1 /*
2 * Copyright (c) 2017 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/mem_internal.h"
26 #include "libavutil/arm/cpu.h"
27 #include "vp9dsp_init.h"
28
29 #define declare_fpel(type, sz, suffix) \
30 void ff_vp9_##type##sz##suffix##_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 decl_mc_func(op, filter, dir, sz, bpp) \
35 void ff_vp9_##op##_##filter##sz##_##dir##_##bpp##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
36 const uint8_t *src, ptrdiff_t src_stride, \
37 int h, int mx, int my)
38
39 #define define_8tap_2d_fn(op, filter, sz, bpp) \
40 static void op##_##filter##sz##_hv_##bpp##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
41 const uint8_t *src, \
42 ptrdiff_t src_stride, \
43 int h, int mx, int my) \
44 { \
45 LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz * 2]); \
46 /* We only need h + 7 lines, but the horizontal filter assumes an \
47 * even number of rows, so filter h + 8 lines here. */ \
48 ff_vp9_put_##filter##sz##_h_##bpp##_neon(temp, 2 * sz, \
49 src - 3 * src_stride, src_stride, \
50 h + 8, mx, 0); \
51 ff_vp9_##op##_##filter##sz##_v_##bpp##_neon(dst, dst_stride, \
52 temp + 3 * 2 * sz, 2 * sz, \
53 h, 0, my); \
54 }
55
56 #define decl_filter_funcs(op, dir, sz, bpp) \
57 decl_mc_func(op, regular, dir, sz, bpp); \
58 decl_mc_func(op, sharp, dir, sz, bpp); \
59 decl_mc_func(op, smooth, dir, sz, bpp)
60
61 #define decl_mc_funcs(sz, bpp) \
62 decl_filter_funcs(put, h, sz, bpp); \
63 decl_filter_funcs(avg, h, sz, bpp); \
64 decl_filter_funcs(put, v, sz, bpp); \
65 decl_filter_funcs(avg, v, sz, bpp); \
66 decl_filter_funcs(put, hv, sz, bpp); \
67 decl_filter_funcs(avg, hv, sz, bpp)
68
69 declare_fpel(copy, 128, );
70 declare_fpel(copy, 64, );
71 declare_fpel(copy, 32, );
72 declare_fpel(copy, 16, );
73 declare_fpel(copy, 8, );
74 declare_fpel(avg, 64, _16);
75 declare_fpel(avg, 32, _16);
76 declare_fpel(avg, 16, _16);
77 declare_fpel(avg, 8, _16);
78 declare_fpel(avg, 4, _16);
79
80 decl_mc_funcs(64, BPP);
81 decl_mc_funcs(32, BPP);
82 decl_mc_funcs(16, BPP);
83 decl_mc_funcs(8, BPP);
84 decl_mc_funcs(4, BPP);
85
86 #define define_8tap_2d_funcs(sz, bpp) \
87 define_8tap_2d_fn(put, regular, sz, bpp) \
88 define_8tap_2d_fn(put, sharp, sz, bpp) \
89 define_8tap_2d_fn(put, smooth, sz, bpp) \
90 define_8tap_2d_fn(avg, regular, sz, bpp) \
91 define_8tap_2d_fn(avg, sharp, sz, bpp) \
92 define_8tap_2d_fn(avg, smooth, sz, bpp)
93
94 define_8tap_2d_funcs(64, BPP)
95 define_8tap_2d_funcs(32, BPP)
96 define_8tap_2d_funcs(16, BPP)
97 define_8tap_2d_funcs(8, BPP)
98 define_8tap_2d_funcs(4, BPP)
99
100
vp9dsp_mc_init_arm(VP9DSPContext * dsp)101 static av_cold void vp9dsp_mc_init_arm(VP9DSPContext *dsp)
102 {
103 int cpu_flags = av_get_cpu_flags();
104
105 if (have_neon(cpu_flags)) {
106 #define init_fpel(idx1, idx2, sz, type, suffix) \
107 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
108 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
109 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
110 dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##suffix##_neon
111
112 #define init_copy_avg(idx, sz1, sz2) \
113 init_fpel(idx, 0, sz2, copy, ); \
114 init_fpel(idx, 1, sz1, avg, _16)
115
116 #define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx, bpp) \
117 dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_##bpp##_neon
118
119 #define init_mc_funcs(idx, dir, mx, my, sz, pfx, bpp) \
120 init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx, bpp); \
121 init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx, bpp); \
122 init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx, bpp); \
123 init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx, bpp); \
124 init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx, bpp); \
125 init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx, bpp)
126
127 #define init_mc_funcs_dirs(idx, sz, bpp) \
128 init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_, bpp); \
129 init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_, bpp); \
130 init_mc_funcs(idx, hv, 1, 1, sz, , bpp)
131
132 init_copy_avg(0, 64, 128);
133 init_copy_avg(1, 32, 64);
134 init_copy_avg(2, 16, 32);
135 init_copy_avg(3, 8, 16);
136 init_copy_avg(4, 4, 8);
137
138 init_mc_funcs_dirs(0, 64, BPP);
