1 /*
2 * Copyright (c) 2016 Google Inc.
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/internal.h"
25 #include "libavutil/aarch64/cpu.h"
26 #include "libavcodec/vp9dsp.h"
27 #include "vp9dsp_init.h"
28
29 #define declare_fpel(type, sz) \
30 void ff_vp9_##type##sz##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
31 const uint8_t *src, ptrdiff_t src_stride, \
32 int h, int mx, int my)
33
34 #define declare_copy_avg(sz) \
35 declare_fpel(copy, sz); \
36 declare_fpel(avg , sz)
37
38 #define decl_mc_func(op, filter, dir, sz) \
39 void ff_vp9_##op##_##filter##sz##_##dir##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
40 const uint8_t *src, ptrdiff_t src_stride, \
41 int h, int mx, int my)
42
43 #define define_8tap_2d_fn(op, filter, sz) \
44 static void op##_##filter##sz##_hv_neon(uint8_t *dst, ptrdiff_t dst_stride, \
45 const uint8_t *src, ptrdiff_t src_stride, \
46 int h, int mx, int my) \
47 { \
48 LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz]); \
49 /* We only need h + 7 lines, but the horizontal filter assumes an \
50 * even number of rows, so filter h + 8 lines here. */ \
51 ff_vp9_put_##filter##sz##_h_neon(temp, sz, \
52 src - 3 * src_stride, src_stride, \
53 h + 8, mx, 0); \
54 ff_vp9_##op##_##filter##sz##_v_neon(dst, dst_stride, \
55 temp + 3 * sz, sz, \
56 h, 0, my); \
57 }
58
59 #define decl_filter_funcs(op, dir, sz) \
60 decl_mc_func(op, regular, dir, sz); \
61 decl_mc_func(op, sharp, dir, sz); \
62 decl_mc_func(op, smooth, dir, sz)
63
64 #define decl_mc_funcs(sz) \
65 decl_filter_funcs(put, h, sz); \
66 decl_filter_funcs(avg, h, sz); \
67 decl_filter_funcs(put, v, sz); \
68 decl_filter_funcs(avg, v, sz); \
69 decl_filter_funcs(put, hv, sz); \
70 decl_filter_funcs(avg, hv, sz)
71
72 #define ff_vp9_copy32_neon ff_vp9_copy32_aarch64
73 #define ff_vp9_copy64_neon ff_vp9_copy64_aarch64
74
75 declare_copy_avg(64);
76 declare_copy_avg(32);
77 declare_copy_avg(16);
78 declare_copy_avg(8);
79 declare_copy_avg(4);
80
81 decl_mc_funcs(64);
82 decl_mc_funcs(32);
83 decl_mc_funcs(16);
84 decl_mc_funcs(8);
85 decl_mc_funcs(4);
86
87 #define define_8tap_2d_funcs(sz) \
88 define_8tap_2d_fn(put, regular, sz) \
89 define_8tap_2d_fn(put, sharp, sz) \
90 define_8tap_2d_fn(put, smooth, sz) \
91 define_8tap_2d_fn(avg, regular, sz) \
92 define_8tap_2d_fn(avg, sharp, sz) \
93 define_8tap_2d_fn(avg, smooth, sz)
94
95 define_8tap_2d_funcs(64)
96 define_8tap_2d_funcs(32)
97 define_8tap_2d_funcs(16)
98 define_8tap_2d_funcs(8)
99 define_8tap_2d_funcs(4)
100
vp9dsp_mc_init_aarch64(VP9DSPContext * dsp)101 static av_cold void vp9dsp_mc_init_aarch64(VP9DSPContext *dsp)
102 {
103 int cpu_flags = av_get_cpu_flags();
104
105 #define init_fpel(idx1, idx2, sz, type, suffix) \
106 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
107 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
108 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
109 dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##suffix
110
111 #define init_copy(idx, sz, suffix) \
112 init_fpel(idx, 0, sz, copy, suffix)
113
114 #define init_avg(idx, sz, suffix) \
115 init_fpel(idx, 1, sz, avg, suffix)
116
117 #define init_copy_avg(idx, sz) \
118 init_copy(idx, sz, _neon); \
119 init_avg (idx, sz, _neon)
120
121 if (have_armv8(cpu_flags)) {
122 init_copy(0, 64, _aarch64);
123 init_copy(1, 32, _aarch64);
124 }
125
126 if (have_neon(cpu_flags)) {
127 #define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx) \
128 dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_neon
129
130 #define init_mc_funcs(idx, dir, mx, my, sz, pfx) \
131 init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
132 init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
133 init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx); \
134 init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \
135 init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \
136 init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx)
137
138 #define init_mc_funcs_dirs(idx, sz) \
139 init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_); \
