arm: vp9lpf: Implement the mix2_44 function with one single filter pass
[libav.git] / libavcodec / arm / vp9dsp_init_arm.c
1 /*
2 * Copyright (c) 2016 Google Inc.
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 "libavutil/attributes.h"
24 #include "libavutil/arm/cpu.h"
25 #include "libavcodec/vp9.h"
26
27 #define declare_fpel(type, sz) \
28 void ff_vp9_##type##sz##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
29 const uint8_t *src, ptrdiff_t src_stride, \
30 int h, int mx, int my)
31
32 #define declare_copy_avg(sz) \
33 declare_fpel(copy, sz); \
34 declare_fpel(avg , sz)
35
36 #define decl_mc_func(op, filter, dir, sz) \
37 void ff_vp9_##op##_##filter##sz##_##dir##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
38 const uint8_t *src, ptrdiff_t src_stride, \
39 int h, int mx, int my)
40
41 #define define_8tap_2d_fn(op, filter, sz) \
42 static void op##_##filter##sz##_hv_neon(uint8_t *dst, ptrdiff_t dst_stride, \
43 const uint8_t *src, ptrdiff_t src_stride, \
44 int h, int mx, int my) \
45 { \
46 LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz]); \
47 /* We only need h + 7 lines, but the horizontal filter assumes an \
48 * even number of rows, so filter h + 8 lines here. */ \
49 ff_vp9_put_##filter##sz##_h_neon(temp, sz, \
50 src - 3 * src_stride, src_stride, \
51 h + 8, mx, 0); \
52 ff_vp9_##op##_##filter##sz##_v_neon(dst, dst_stride, \
53 temp + 3 * sz, sz, \
54 h, 0, my); \
55 }
56
57 #define decl_filter_funcs(op, dir, sz) \
58 decl_mc_func(op, regular, dir, sz); \
59 decl_mc_func(op, sharp, dir, sz); \
60 decl_mc_func(op, smooth, dir, sz)
61
62 #define decl_mc_funcs(sz) \
63 decl_filter_funcs(put, h, sz); \
64 decl_filter_funcs(avg, h, sz); \
65 decl_filter_funcs(put, v, sz); \
66 decl_filter_funcs(avg, v, sz); \
67 decl_filter_funcs(put, hv, sz); \
68 decl_filter_funcs(avg, hv, sz)
69
70 declare_copy_avg(64);
71 declare_copy_avg(32);
72 declare_copy_avg(16);
73 declare_copy_avg(8);
74 declare_copy_avg(4);
75
76 decl_mc_funcs(64);
77 decl_mc_funcs(32);
78 decl_mc_funcs(16);
79 decl_mc_funcs(8);
80 decl_mc_funcs(4);
81
82 #define define_8tap_2d_funcs(sz) \
83 define_8tap_2d_fn(put, regular, sz) \
84 define_8tap_2d_fn(put, sharp, sz) \
85 define_8tap_2d_fn(put, smooth, sz) \
86 define_8tap_2d_fn(avg, regular, sz) \
87 define_8tap_2d_fn(avg, sharp, sz) \
88 define_8tap_2d_fn(avg, smooth, sz)
89
90 define_8tap_2d_funcs(64)
91 define_8tap_2d_funcs(32)
92 define_8tap_2d_funcs(16)
93 define_8tap_2d_funcs(8)
94 define_8tap_2d_funcs(4)
95
96
97 static av_cold void vp9dsp_mc_init_arm(VP9DSPContext *dsp)
98 {
99 int cpu_flags = av_get_cpu_flags();
100
101 if (have_neon(cpu_flags)) {
102 #define init_fpel(idx1, idx2, sz, type) \
103 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
104 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
105 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
106 dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##_neon
107
108 #define init_copy_avg(idx, sz) \
109 init_fpel(idx, 0, sz, copy); \
110 init_fpel(idx, 1, sz, avg)
111
112 #define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx) \
113 dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_neon
114
115 #define init_mc_funcs(idx, dir, mx, my, sz, pfx) \
116 init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
117 init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
118 init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx); \
119 init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
120 init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
121 init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx)
122
123 #define init_mc_funcs_dirs(idx, sz) \
124 init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_); \
125 init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_); \
126 init_mc_funcs(idx, hv, 1, 1, sz,)
127
128 init_copy_avg(0, 64);
129 init_copy_avg(1, 32);
130 init_copy_avg(2, 16);
131 init_copy_avg(3, 8);