139 init_mc_funcs_dirs(1, 32, BPP);
140 init_mc_funcs_dirs(2, 16, BPP);
141 init_mc_funcs_dirs(3, 8, BPP);
142 init_mc_funcs_dirs(4, 4, BPP);
143 }
144 }
145
146 #define define_itxfm2(type_a, type_b, sz, bpp) \
147 void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_##bpp##_neon(uint8_t *_dst, \
148 ptrdiff_t stride, \
149 int16_t *_block, int eob)
150 #define define_itxfm(type_a, type_b, sz, bpp) define_itxfm2(type_a, type_b, sz, bpp)
151
152 #define define_itxfm_funcs(sz, bpp) \
153 define_itxfm(idct, idct, sz, bpp); \
154 define_itxfm(iadst, idct, sz, bpp); \
155 define_itxfm(idct, iadst, sz, bpp); \
156 define_itxfm(iadst, iadst, sz, bpp)
157
158 define_itxfm_funcs(4, BPP);
159 define_itxfm_funcs(8, BPP);
160 define_itxfm_funcs(16, BPP);
161 define_itxfm(idct, idct, 32, BPP);
162 define_itxfm(iwht, iwht, 4, BPP);
163
164
vp9dsp_itxfm_init_arm(VP9DSPContext * dsp)165 static av_cold void vp9dsp_itxfm_init_arm(VP9DSPContext *dsp)
166 {
167 int cpu_flags = av_get_cpu_flags();
168
169 if (have_neon(cpu_flags)) {
170 #define init_itxfm2(tx, sz, bpp) \
171 dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_##bpp##_neon; \
172 dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_##bpp##_neon; \
173 dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_##bpp##_neon; \
174 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_##bpp##_neon
175 #define init_itxfm(tx, sz, bpp) init_itxfm2(tx, sz, bpp)
176
177 #define init_idct2(tx, nm, bpp) \
178 dsp->itxfm_add[tx][DCT_DCT] = \
179 dsp->itxfm_add[tx][ADST_DCT] = \
180 dsp->itxfm_add[tx][DCT_ADST] = \
181 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_##bpp##_neon
182 #define init_idct(tx, nm, bpp) init_idct2(tx, nm, bpp)
183
184 init_itxfm(TX_4X4, 4x4, BPP);
185 init_itxfm(TX_8X8, 8x8, BPP);
186 init_itxfm(TX_16X16, 16x16, BPP);
187 init_idct(TX_32X32, idct_idct_32x32, BPP);
188 init_idct(4, iwht_iwht_4x4, BPP);
189 }
190 }
191
192 #define define_loop_filter(dir, wd, size, bpp) \
193 void ff_vp9_loop_filter_##dir##_##wd##_##size##_##bpp##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H)
194
195 #define define_loop_filters(wd, size, bpp) \
196 define_loop_filter(h, wd, size, bpp); \
197 define_loop_filter(v, wd, size, bpp)
198
199 define_loop_filters(4, 8, BPP);
200 define_loop_filters(8, 8, BPP);
201 define_loop_filters(16, 8, BPP);
202
203 define_loop_filters(16, 16, BPP);
204
205 define_loop_filters(44, 16, BPP);
206 define_loop_filters(48, 16, BPP);
207 define_loop_filters(84, 16, BPP);
208 define_loop_filters(88, 16, BPP);
209
vp9dsp_loopfilter_init_arm(VP9DSPContext * dsp)210 static av_cold void vp9dsp_loopfilter_init_arm(VP9DSPContext *dsp)
211 {
212 int cpu_flags = av_get_cpu_flags();
213
214 if (have_neon(cpu_flags)) {
215 #define init_lpf_func_8(idx1, idx2, dir, wd, bpp) \
216 dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_8_##bpp##_neon
217
218 #define init_lpf_func_16(idx, dir, bpp) \
219 dsp->loop_filter_16[idx] = ff_vp9_loop_filter_##dir##_16_16_##bpp##_neon
220
221 #define init_lpf_func_mix2(idx1, idx2, idx3, dir, wd, bpp) \
222 dsp->loop_filter_mix2[idx1][idx2][idx3] = ff_vp9_loop_filter_##dir##_##wd##_16_##bpp##_neon
223
224 #define init_lpf_funcs_8_wd(idx, wd, bpp) \
225 init_lpf_func_8(idx, 0, h, wd, bpp); \
226 init_lpf_func_8(idx, 1, v, wd, bpp)
227
228 #define init_lpf_funcs_16(bpp) \
229 init_lpf_func_16(0, h, bpp); \
230 init_lpf_func_16(1, v, bpp)
231
232 #define init_lpf_funcs_mix2_wd(idx1, idx2, wd, bpp) \
233 init_lpf_func_mix2(idx1, idx2, 0, h, wd, bpp); \
234 init_lpf_func_mix2(idx1, idx2, 1, v, wd, bpp)
235
236 #define init_lpf_funcs_8(bpp) \
237 init_lpf_funcs_8_wd(0, 4, bpp); \
238 init_lpf_funcs_8_wd(1, 8, bpp); \
239 init_lpf_funcs_8_wd(2, 16, bpp)
240
241 #define init_lpf_funcs_mix2(bpp) \
242 init_lpf_funcs_mix2_wd(0, 0, 44, bpp); \
243 init_lpf_funcs_mix2_wd(0, 1, 48, bpp); \
244 init_lpf_funcs_mix2_wd(1, 0, 84, bpp); \
245 init_lpf_funcs_mix2_wd(1, 1, 88, bpp)
246
247 init_lpf_funcs_8(BPP);
248 init_lpf_funcs_16(BPP);
249 init_lpf_funcs_mix2(BPP);
250 }
251 }
252
INIT_FUNC(VP9DSPContext * dsp)253 av_cold void INIT_FUNC(VP9DSPContext *dsp)
254 {
255 vp9dsp_mc_init_arm(dsp);
256 vp9dsp_loopfilter_init_arm(dsp);
257 vp9dsp_itxfm_init_arm(dsp);
258 }
259