FFmpeg
hevcdsp_init_loongarch.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2022 Loongson Technology Corporation Limited
3  * Contributed by Lu Wang <wanglu@loongson.cn>
4  * Hao Chen <chenhao@loongson.cn>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include "libavutil/loongarch/cpu.h"
24 #include "hevcdsp_lsx.h"
25 
26 void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
27 {
28  int cpu_flags = av_get_cpu_flags();
29 
30  if (have_lsx(cpu_flags)) {
31  if (bit_depth == 8) {
32  c->put_hevc_qpel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_lsx;
33  c->put_hevc_qpel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_lsx;
34  c->put_hevc_qpel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_lsx;
35  c->put_hevc_qpel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_lsx;
36  c->put_hevc_qpel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_lsx;
37  c->put_hevc_qpel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_lsx;
38  c->put_hevc_qpel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_lsx;
39  c->put_hevc_qpel[8][0][0] = ff_hevc_put_hevc_pel_pixels48_8_lsx;
40  c->put_hevc_qpel[9][0][0] = ff_hevc_put_hevc_pel_pixels64_8_lsx;
41 
42  c->put_hevc_epel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_lsx;
43  c->put_hevc_epel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_lsx;
44  c->put_hevc_epel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_lsx;
45  c->put_hevc_epel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_lsx;
46  c->put_hevc_epel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_lsx;
47  c->put_hevc_epel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_lsx;
48  c->put_hevc_epel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_lsx;
49 
50  c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_lsx;
51  c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_lsx;
52  c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_lsx;
53  c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_lsx;
54  c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_lsx;
55  c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_lsx;
56  c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_lsx;
57  c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_lsx;
58 
59  c->put_hevc_qpel[1][1][0] = ff_hevc_put_hevc_qpel_v4_8_lsx;
60  c->put_hevc_qpel[3][1][0] = ff_hevc_put_hevc_qpel_v8_8_lsx;
61  c->put_hevc_qpel[4][1][0] = ff_hevc_put_hevc_qpel_v12_8_lsx;
62  c->put_hevc_qpel[5][1][0] = ff_hevc_put_hevc_qpel_v16_8_lsx;
63  c->put_hevc_qpel[6][1][0] = ff_hevc_put_hevc_qpel_v24_8_lsx;
64  c->put_hevc_qpel[7][1][0] = ff_hevc_put_hevc_qpel_v32_8_lsx;
65  c->put_hevc_qpel[8][1][0] = ff_hevc_put_hevc_qpel_v48_8_lsx;
66  c->put_hevc_qpel[9][1][0] = ff_hevc_put_hevc_qpel_v64_8_lsx;
67 
68  c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_lsx;
69  c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_lsx;
70  c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_lsx;
71  c->put_hevc_qpel[5][1][1] = ff_hevc_put_hevc_qpel_hv16_8_lsx;
72  c->put_hevc_qpel[6][1][1] = ff_hevc_put_hevc_qpel_hv24_8_lsx;
73  c->put_hevc_qpel[7][1][1] = ff_hevc_put_hevc_qpel_hv32_8_lsx;
74  c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_lsx;
75  c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_lsx;
76 
77  c->put_hevc_epel[7][0][1] = ff_hevc_put_hevc_epel_h32_8_lsx;
78 
79  c->put_hevc_epel[5][1][0] = ff_hevc_put_hevc_epel_v16_8_lsx;
80  c->put_hevc_epel[6][1][0] = ff_hevc_put_hevc_epel_v24_8_lsx;
81  c->put_hevc_epel[7][1][0] = ff_hevc_put_hevc_epel_v32_8_lsx;
82 
83  c->put_hevc_epel[3][1][1] = ff_hevc_put_hevc_epel_hv8_8_lsx;
84  c->put_hevc_epel[4][1][1] = ff_hevc_put_hevc_epel_hv12_8_lsx;
85  c->put_hevc_epel[5][1][1] = ff_hevc_put_hevc_epel_hv16_8_lsx;
86  c->put_hevc_epel[6][1][1] = ff_hevc_put_hevc_epel_hv24_8_lsx;
