Merge commit '502512ea0f179165782777265f441d4061ff6012'
[ffmpeg.git] / 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
58 #if HAVE_INLINE_ASM
59 static void ff_acelp_interpolatef_mips(float *out, const float *in,
60                            const float *filter_coeffs, int precision,
61                            int frac_pos, int filter_length, int length)
62 {
63     int n, i;
64     int prec = precision * 4;
65     int fc_offset = precision - frac_pos;
66     float in_val_p, in_val_m, fc_val_p, fc_val_m;
67
68     for (n = 0; n < length; n++) {
69         /**
70         * four pointers are defined in order to minimize number of
71         * computations done in inner loop
72         */
73         const float *p_in_p = &in[n];
74         const float *p_in_m = &in[n-1];
75         const float *p_filter_coeffs_p = &filter_coeffs[frac_pos];
76         const float *p_filter_coeffs_m = filter_coeffs + fc_offset;
77         float v = 0;
78
79         for (i = 0; i < filter_length;i++) {
80             __asm__ volatile (
81                 "lwc1   %[in_val_p],           0(%[p_in_p])                    \n\t"
82                 "lwc1   %[fc_val_p],           0(%[p_filter_coeffs_p])         \n\t"
83                 "lwc1   %[in_val_m],           0(%[p_in_m])                    \n\t"
84                 "lwc1   %[fc_val_m],           0(%[p_filter_coeffs_m])         \n\t"
85                 "addiu  %[p_in_p],             %[p_in_p],              4       \n\t"
86                 "madd.s %[v],%[v],             %[in_val_p],%[fc_val_p]         \n\t"
87                 "addiu  %[p_in_m],             %[p_in_m],              -4      \n\t"
88                 "addu   %[p_filter_coeffs_p],  %[p_filter_coeffs_p],   %[prec] \n\t"
89                 "addu   %[p_filter_coeffs_m],  %[p_filter_coeffs_m],   %[prec] \n\t"
90                 "madd.s %[v],%[v],%[in_val_m], %[fc_val_m]                     \n\t"
91
92                 : [v] "=&f" (v),[p_in_p] "+r" (p_in_p), [p_in_m] "+r" (p_in_m),
93                   [p_filter_coeffs_p] "+r" (p_filter_coeffs_p),
94                   [in_val_p] "=&f" (in_val_p), [in_val_m] "=&f" (in_val_m),
95                   [fc_val_p] "=&f" (fc_val_p), [fc_val_m] "=&f" (fc_val_m),
96                   [p_filter_coeffs_m] "+r" (p_filter_coeffs_m)
97                 : [prec] "r" (prec)
98                 : "memory"
99             );
100         }
101         out[n] = v;
102     }
103 }
104
105 static void ff_acelp_apply_order_2_transfer_function_mips(float *out, const float *in,
106                                               const float zero_coeffs[2],
107                                               const float pole_coeffs[2],
108                                               float gain, float mem[2], int n)
109 {
110     /**
111     * loop is unrolled eight times
112     */
113
114     __asm__ volatile (
115         "lwc1   $f0,    0(%[mem])                                              \n\t"
116         "blez   %[n],   ff_acelp_apply_order_2_transfer_function_end%=         \n\t"
117         "lwc1   $f1,    4(%[mem])                                              \n\t"
118         "lwc1   $f2,    0(%[pole_coeffs])                                      \n\t"
119         "lwc1   $f3,    4(%[pole_coeffs])                                      \n\t"
120         "lwc1   $f4,    0(%[zero_coeffs])                                      \n\t"
121         "lwc1   $f5,    4(%[zero_coeffs])                                      \n\t"
122
123         "ff_acelp_apply_order_2_transfer_function_madd%=:                      \n\t"
124
125         "lwc1   $f6,    0(%[in])                                               \n\t"
126         "mul.s  $f9,    $f3,      $f1                                          \n\t"
127         "mul.s  $f7,    $f2,      $f0                                          \n\t"
128         "msub.s $f7,    $f7,      %[gain], $f6                                 \n\t"
129         "sub.s  $f7,    $f7,      $f9                                          \n\t"
130         "madd.s $f8,    $f7,      $f4,     $f0                                 \n\t"
131         "madd.s $f8,    $f8,      $f5,     $f1                                 \n\t"
132         "lwc1   $f11,   4(%[in])                                               \n\t"
133         "mul.s  $f12,   $f3,      $f0                                          \n\t"
134         "mul.s  $f13,   $f2,      $f7                                          \n\t"
135         "msub.s $f13,   $f13,     %[gain], $f11                                \n\t"
136         "sub.s  $f13,   $f13,     $f12                                         \n\t"
137         "madd.s $f14,   $f13,     $f4,     $f7                                 \n\t"
138         "madd.s $f14,   $f14,     $f5,     $f0                                 \n\t"
139         "swc1   $f8,    0(%[out])                                              \n\t"
140         "lwc1   $f6,    8(%[in])                                               \n\t"
141         "mul.s  $f9,    $f3,      $f7                                          \n\t"
142         "mul.s  $f15,   $f2,      $f13                                         \n\t"
143         "msub.s $f15,   $f15,     %[gain], $f6                                 \n\t"
144         "sub.s  $f15,   $f15,     $f9                                          \n\t"
