2 * Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
4 * This file is part of Libav.
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.
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.
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
23 #include "libavutil/attributes.h"
24 #include "libavutil/arm/cpu.h"
25 #include "libavcodec/h264dsp.h"
27 void ff_h264_v_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
28 int beta, int8_t *tc0);
29 void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
30 int beta, int8_t *tc0);
31 void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
32 int beta, int8_t *tc0);
33 void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
34 int beta, int8_t *tc0);
36 void ff_weight_h264_pixels_16_neon(uint8_t *dst, int stride, int height,
37 int log2_den, int weight, int offset);
38 void ff_weight_h264_pixels_8_neon(uint8_t *dst, int stride, int height,
39 int log2_den, int weight, int offset);
40 void ff_weight_h264_pixels_4_neon(uint8_t *dst, int stride, int height,
41 int log2_den, int weight, int offset);
43 void ff_biweight_h264_pixels_16_neon(uint8_t *dst, uint8_t *src, int stride,
44 int height, int log2_den, int weightd,
45 int weights, int offset);
46 void ff_biweight_h264_pixels_8_neon(uint8_t *dst, uint8_t *src, int stride,
47 int height, int log2_den, int weightd,
48 int weights, int offset);
49 void ff_biweight_h264_pixels_4_neon(uint8_t *dst, uint8_t *src, int stride,
50 int height, int log2_den, int weightd,
51 int weights, int offset);
53 void ff_h264_idct_add_neon(uint8_t *dst, int16_t *block, int stride);
54 void ff_h264_idct_dc_add_neon(uint8_t *dst, int16_t *block, int stride);
55 void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
56 int16_t *block, int stride,
57 const uint8_t nnzc[6*8]);
58 void ff_h264_idct_add16intra_neon(uint8_t *dst, const int *block_offset,
59 int16_t *block, int stride,
60 const uint8_t nnzc[6*8]);
61 void ff_h264_idct_add8_neon(uint8_t **dest, const int *block_offset,
62 int16_t *block, int stride,
63 const uint8_t nnzc[6*8]);
65 void ff_h264_idct8_add_neon(uint8_t *dst, int16_t *block, int stride);
66 void ff_h264_idct8_dc_add_neon(uint8_t *dst, int16_t *block, int stride);
67 void ff_h264_idct8_add4_neon(uint8_t *dst, const int *block_offset,
68 int16_t *block, int stride,
69 const uint8_t nnzc[6*8]);
71 static av_cold void h264dsp_init_neon(H264DSPContext *c, const int bit_depth,
72 const int chroma_format_idc)
75 c->h264_v_loop_filter_luma = ff_h264_v_loop_filter_luma_neon;
76 c->h264_h_loop_filter_luma = ff_h264_h_loop_filter_luma_neon;
77 c->h264_v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
78 c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
80 c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels_16_neon;
81 c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels_8_neon;
82 c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels_4_neon;
84 c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels_16_neon;
85 c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels_8_neon;
86 c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels_4_neon;
88 c->h264_idct_add = ff_h264_idct_add_neon;
89 c->h264_idct_dc_add = ff_h264_idct_dc_add_neon;
90 c->h264_idct_add16 = ff_h264_idct_add16_neon;
91 c->h264_idct_add16intra = ff_h264_idct_add16intra_neon;
92 if (chroma_format_idc == 1)
93 c->h264_idct_add8 = ff_h264_idct_add8_neon;
94 c->h264_idct8_add = ff_h264_idct8_add_neon;
95 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_neon;
96 c->h264_idct8_add4 = ff_h264_idct8_add4_neon;
100 av_cold void ff_h264dsp_init_arm(H264DSPContext *c, const int bit_depth,
101 const int chroma_format_idc)
103 int cpu_flags = av_get_cpu_flags();
105 if (have_neon(cpu_flags))
106 h264dsp_init_neon(c, bit_depth, chroma_format_idc);