87  c->put_hevc_epel[7][1][1] = ff_hevc_put_hevc_epel_hv32_8_lsx;
88 
89  c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_lsx;
90  c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_lsx;
91  c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_lsx;
92  c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_lsx;
93  c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_lsx;
94  c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_lsx;
95  c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_lsx;
96  c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_lsx;
97 
98  c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_lsx;
99  c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_lsx;
100  c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_lsx;
101  c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_lsx;
102  c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_lsx;
103  c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_lsx;
104  c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_lsx;
105 
106  c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_lsx;
107  c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_lsx;
108  c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_lsx;
109  c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_lsx;
110  c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_lsx;
111  c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_lsx;
112 
113  c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_lsx;
114  c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_lsx;
115  c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_lsx;
116  c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_lsx;
117  c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_lsx;
118  c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_lsx;
119 
120  c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_lsx;
121  c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_lsx;
122  c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_lsx;
123  c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_lsx;
124  c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_lsx;
125 
126  c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_lsx;
127  c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_lsx;
128 
129  c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_lsx;
130  c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_lsx;
131  c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_lsx;
132  c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_lsx;
133 
134  c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_lsx;
135  c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_lsx;
136  c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_lsx;
137  c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_lsx;
138  c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_lsx;
139 
140  c->put_hevc_qpel_uni[9][0][1] = ff_hevc_put_hevc_uni_qpel_h64_8_lsx;
141 
142  c->put_hevc_qpel_uni[6][1][0] = ff_hevc_put_hevc_uni_qpel_v24_8_lsx;
143  c->put_hevc_qpel_uni[7][1][0] = ff_hevc_put_hevc_uni_qpel_v32_8_lsx;
144  c->put_hevc_qpel_uni[8][1][0] = ff_hevc_put_hevc_uni_qpel_v48_8_lsx;
145  c->put_hevc_qpel_uni[9][1][0] = ff_hevc_put_hevc_uni_qpel_v64_8_lsx;
146 
147  c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_uni_qpel_hv8_8_lsx;
148  c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_uni_qpel_hv16_8_lsx;
149  c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_uni_qpel_hv24_8_lsx;
150  c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_lsx;
151  c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_lsx;
152  c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_lsx;
153 
154  c->put_hevc_epel_uni[6][1][0] = ff_hevc_put_hevc_uni_epel_v24_8_lsx;
155  c->put_hevc_epel_uni[7][1][0] = ff_hevc_put_hevc_uni_epel_v32_8_lsx;
156 
