2 * Copyright (c) 2017 Google Inc.
4 * This file is part of FFmpeg.
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.
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.
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
23 #include "libavutil/attributes.h"
24 #include "libavutil/aarch64/cpu.h"
25 #include "vp9dsp_init.h"
27 #define declare_fpel(type, sz, suffix) \
28 void ff_vp9_##type##sz##suffix##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
29 const uint8_t *src, ptrdiff_t src_stride, \
30 int h, int mx, int my)
32 #define decl_mc_func(op, filter, dir, sz, bpp) \
33 void ff_vp9_##op##_##filter##sz##_##dir##_##bpp##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
34 const uint8_t *src, ptrdiff_t src_stride, \
35 int h, int mx, int my)
37 #define define_8tap_2d_fn(op, filter, sz, bpp) \
38 static void op##_##filter##sz##_hv_##bpp##_neon(uint8_t *dst, ptrdiff_t dst_stride, \
40 ptrdiff_t src_stride, \
41 int h, int mx, int my) \
43 LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz * 2]); \
44 /* We only need h + 7 lines, but the horizontal filter assumes an \
45 * even number of rows, so filter h + 8 lines here. */ \
46 ff_vp9_put_##filter##sz##_h_##bpp##_neon(temp, 2 * sz, \
47 src - 3 * src_stride, src_stride, \
49 ff_vp9_##op##_##filter##sz##_v_##bpp##_neon(dst, dst_stride, \
50 temp + 3 * 2 * sz, 2 * sz, \
54 #define decl_filter_funcs(op, dir, sz, bpp) \
55 decl_mc_func(op, regular, dir, sz, bpp); \
56 decl_mc_func(op, sharp, dir, sz, bpp); \
57 decl_mc_func(op, smooth, dir, sz, bpp)
59 #define decl_mc_funcs(sz, bpp) \
60 decl_filter_funcs(put, h, sz, bpp); \
61 decl_filter_funcs(avg, h, sz, bpp); \
62 decl_filter_funcs(put, v, sz, bpp); \
63 decl_filter_funcs(avg, v, sz, bpp); \
64 decl_filter_funcs(put, hv, sz, bpp); \
65 decl_filter_funcs(avg, hv, sz, bpp)
67 #define ff_vp9_copy32_neon ff_vp9_copy32_aarch64
68 #define ff_vp9_copy64_neon ff_vp9_copy64_aarch64
69 #define ff_vp9_copy128_neon ff_vp9_copy128_aarch64
71 declare_fpel(copy, 128, );
72 declare_fpel(copy, 64, );
73 declare_fpel(copy, 32, );
74 declare_fpel(copy, 16, );
75 declare_fpel(copy, 8, );
76 declare_fpel(avg, 64, _16);
77 declare_fpel(avg, 32, _16);
78 declare_fpel(avg, 16, _16);
79 declare_fpel(avg, 8, _16);
80 declare_fpel(avg, 4, _16);
82 decl_mc_funcs(64, BPP);
83 decl_mc_funcs(32, BPP);
84 decl_mc_funcs(16, BPP);
85 decl_mc_funcs(8, BPP);
86 decl_mc_funcs(4, BPP);
88 #define define_8tap_2d_funcs(sz, bpp) \
89 define_8tap_2d_fn(put, regular, sz, bpp) \
90 define_8tap_2d_fn(put, sharp, sz, bpp) \
91 define_8tap_2d_fn(put, smooth, sz, bpp) \
92 define_8tap_2d_fn(avg, regular, sz, bpp) \
93 define_8tap_2d_fn(avg, sharp, sz, bpp) \
94 define_8tap_2d_fn(avg, smooth, sz, bpp)
96 define_8tap_2d_funcs(64, BPP)
97 define_8tap_2d_funcs(32, BPP)
98 define_8tap_2d_funcs(16, BPP)
99 define_8tap_2d_funcs(8, BPP)
100 define_8tap_2d_funcs(4, BPP)
102 static av_cold void vp9dsp_mc_init_aarch64(VP9DSPContext *dsp)
104 int cpu_flags = av_get_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
112 #define init_copy(idx, sz, suffix) \
113 init_fpel(idx, 0, sz, copy, suffix)
115 #define init_avg(idx, sz, suffix) \
116 init_fpel(idx, 1, sz, avg, suffix)
118 #define init_copy_avg(idx, sz1, sz2) \
119 init_copy(idx, sz2, _neon); \
120 init_avg (idx, sz1, _16_neon)
122 if (have_armv8(cpu_flags)) {
123 init_copy(0, 128, _aarch64);
124 init_copy(1, 64, _aarch64);
125 init_copy(2, 32, _aarch64);
128 if (have_neon(cpu_flags)) {
129 #define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx, bpp) \
130 dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_##bpp##_neon
132 #define init_mc_funcs(idx, dir, mx, my, sz, pfx, bpp) \
133 init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx, bpp); \
134 init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx, bpp); \
135 init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx, bpp); \
136 init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx, bpp); \
137 init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx, bpp); \
