1/*
2 * Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
3 *
4 * This file is part of Libav.
5 *
6 * Libav 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 * Libav 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 Libav; 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 "libavcodec/dsputil.h"
24#include "libavcodec/h264dsp.h"
25
26void ff_h264_v_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
27                                     int beta, int8_t *tc0);
28void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
29                                     int beta, int8_t *tc0);
30void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
31                                       int beta, int8_t *tc0);
32void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
33                                       int beta, int8_t *tc0);
34
35void ff_weight_h264_pixels_16_neon(uint8_t *dst, int stride, int height,
36                                   int log2_den, int weight, int offset);
37void ff_weight_h264_pixels_8_neon(uint8_t *dst, int stride, int height,
38                                  int log2_den, int weight, int offset);
39void ff_weight_h264_pixels_4_neon(uint8_t *dst, int stride, int height,
40                                  int log2_den, int weight, int offset);
41
42void ff_biweight_h264_pixels_16_neon(uint8_t *dst, uint8_t *src, int stride,
43                                     int height, int log2_den, int weightd,
44                                     int weights, int offset);
45void ff_biweight_h264_pixels_8_neon(uint8_t *dst, uint8_t *src, int stride,
46                                    int height, int log2_den, int weightd,
47                                    int weights, int offset);
48void ff_biweight_h264_pixels_4_neon(uint8_t *dst, uint8_t *src, int stride,
49                                    int height, int log2_den, int weightd,
50                                    int weights, int offset);
51
52void ff_h264_idct_add_neon(uint8_t *dst, DCTELEM *block, int stride);
53void ff_h264_idct_dc_add_neon(uint8_t *dst, DCTELEM *block, int stride);
54void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
55                             DCTELEM *block, int stride,
56                             const uint8_t nnzc[6*8]);
57void ff_h264_idct_add16intra_neon(uint8_t *dst, const int *block_offset,
58                                  DCTELEM *block, int stride,
59                                  const uint8_t nnzc[6*8]);
60void ff_h264_idct_add8_neon(uint8_t **dest, const int *block_offset,
61                            DCTELEM *block, int stride,
62                            const uint8_t nnzc[6*8]);
63
64void ff_h264_idct8_add_neon(uint8_t *dst, DCTELEM *block, int stride);
65void ff_h264_idct8_dc_add_neon(uint8_t *dst, DCTELEM *block, int stride);
66void ff_h264_idct8_add4_neon(uint8_t *dst, const int *block_offset,
67                             DCTELEM *block, int stride,
68                             const uint8_t nnzc[6*8]);
69
70static void ff_h264dsp_init_neon(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
71{
72    if (bit_depth == 8) {
73    c->h264_v_loop_filter_luma   = ff_h264_v_loop_filter_luma_neon;
74    c->h264_h_loop_filter_luma   = ff_h264_h_loop_filter_luma_neon;
75    c->h264_v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
76    c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
77
78    c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels_16_neon;
79    c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels_8_neon;
80    c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels_4_neon;
81
82    c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels_16_neon;
83    c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels_8_neon;
84    c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels_4_neon;
85
86    c->h264_idct_add        = ff_h264_idct_add_neon;
87    c->h264_idct_dc_add     = ff_h264_idct_dc_add_neon;
88    c->h264_idct_add16      = ff_h264_idct_add16_neon;
89    c->h264_idct_add16intra = ff_h264_idct_add16intra_neon;
90    if (chroma_format_idc == 1)
91        c->h264_idct_add8   = ff_h264_idct_add8_neon;
92    c->h264_idct8_add       = ff_h264_idct8_add_neon;
93    c->h264_idct8_dc_add    = ff_h264_idct8_dc_add_neon;
94    c->h264_idct8_add4      = ff_h264_idct8_add4_neon;
95    }
96}
97
98void ff_h264dsp_init_arm(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
99{
100    if (HAVE_NEON) ff_h264dsp_init_neon(c, bit_depth, chroma_format_idc);
101}
102