157  c->put_hevc_epel_uni[3][1][1] = ff_hevc_put_hevc_uni_epel_hv8_8_lsx;
158  c->put_hevc_epel_uni[4][1][1] = ff_hevc_put_hevc_uni_epel_hv12_8_lsx;
159  c->put_hevc_epel_uni[5][1][1] = ff_hevc_put_hevc_uni_epel_hv16_8_lsx;
160  c->put_hevc_epel_uni[6][1][1] = ff_hevc_put_hevc_uni_epel_hv24_8_lsx;
161  c->put_hevc_epel_uni[7][1][1] = ff_hevc_put_hevc_uni_epel_hv32_8_lsx;
162 
163  c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_lsx;
164  c->put_hevc_qpel_uni_w[5][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv16_8_lsx;
165  c->put_hevc_qpel_uni_w[6][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv24_8_lsx;
166  c->put_hevc_qpel_uni_w[7][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv32_8_lsx;
167  c->put_hevc_qpel_uni_w[8][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv48_8_lsx;
168  c->put_hevc_qpel_uni_w[9][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv64_8_lsx;
169 
170  c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_8_lsx;
171 
172  c->hevc_h_loop_filter_luma = ff_hevc_loop_filter_luma_h_8_lsx;
173  c->hevc_v_loop_filter_luma = ff_hevc_loop_filter_luma_v_8_lsx;
174 
175  c->hevc_h_loop_filter_luma_c = ff_hevc_loop_filter_luma_h_8_lsx;
176  c->hevc_v_loop_filter_luma_c = ff_hevc_loop_filter_luma_v_8_lsx;
177 
178  c->hevc_h_loop_filter_chroma = ff_hevc_loop_filter_chroma_h_8_lsx;
179  c->hevc_v_loop_filter_chroma = ff_hevc_loop_filter_chroma_v_8_lsx;
180 
181  c->hevc_h_loop_filter_chroma_c = ff_hevc_loop_filter_chroma_h_8_lsx;
182  c->hevc_v_loop_filter_chroma_c = ff_hevc_loop_filter_chroma_v_8_lsx;
183 
184  c->idct[0] = ff_hevc_idct_4x4_lsx;
185  c->idct[1] = ff_hevc_idct_8x8_lsx;
186  c->idct[2] = ff_hevc_idct_16x16_lsx;
187  c->idct[3] = ff_hevc_idct_32x32_lsx;
188  }
189  }
190 }
ff_hevc_loop_filter_chroma_h_8_lsx
void ff_hevc_loop_filter_chroma_h_8_lsx(uint8_t *src, ptrdiff_t stride, const int32_t *tc, const uint8_t *p_is_pcm, const uint8_t *q_is_pcm)
Definition: hevc_lpf_sao_lsx.c:863
ff_hevc_idct_32x32_lsx
void ff_hevc_idct_32x32_lsx(int16_t *coeffs, int col_limit)
Definition: hevc_idct_lsx.c:811
ff_hevc_idct_16x16_lsx
void ff_hevc_idct_16x16_lsx(int16_t *coeffs, int col_limit)
Definition: hevc_idct_lsx.c:386
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
ff_hevc_dsp_init_loongarch
void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth)
Definition: hevcdsp_init_loongarch.c:26
bit_depth
static void bit_depth(AudioStatsContext *s, const uint64_t *const mask, uint8_t *depth)
Definition: af_astats.c:245
cpu_flags
static atomic_int cpu_flags
Definition: cpu.c:52
ff_hevc_loop_filter_luma_v_8_lsx
void ff_hevc_loop_filter_luma_v_8_lsx(uint8_t *src, ptrdiff_t stride, int32_t beta, const int32_t *tc, const uint8_t *p_is_pcm, const uint8_t *q_is_pcm)
Definition: hevc_lpf_sao_lsx.c:438
ff_hevc_sao_edge_filter_8_lsx
void ff_hevc_sao_edge_filter_8_lsx(uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst, const int16_t *sao_offset_val, int eo, int width, int height)
Definition: hevc_lpf_sao_lsx.c:2367
ff_hevc_idct_4x4_lsx
void ff_hevc_idct_4x4_lsx(int16_t *coeffs, int col_limit)
Definition: hevc_idct_lsx.c:330
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
have_lsx
#define have_lsx(flags)
Definition: cpu.h:28
ff_hevc_loop_filter_luma_h_8_lsx
void ff_hevc_loop_filter_luma_h_8_lsx(uint8_t *src, ptrdiff_t stride, int32_t beta, const int32_t *tc, const uint8_t *p_is_pcm, const uint8_t *q_is_pcm)
Definition: hevc_lpf_sao_lsx.c:26
HEVCDSPContext
Definition: hevcdsp.h:47
ff_hevc_idct_8x8_lsx
void ff_hevc_idct_8x8_lsx(int16_t *coeffs, int col_limit)
Definition: hevc_idct_lsx.c:360
ff_hevc_loop_filter_chroma_v_8_lsx
void ff_hevc_loop_filter_chroma_v_8_lsx(uint8_t *src, ptrdiff_t stride, const int32_t *tc, const uint8_t *p_is_pcm, const uint8_t *q_is_pcm)
Definition: hevc_lpf_sao_lsx.c:917
hevcdsp_lsx.h
cpu.h
Generated on Thu Apr 18 2024 22:41:55 for FFmpeg by   doxygen 1.8.17