1 /*
2 * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
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/aarch64/cpu.h"
25 #include "libavcodec/avcodec.h"
26 #include "libavcodec/h264pred.h"
27
28 void ff_pred16x16_vert_neon(uint8_t *src, ptrdiff_t stride);
29 void ff_pred16x16_hor_neon(uint8_t *src, ptrdiff_t stride);
30 void ff_pred16x16_plane_neon(uint8_t *src, ptrdiff_t stride);
31 void ff_pred16x16_dc_neon(uint8_t *src, ptrdiff_t stride);
32 void ff_pred16x16_128_dc_neon(uint8_t *src, ptrdiff_t stride);
33 void ff_pred16x16_left_dc_neon(uint8_t *src, ptrdiff_t stride);
34 void ff_pred16x16_top_dc_neon(uint8_t *src, ptrdiff_t stride);
35
36 void ff_pred8x8_vert_neon(uint8_t *src, ptrdiff_t stride);
37 void ff_pred8x8_hor_neon(uint8_t *src, ptrdiff_t stride);
38 void ff_pred8x8_plane_neon(uint8_t *src, ptrdiff_t stride);
39 void ff_pred8x8_dc_neon(uint8_t *src, ptrdiff_t stride);
40 void ff_pred8x8_128_dc_neon(uint8_t *src, ptrdiff_t stride);
41 void ff_pred8x8_left_dc_neon(uint8_t *src, ptrdiff_t stride);
42 void ff_pred8x8_top_dc_neon(uint8_t *src, ptrdiff_t stride);
43 void ff_pred8x8_l0t_dc_neon(uint8_t *src, ptrdiff_t stride);
44 void ff_pred8x8_0lt_dc_neon(uint8_t *src, ptrdiff_t stride);
45 void ff_pred8x8_l00_dc_neon(uint8_t *src, ptrdiff_t stride);
46 void ff_pred8x8_0l0_dc_neon(uint8_t *src, ptrdiff_t stride);
47
48 void ff_pred16x16_vert_neon_10(uint8_t *src, ptrdiff_t stride);
49 void ff_pred16x16_hor_neon_10(uint8_t *src, ptrdiff_t stride);
50 void ff_pred16x16_plane_neon_10(uint8_t *src, ptrdiff_t stride);
51 void ff_pred16x16_dc_neon_10(uint8_t *src, ptrdiff_t stride);
52 void ff_pred16x16_top_dc_neon_10(uint8_t *src, ptrdiff_t stride);
53
54 void ff_pred8x8_vert_neon_10(uint8_t *src, ptrdiff_t stride);
55 void ff_pred8x8_hor_neon_10(uint8_t *src, ptrdiff_t stride);
56 void ff_pred8x8_plane_neon_10(uint8_t *src, ptrdiff_t stride);
57 void ff_pred8x8_dc_neon_10(uint8_t *src, ptrdiff_t stride);
58 void ff_pred8x8_128_dc_neon_10(uint8_t *src, ptrdiff_t stride);
59 void ff_pred8x8_left_dc_neon_10(uint8_t *src, ptrdiff_t stride);
60 void ff_pred8x8_top_dc_neon_10(uint8_t *src, ptrdiff_t stride);
61 void ff_pred8x8_l0t_dc_neon_10(uint8_t *src, ptrdiff_t stride);
62 void ff_pred8x8_0lt_dc_neon_10(uint8_t *src, ptrdiff_t stride);
63 void ff_pred8x8_l00_dc_neon_10(uint8_t *src, ptrdiff_t stride);
64 void ff_pred8x8_0l0_dc_neon_10(uint8_t *src, ptrdiff_t stride);
65
h264_pred_init_neon(H264PredContext * h,int codec_id,const int bit_depth,const int chroma_format_idc)66 static av_cold void h264_pred_init_neon(H264PredContext *h, int codec_id,
67 const int bit_depth,
68 const int chroma_format_idc)
69 {
70 if (bit_depth == 8) {
71 if (chroma_format_idc <= 1) {
72 h->pred8x8[VERT_PRED8x8 ] = ff_pred8x8_vert_neon;
73 h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_hor_neon;
74 if (codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8)
75 h->pred8x8[PLANE_PRED8x8] = ff_pred8x8_plane_neon;
76 h->pred8x8[DC_128_PRED8x8 ] = ff_pred8x8_128_dc_neon;
77 if (codec_id != AV_CODEC_ID_RV40 && codec_id != AV_CODEC_ID_VP7 &&
78 codec_id != AV_CODEC_ID_VP8) {
79 h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_neon;
80 h->pred8x8[LEFT_DC_PRED8x8] = ff_pred8x8_left_dc_neon;
81 h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_neon;
82 h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8] = ff_pred8x8_l0t_dc_neon;
83 h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8] = ff_pred8x8_0lt_dc_neon;
84 h->pred8x8[ALZHEIMER_DC_L00_PRED8x8] = ff_pred8x8_l00_dc_neon;
85 h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8] = ff_pred8x8_0l0_dc_neon;
86 }
87 }
88
89 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_neon;
90 h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vert_neon;
91 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_hor_neon;
92 h->pred16x16[LEFT_DC_PRED8x8] = ff_pred16x16_left_dc_neon;
93 h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_neon;
94 h->pred16x16[DC_128_PRED8x8 ] = ff_pred16x16_128_dc_neon;
95 if (codec_id != AV_CODEC_ID_SVQ3 && codec_id != AV_CODEC_ID_RV40 &&
96 codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8)
97 h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_neon;
98 }
99 if (bit_depth == 10) {
100 if (chroma_format_idc <= 1) {
101 h->pred8x8[VERT_PRED8x8 ] = ff_pred8x8_vert_neon_10;
102 h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_hor_neon_10;
103 if (codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8)
104 h->pred8x8[PLANE_PRED8x8] = ff_pred8x8_plane_neon_10;
105 h->pred8x8[DC_128_PRED8x8 ] = ff_pred8x8_128_dc_neon_10;
106 if (codec_id != AV_CODEC_ID_RV40 && codec_id != AV_CODEC_ID_VP7 &&
107 codec_id != AV_CODEC_ID_VP8) {
108 h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_neon_10;
109 h->pred8x8[LEFT_DC_PRED8x8] = ff_pred8x8_left_dc_neon_10;
110 h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_neon_10;
111 h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8] = ff_pred8x8_l0t_dc_neon_10;
112 h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8] = ff_pred8x8_0lt_dc_neon_10;
113 h->pred8x8[ALZHEIMER_DC_L00_PRED8x8] = ff_pred8x8_l00_dc_neon_10;
114 h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8] = ff_pred8x8_0l0_dc_neon_10;
115 }
116 }
117
118 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_neon_10;
119 h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vert_neon_10;
120 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_hor_neon_10;
121 h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_neon_10;
122 if (codec_id != AV_CODEC_ID_SVQ3 && codec_id != AV_CODEC_ID_RV40 &&
123 codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8)
124 h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_neon_10;
125 }
126 }
127
ff_h264_pred_init_aarch64(H264PredContext * h,int codec_id,int bit_depth,const int chroma_format_idc)128 av_cold void ff_h264_pred_init_aarch64(H264PredContext *h, int codec_id,
129 int bit_depth, const int chroma_format_idc)
130 {
131 int cpu_flags = av_get_cpu_flags();
132
133 if (have_neon(cpu_flags))
134 h264_pred_init_neon(h, codec_id, bit_depth, chroma_format_idc);
135 }
136