145         "madd.s $f8,    $f15,     $f4,     $f13                                \n\t"
146         "madd.s $f8,    $f8,      $f5,     $f7                                 \n\t"
147         "swc1   $f14,   4(%[out])                                              \n\t"
148         "lwc1   $f11,   12(%[in])                                              \n\t"
149         "mul.s  $f12,   $f3,      $f13                                         \n\t"
150         "mul.s  $f16,   $f2,      $f15                                         \n\t"
151         "msub.s $f16,   $f16,     %[gain], $f11                                \n\t"
152         "sub.s  $f16,   $f16,     $f12                                         \n\t"
153         "madd.s $f14,   $f16,     $f4,     $f15                                \n\t"
154         "madd.s $f14,   $f14,     $f5,     $f13                                \n\t"
155         "swc1   $f8,    8(%[out])                                              \n\t"
156         "lwc1   $f6,    16(%[in])                                              \n\t"
157         "mul.s  $f9,    $f3,      $f15                                         \n\t"
158         "mul.s  $f7,    $f2,      $f16                                         \n\t"
159         "msub.s $f7,    $f7,      %[gain], $f6                                 \n\t"
160         "sub.s  $f7,    $f7,      $f9                                          \n\t"
161         "madd.s $f8,    $f7,      $f4,     $f16                                \n\t"
162         "madd.s $f8,    $f8,      $f5,     $f15                                \n\t"
163         "swc1   $f14,   12(%[out])                                             \n\t"
164         "lwc1   $f11,   20(%[in])                                              \n\t"
165         "mul.s  $f12,   $f3,      $f16                                         \n\t"
166         "mul.s  $f13,   $f2,      $f7                                          \n\t"
167         "msub.s $f13,   $f13,     %[gain], $f11                                \n\t"
168         "sub.s  $f13,   $f13,     $f12                                         \n\t"
169         "madd.s $f14,   $f13,     $f4,     $f7                                 \n\t"
170         "madd.s $f14,   $f14,     $f5,     $f16                                \n\t"
171         "swc1   $f8,    16(%[out])                                             \n\t"
172         "lwc1   $f6,    24(%[in])                                              \n\t"
173         "mul.s  $f9,    $f3,      $f7                                          \n\t"
174         "mul.s  $f15,   $f2,      $f13                                         \n\t"
175         "msub.s $f15,   $f15,     %[gain], $f6                                 \n\t"
176         "sub.s  $f1,    $f15,     $f9                                          \n\t"
177         "madd.s $f8,    $f1,      $f4,     $f13                                \n\t"
178         "madd.s $f8,    $f8,      $f5,     $f7                                 \n\t"
179         "swc1   $f14,   20(%[out])                                             \n\t"
180         "lwc1   $f11,   28(%[in])                                              \n\t"
181         "mul.s  $f12,   $f3,      $f13                                         \n\t"
182         "mul.s  $f16,   $f2,      $f1                                          \n\t"
183         "msub.s $f16,   $f16,     %[gain], $f11                                \n\t"
184         "sub.s  $f0,    $f16,     $f12                                         \n\t"
185         "madd.s $f14,   $f0,      $f4,     $f1                                 \n\t"
186         "madd.s $f14,   $f14,     $f5,     $f13                                \n\t"
187         "swc1   $f8,    24(%[out])                                             \n\t"
188         "addiu  %[out], 32                                                     \n\t"
189         "addiu  %[in],  32                                                     \n\t"
190         "addiu  %[n],   -8                                                     \n\t"
191         "swc1   $f14,   -4(%[out])                                             \n\t"
192         "bnez   %[n],   ff_acelp_apply_order_2_transfer_function_madd%=        \n\t"
193         "swc1   $f1,    4(%[mem])                                              \n\t"
194         "swc1   $f0,    0(%[mem])                                              \n\t"
195
196         "ff_acelp_apply_order_2_transfer_function_end%=:                       \n\t"
197
198          : [out] "+r" (out),
199            [in] "+r" (in), [gain] "+f" (gain),
200            [n] "+r" (n), [mem] "+r" (mem)
201          : [zero_coeffs] "r" (zero_coeffs),
202            [pole_coeffs] "r" (pole_coeffs)
203          : "$f0", "$f1", "$f2", "$f3", "$f4", "$f5",
204            "$f6", "$f7",  "$f8", "$f9", "$f10", "$f11",
205            "$f12", "$f13", "$f14", "$f15", "$f16", "memory"
206     );
207 }
208 #endif /* HAVE_INLINE_ASM */
209
210 void ff_acelp_filter_init_mips(ACELPFContext *c)
211 {
212 #if HAVE_INLINE_ASM
213     c->acelp_interpolatef                      = ff_acelp_interpolatef_mips;
214     c->acelp_apply_order_2_transfer_function   = ff_acelp_apply_order_2_transfer_function_mips;
215 #endif
216 }