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