1 /*
2 * Copyright (c) 2020 Reimar Döffinger
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include <stdint.h>
22
23 #include "libavutil/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/aarch64/cpu.h"
26 #include "libavcodec/hevcdsp.h"
27
28 void ff_hevc_add_residual_4x4_8_neon(uint8_t *_dst, int16_t *coeffs,
29 ptrdiff_t stride);
30 void ff_hevc_add_residual_4x4_10_neon(uint8_t *_dst, int16_t *coeffs,
31 ptrdiff_t stride);
32 void ff_hevc_add_residual_8x8_8_neon(uint8_t *_dst, int16_t *coeffs,
33 ptrdiff_t stride);
34 void ff_hevc_add_residual_8x8_10_neon(uint8_t *_dst, int16_t *coeffs,
35 ptrdiff_t stride);
36 void ff_hevc_add_residual_16x16_8_neon(uint8_t *_dst, int16_t *coeffs,
37 ptrdiff_t stride);
38 void ff_hevc_add_residual_16x16_10_neon(uint8_t *_dst, int16_t *coeffs,
39 ptrdiff_t stride);
40 void ff_hevc_add_residual_32x32_8_neon(uint8_t *_dst, int16_t *coeffs,
41 ptrdiff_t stride);
42 void ff_hevc_add_residual_32x32_10_neon(uint8_t *_dst, int16_t *coeffs,
43 ptrdiff_t stride);
44 void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
45 void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
46 void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
47 void ff_hevc_idct_16x16_10_neon(int16_t *coeffs, int col_limit);
48 void ff_hevc_idct_4x4_dc_8_neon(int16_t *coeffs);
49 void ff_hevc_idct_8x8_dc_8_neon(int16_t *coeffs);
50 void ff_hevc_idct_16x16_dc_8_neon(int16_t *coeffs);
51 void ff_hevc_idct_32x32_dc_8_neon(int16_t *coeffs);
52 void ff_hevc_idct_4x4_dc_10_neon(int16_t *coeffs);
53 void ff_hevc_idct_8x8_dc_10_neon(int16_t *coeffs);
54 void ff_hevc_idct_16x16_dc_10_neon(int16_t *coeffs);
55 void ff_hevc_idct_32x32_dc_10_neon(int16_t *coeffs);
56 void ff_hevc_sao_band_filter_8x8_8_neon(uint8_t *_dst, uint8_t *_src,
57 ptrdiff_t stride_dst, ptrdiff_t stride_src,
58 int16_t *sao_offset_val, int sao_left_class,
59 int width, int height);
60 void ff_hevc_sao_edge_filter_16x16_8_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst,
61 int16_t *sao_offset_val, int eo, int width, int height);
62 void ff_hevc_sao_edge_filter_8x8_8_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst,
63 int16_t *sao_offset_val, int eo, int width, int height);
64
ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)65 av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
66 {
67 if (!have_neon(av_get_cpu_flags())) return;
68
69 if (bit_depth == 8) {
70 c->add_residual[0] = ff_hevc_add_residual_4x4_8_neon;
71 c->add_residual[1] = ff_hevc_add_residual_8x8_8_neon;
72 c->add_residual[2] = ff_hevc_add_residual_16x16_8_neon;
73 c->add_residual[3] = ff_hevc_add_residual_32x32_8_neon;
74 c->idct[1] = ff_hevc_idct_8x8_8_neon;
75 c->idct[2] = ff_hevc_idct_16x16_8_neon;
76 c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_neon;
77 c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_neon;
78 c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_neon;
79 c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_neon;
80 c->sao_band_filter[0] =
81 c->sao_band_filter[1] =
82 c->sao_band_filter[2] =
83 c->sao_band_filter[3] =
84 c->sao_band_filter[4] = ff_hevc_sao_band_filter_8x8_8_neon;
85 c->sao_edge_filter[0] = ff_hevc_sao_edge_filter_8x8_8_neon;
86 c->sao_edge_filter[1] =
87 c->sao_edge_filter[2] =
88 c->sao_edge_filter[3] =
89 c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_16x16_8_neon;
90 }
91 if (bit_depth == 10) {
92 c->add_residual[0] = ff_hevc_add_residual_4x4_10_neon;
93 c->add_residual[1] = ff_hevc_add_residual_8x8_10_neon;
94 c->add_residual[2] = ff_hevc_add_residual_16x16_10_neon;
95 c->add_residual[3] = ff_hevc_add_residual_32x32_10_neon;
96 c->idct[1] = ff_hevc_idct_8x8_10_neon;
97 c->idct[2] = ff_hevc_idct_16x16_10_neon;
98 c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_neon;
99 c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_neon;
100 c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_neon;
101 c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_neon;
102 }
103 }
104