140 init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_); \
141 init_mc_funcs(idx, hv, 1, 1, sz,)
142
143 init_avg(0, 64, _neon);
144 init_avg(1, 32, _neon);
145 init_copy_avg(2, 16);
146 init_copy_avg(3, 8);
147 init_copy_avg(4, 4);
148
149 init_mc_funcs_dirs(0, 64);
150 init_mc_funcs_dirs(1, 32);
151 init_mc_funcs_dirs(2, 16);
152 init_mc_funcs_dirs(3, 8);
153 init_mc_funcs_dirs(4, 4);
154 }
155 }
156
157 #define define_itxfm(type_a, type_b, sz) \
158 void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_neon(uint8_t *_dst, \
159 ptrdiff_t stride, \
160 int16_t *_block, int eob)
161
162 #define define_itxfm_funcs(sz) \
163 define_itxfm(idct, idct, sz); \
164 define_itxfm(iadst, idct, sz); \
165 define_itxfm(idct, iadst, sz); \
166 define_itxfm(iadst, iadst, sz)
167
168 define_itxfm_funcs(4);
169 define_itxfm_funcs(8);
170 define_itxfm_funcs(16);
171 define_itxfm(idct, idct, 32);
172 define_itxfm(iwht, iwht, 4);
173
174
vp9dsp_itxfm_init_aarch64(VP9DSPContext * dsp)175 static av_cold void vp9dsp_itxfm_init_aarch64(VP9DSPContext *dsp)
176 {
177 int cpu_flags = av_get_cpu_flags();
178
179 if (have_neon(cpu_flags)) {
180 #define init_itxfm(tx, sz) \
181 dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_neon; \
182 dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_neon; \
183 dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_neon; \
184 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_neon
185
186 #define init_idct(tx, nm) \
187 dsp->itxfm_add[tx][DCT_DCT] = \
188 dsp->itxfm_add[tx][ADST_DCT] = \
189 dsp->itxfm_add[tx][DCT_ADST] = \
190 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_neon
191
192 init_itxfm(TX_4X4, 4x4);
193 init_itxfm(TX_8X8, 8x8);
194 init_itxfm(TX_16X16, 16x16);
195 init_idct(TX_32X32, idct_idct_32x32);
196 init_idct(4, iwht_iwht_4x4);
197 }
198 }
199
200 #define define_loop_filter(dir, wd, len) \
201 void ff_vp9_loop_filter_##dir##_##wd##_##len##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H)
202
203 #define define_loop_filters(wd, len) \
204 define_loop_filter(h, wd, len); \
205 define_loop_filter(v, wd, len)
206
207 define_loop_filters(4, 8);
208 define_loop_filters(8, 8);
209 define_loop_filters(16, 8);
210
211 define_loop_filters(16, 16);
212
213 define_loop_filters(44, 16);
214 define_loop_filters(48, 16);
215 define_loop_filters(84, 16);
216 define_loop_filters(88, 16);
217
vp9dsp_loopfilter_init_aarch64(VP9DSPContext * dsp)218 static av_cold void vp9dsp_loopfilter_init_aarch64(VP9DSPContext *dsp)
219 {
220 int cpu_flags = av_get_cpu_flags();
221
222 if (have_neon(cpu_flags)) {
223 dsp->loop_filter_8[0][1] = ff_vp9_loop_filter_v_4_8_neon;
224 dsp->loop_filter_8[0][0] = ff_vp9_loop_filter_h_4_8_neon;
225 dsp->loop_filter_8[1][1] = ff_vp9_loop_filter_v_8_8_neon;
226 dsp->loop_filter_8[1][0] = ff_vp9_loop_filter_h_8_8_neon;
227 dsp->loop_filter_8[2][1] = ff_vp9_loop_filter_v_16_8_neon;
228 dsp->loop_filter_8[2][0] = ff_vp9_loop_filter_h_16_8_neon;
229
230 dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_neon;
231 dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_neon;
232
233 dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_neon;
234 dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_neon;
235 dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_neon;
236 dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_neon;
237 dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_neon;
238 dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_neon;
239 dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_neon;
240 dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_neon;
241 }
242 }
243
ff_vp9dsp_init_aarch64(VP9DSPContext * dsp,int bpp)244 av_cold void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp, int bpp)
245 {
246 if (bpp == 10) {
247 ff_vp9dsp_init_10bpp_aarch64(dsp);
248 return;
249 } else if (bpp == 12) {
250 ff_vp9dsp_init_12bpp_aarch64(dsp);
251 return;
252 } else if (bpp != 8)
253 return;
254
255 vp9dsp_mc_init_aarch64(dsp);
256 vp9dsp_loopfilter_init_aarch64(dsp);
257 vp9dsp_itxfm_init_aarch64(dsp);
258 }
259