OSDN Git Service

libavcodec/vp9: ipred_dl_32x32_16 avx2 implementation
[android-x86/external-ffmpeg.git] / libavcodec / mlpdsp.c
1 /*
2  * Copyright (c) 2007-2008 Ian Caulfield
3  *               2009 Ramiro Polla
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 #include "config.h"
23 #include "libavutil/attributes.h"
24 #include "mlpdsp.h"
25 #include "mlp.h"
26
27 static void mlp_filter_channel(int32_t *state, const int32_t *coeff,
28                                int firorder, int iirorder,
29                                unsigned int filter_shift, int32_t mask,
30                                int blocksize, int32_t *sample_buffer)
31 {
32     int32_t *firbuf = state;
33     int32_t *iirbuf = state + MAX_BLOCKSIZE + MAX_FIR_ORDER;
34     const int32_t *fircoeff = coeff;
35     const int32_t *iircoeff = coeff + MAX_FIR_ORDER;
36     int i;
37
38     for (i = 0; i < blocksize; i++) {
39         int32_t residual = *sample_buffer;
40         unsigned int order;
41         int64_t accum = 0;
42         int32_t result;
43
44         for (order = 0; order < firorder; order++)
45             accum += (int64_t) firbuf[order] * fircoeff[order];
46         for (order = 0; order < iirorder; order++)
47             accum += (int64_t) iirbuf[order] * iircoeff[order];
48
49         accum  = accum >> filter_shift;
50         result = (accum + residual) & mask;
51
52         *--firbuf = result;
53         *--iirbuf = result - accum;
54
55         *sample_buffer = result;
56         sample_buffer += MAX_CHANNELS;
57     }
58 }
59
60 void ff_mlp_rematrix_channel(int32_t *samples,
61                              const int32_t *coeffs,
62                              const uint8_t *bypassed_lsbs,
63                              const int8_t *noise_buffer,
64                              int index,
65                              unsigned int dest_ch,
66                              uint16_t blockpos,
67                              unsigned int maxchan,
68                              int matrix_noise_shift,
69                              int access_unit_size_pow2,
70                              int32_t mask)
71 {
72     unsigned int src_ch, i;
73     int index2 = 2 * index + 1;
74     for (i = 0; i < blockpos; i++) {
75         int64_t accum = 0;
76
77         for (src_ch = 0; src_ch <= maxchan; src_ch++)
78             accum += (int64_t) samples[src_ch] * coeffs[src_ch];
79
80         if (matrix_noise_shift) {
81             index &= access_unit_size_pow2 - 1;
82             accum += noise_buffer[index] << (matrix_noise_shift + 7);
83             index += index2;
84         }
85
86         samples[dest_ch] = ((accum >> 14) & mask) + *bypassed_lsbs;
87         bypassed_lsbs += MAX_CHANNELS;
88         samples += MAX_CHANNELS;
89     }
90 }
91
92 static int32_t (*mlp_select_pack_output(uint8_t *ch_assign,
93                                         int8_t *output_shift,
94                                         uint8_t max_matrix_channel,
95                                         int is32))(int32_t, uint16_t, int32_t (*)[], void *, uint8_t*, int8_t *, uint8_t, int)
96 {
97     return ff_mlp_pack_output;
98 }
99
100 int32_t ff_mlp_pack_output(int32_t lossless_check_data,
101                            uint16_t blockpos,
102                            int32_t (*sample_buffer)[MAX_CHANNELS],
103                            void *data,
104                            uint8_t *ch_assign,
105                            int8_t *output_shift,
106                            uint8_t max_matrix_channel,
107                            int is32)
108 {
109     unsigned int i, out_ch = 0;
110     int32_t *data_32 = data;
111     int16_t *data_16 = data;
112
113     for (i = 0; i < blockpos; i++) {
114         for (out_ch = 0; out_ch <= max_matrix_channel; out_ch++) {
115             int mat_ch = ch_assign[out_ch];
116             int32_t sample = sample_buffer[i][mat_ch] *
117                           (1U << output_shift[mat_ch]);
118             lossless_check_data ^= (sample & 0xffffff) << mat_ch;
119             if (is32)
120                 *data_32++ = sample << 8;
121             else
122                 *data_16++ = sample >> 8;
123         }
124     }
125     return lossless_check_data;
126 }
127
128 av_cold void ff_mlpdsp_init(MLPDSPContext *c)
129 {
130     c->mlp_filter_channel = mlp_filter_channel;
131     c->mlp_rematrix_channel = ff_mlp_rematrix_channel;
132     c->mlp_select_pack_output = mlp_select_pack_output;
133     c->mlp_pack_output = ff_mlp_pack_output;
134     if (ARCH_ARM)
135         ff_mlpdsp_init_arm(c);
136     if (ARCH_X86)
137         ff_mlpdsp_init_x86(c);
138 }