \n\t"
 162         "sub.s  $f7,    $f7,      $f9                                          \n\t"
 163         "madd.s $f8,    $f7,      $f4,     $f16                                \n\t"
 164         "madd.s $f8,    $f8,      $f5,     $f15                                \n\t"
 165         "swc1   $f14,   12(%[out])                                             \n\t"
 166         "lwc1   $f11,   20(%[in])                                              \n\t"
 167         "mul.s  $f12,   $f3,      $f16                                         \n\t"
 168         "mul.s  $f13,   $f2,      $f7                                          \n\t"
 169         "msub.s $f13,   $f13,     %[gain], $f11                                \n\t"
 170         "sub.s  $f13,   $f13,     $f12                                         \n\t"
 171         "madd.s $f14,   $f13,     $f4,     $f7                                 \n\t"
 172         "madd.s $f14,   $f14,     $f5,     $f16                                \n\t"
 173         "swc1   $f8,    16(%[out])                                             \n\t"
 174         "lwc1   $f6,    24(%[in])                                              \n\t"
 175         "mul.s  $f9,    $f3,      $f7                                          \n\t"
 176         "mul.s  $f15,   $f2,      $f13                                         \n\t"
 177         "msub.s $f15,   $f15,     %[gain], $f6                                 \n\t"
 178         "sub.s  $f1,    $f15,     $f9                                          \n\t"
 179         "madd.s $f8,    $f1,      $f4,     $f13                                \n\t"
 180         "madd.s $f8,    $f8,      $f5,     $f7                                 \n\t"
 181         "swc1   $f14,   20(%[out])                                             \n\t"
 182         "lwc1   $f11,   28(%[in])                                              \n\t"
 183         "mul.s  $f12,   $f3,      $f13                                         \n\t"
 184         "mul.s  $f16,   $f2,      $f1                                          \n\t"
 185         "msub.s $f16,   $f16,     %[gain], $f11                                \n\t"
 186         "sub.s  $f0,    $f16,     $f12                                         \n\t"
 187         "madd.s $f14,   $f0,      $f4,     $f1                                 \n\t"
 188         "madd.s $f14,   $f14,     $f5,     $f13                                \n\t"
 189         "swc1   $f8,    24(%[out])                                             \n\t"
 190         PTR_ADDIU "%[out], 32                                                  \n\t"
 191         PTR_ADDIU "%[in],  32                                                  \n\t"
 192         "addiu  %[n],   -8                                                     \n\t"
 193         "swc1   $f14,   -4(%[out])                                             \n\t"
 194         "bnez   %[n],   ff_acelp_apply_order_2_transfer_function_madd%=        \n\t"
 195         "swc1   $f1,    4(%[mem])                                              \n\t"
 196         "swc1   $f0,    0(%[mem])                                              \n\t"
 197
 198         "ff_acelp_apply_order_2_transfer_function_end%=:                       \n\t"
 199
 200          : [out] "+r" (out),
 201            [in] "+r" (in), [gain] "+f" (gain),
 202            [n] "+r" (n), [mem] "+r" (mem)
 203          : [zero_coeffs] "r" (zero_coeffs),
 204            [pole_coeffs] "r" (pole_coeffs)
 205          : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5",
 206            "$f6", "$f7",  "$f8", "$f9", "$f10", "$f11",
 207            "$f12", "$f13", "$f14", "$f15", "$f16", "memory"
 208     );
 209 }
 210 #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
 211 #endif /* HAVE_INLINE_ASM */
 212
 213 void ff_acelp_filter_init_mips(ACELPFContext *c)
 214 {
 215 #if HAVE_INLINE_ASM
 216 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
 217     c->acelp_interpolatef                      = ff_acelp_interpolatef_mips;
 218     c->acelp_apply_order_2_transfer_function   = ff_acelp_apply_order_2_transfer_function_mips;
 219 #endif
 220 #endif
 221 }