132 init_copy_avg(4, 4);
133
134 init_mc_funcs_dirs(0, 64);
135 init_mc_funcs_dirs(1, 32);
136 init_mc_funcs_dirs(2, 16);
137 init_mc_funcs_dirs(3, 8);
138 init_mc_funcs_dirs(4, 4);
139 }
140 }
141
142 #define define_itxfm(type_a, type_b, sz) \
143 void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_neon(uint8_t *_dst, \
144 ptrdiff_t stride, \
145 int16_t *_block, int eob)
146
147 #define define_itxfm_funcs(sz) \
148 define_itxfm(idct, idct, sz); \
149 define_itxfm(iadst, idct, sz); \
150 define_itxfm(idct, iadst, sz); \
151 define_itxfm(iadst, iadst, sz)
152
153 define_itxfm_funcs(4);
154 define_itxfm_funcs(8);
155 define_itxfm_funcs(16);
156 define_itxfm(idct, idct, 32);
157 define_itxfm(iwht, iwht, 4);
158
159
160 static av_cold void vp9dsp_itxfm_init_arm(VP9DSPContext *dsp)
161 {
162 int cpu_flags = av_get_cpu_flags();
163
164 if (have_neon(cpu_flags)) {
165 #define init_itxfm(tx, sz) \
166 dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_neon; \
167 dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_neon; \
168 dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_neon; \
169 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_neon
170
171 #define init_idct(tx, nm) \
172 dsp->itxfm_add[tx][DCT_DCT] = \
173 dsp->itxfm_add[tx][ADST_DCT] = \
174 dsp->itxfm_add[tx][DCT_ADST] = \
175 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_neon
176
177 init_itxfm(TX_4X4, 4x4);
178 init_itxfm(TX_8X8, 8x8);
179 init_itxfm(TX_16X16, 16x16);
180 init_idct(TX_32X32, idct_idct_32x32);
181 init_idct(4, iwht_iwht_4x4);
182 }
183 }
184
185 #define define_loop_filter(dir, wd, size) \
186 void ff_vp9_loop_filter_##dir##_##wd##_##size##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H)
187
188 #define define_loop_filters(wd, size) \
189 define_loop_filter(h, wd, size); \
190 define_loop_filter(v, wd, size)
191
192 define_loop_filters(4, 8);
193 define_loop_filters(8, 8);
194 define_loop_filters(16, 8);
195 define_loop_filters(16, 16);
196
197 define_loop_filters(44, 16);
198
199 #define lf_mix_fn(dir, wd1, wd2, stridea) \
200 static void loop_filter_##dir##_##wd1##wd2##_16_neon(uint8_t *dst, \
201 ptrdiff_t stride, \
202 int E, int I, int H) \
203 { \
204 ff_vp9_loop_filter_##dir##_##wd1##_8_neon(dst, stride, E & 0xff, I & 0xff, H & 0xff); \
205 ff_vp9_loop_filter_##dir##_##wd2##_8_neon(dst + 8 * stridea, stride, E >> 8, I >> 8, H >> 8); \
206 }
207
208 #define lf_mix_fns(wd1, wd2) \
209 lf_mix_fn(h, wd1, wd2, stride) \
210 lf_mix_fn(v, wd1, wd2, sizeof(uint8_t))
211
212 lf_mix_fns(4, 8)
213 lf_mix_fns(8, 4)
214 lf_mix_fns(8, 8)
215
216 static av_cold void vp9dsp_loopfilter_init_arm(VP9DSPContext *dsp)
217 {
218 int cpu_flags = av_get_cpu_flags();
219
220 if (have_neon(cpu_flags)) {
221 dsp->loop_filter_8[0][1] = ff_vp9_loop_filter_v_4_8_neon;
222 dsp->loop_filter_8[0][0] = ff_vp9_loop_filter_h_4_8_neon;
223 dsp->loop_filter_8[1][1] = ff_vp9_loop_filter_v_8_8_neon;
224 dsp->loop_filter_8[1][0] = ff_vp9_loop_filter_h_8_8_neon;
225 dsp->loop_filter_8[2][1] = ff_vp9_loop_filter_v_16_8_neon;
226 dsp->loop_filter_8[2][0] = ff_vp9_loop_filter_h_16_8_neon;
227
228 dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_neon;
229 dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_neon;
230
231 dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_neon;
232 dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_neon;
233 dsp->loop_filter_mix2[0][1][0] = loop_filter_h_48_16_neon;
234 dsp->loop_filter_mix2[0][1][1] = loop_filter_v_48_16_neon;
235 dsp->loop_filter_mix2[1][0][0] = loop_filter_h_84_16_neon;
236 dsp->loop_filter_mix2[1][0][1] = loop_filter_v_84_16_neon;
237 dsp->loop_filter_mix2[1][1][0] = loop_filter_h_88_16_neon;
238 dsp->loop_filter_mix2[1][1][1] = loop_filter_v_88_16_neon;
239 }
240 }
241
242 av_cold void ff_vp9dsp_init_arm(VP9DSPContext *dsp)
243 {
244 vp9dsp_mc_init_arm(dsp);
245 vp9dsp_loopfilter_init_arm(dsp);
246 vp9dsp_itxfm_init_arm(dsp);
247 }