138 init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx, bpp)
140 #define init_mc_funcs_dirs(idx, sz, bpp) \
141 init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_, bpp); \
142 init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_, bpp); \
143 init_mc_funcs(idx, hv, 1, 1, sz, , bpp)
146 init_avg(0, 64, _16_neon);
147 init_avg(1, 32, _16_neon);
148 init_avg(2, 16, _16_neon);
149 init_copy_avg(3, 8, 16);
150 init_copy_avg(4, 4, 8);
152 init_mc_funcs_dirs(0, 64, BPP);
153 init_mc_funcs_dirs(1, 32, BPP);
154 init_mc_funcs_dirs(2, 16, BPP);
155 init_mc_funcs_dirs(3, 8, BPP);
156 init_mc_funcs_dirs(4, 4, BPP);
160 #define define_itxfm2(type_a, type_b, sz, bpp) \
161 void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_##bpp##_neon(uint8_t *_dst, \
163 int16_t *_block, int eob)
164 #define define_itxfm(type_a, type_b, sz, bpp) define_itxfm2(type_a, type_b, sz, bpp)
166 #define define_itxfm_funcs(sz, bpp) \
167 define_itxfm(idct, idct, sz, bpp); \
168 define_itxfm(iadst, idct, sz, bpp); \
169 define_itxfm(idct, iadst, sz, bpp); \
170 define_itxfm(iadst, iadst, sz, bpp)
172 define_itxfm_funcs(4, BPP);
173 define_itxfm_funcs(8, BPP);
174 define_itxfm_funcs(16, BPP);
175 define_itxfm(idct, idct, 32, BPP);
176 define_itxfm(iwht, iwht, 4, BPP);
179 static av_cold void vp9dsp_itxfm_init_aarch64(VP9DSPContext *dsp)
181 int cpu_flags = av_get_cpu_flags();
183 if (have_neon(cpu_flags)) {
184 #define init_itxfm2(tx, sz, bpp) \
185 dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_##bpp##_neon; \
186 dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_##bpp##_neon; \
187 dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_##bpp##_neon; \
188 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_##bpp##_neon
189 #define init_itxfm(tx, sz, bpp) init_itxfm2(tx, sz, bpp)
191 #define init_idct2(tx, nm, bpp) \
192 dsp->itxfm_add[tx][DCT_DCT] = \
193 dsp->itxfm_add[tx][ADST_DCT] = \
194 dsp->itxfm_add[tx][DCT_ADST] = \
195 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_##bpp##_neon
196 #define init_idct(tx, nm, bpp) init_idct2(tx, nm, bpp)
198 init_itxfm(TX_4X4, 4x4, BPP);
199 init_itxfm(TX_8X8, 8x8, BPP);
200 init_itxfm(TX_16X16, 16x16, BPP);
201 init_idct(TX_32X32, idct_idct_32x32, BPP);
202 init_idct(4, iwht_iwht_4x4, BPP);
206 #define define_loop_filter(dir, wd, size, bpp) \
207 void ff_vp9_loop_filter_##dir##_##wd##_##size##_##bpp##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H)
209 #define define_loop_filters(wd, size, bpp) \
210 define_loop_filter(h, wd, size, bpp); \
211 define_loop_filter(v, wd, size, bpp)
213 define_loop_filters(4, 8, BPP);
214 define_loop_filters(8, 8, BPP);
215 define_loop_filters(16, 8, BPP);
217 define_loop_filters(16, 16, BPP);
219 define_loop_filters(44, 16, BPP);
220 define_loop_filters(48, 16, BPP);
221 define_loop_filters(84, 16, BPP);
222 define_loop_filters(88, 16, BPP);
224 static av_cold void vp9dsp_loopfilter_init_aarch64(VP9DSPContext *dsp)
226 int cpu_flags = av_get_cpu_flags();
228 if (have_neon(cpu_flags)) {
229 #define init_lpf_func_8(idx1, idx2, dir, wd, bpp) \
230 dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_8_##bpp##_neon
232 #define init_lpf_func_16(idx, dir, bpp) \
233 dsp->loop_filter_16[idx] = ff_vp9_loop_filter_##dir##_16_16_##bpp##_neon
235 #define init_lpf_func_mix2(idx1, idx2, idx3, dir, wd, bpp) \
236 dsp->loop_filter_mix2[idx1][idx2][idx3] = ff_vp9_loop_filter_##dir##_##wd##_16_##bpp##_neon
238 #define init_lpf_funcs_8_wd(idx, wd, bpp) \
239 init_lpf_func_8(idx, 0, h, wd, bpp); \
240 init_lpf_func_8(idx, 1, v, wd, bpp)
242 #define init_lpf_funcs_16(bpp) \
243 init_lpf_func_16(0, h, bpp); \
244 init_lpf_func_16(1, v, bpp)
246 #define init_lpf_funcs_mix2_wd(idx1, idx2, wd, bpp) \
247 init_lpf_func_mix2(idx1, idx2, 0, h, wd, bpp); \
248 init_lpf_func_mix2(idx1, idx2, 1, v, wd, bpp)
250 #define init_lpf_funcs_8(bpp) \
251 init_lpf_funcs_8_wd(0, 4, bpp); \
252 init_lpf_funcs_8_wd(1, 8, bpp); \
253 init_lpf_funcs_8_wd(2, 16, bpp)
255 #define init_lpf_funcs_mix2(bpp) \
256 init_lpf_funcs_mix2_wd(0, 0, 44, bpp); \
257 init_lpf_funcs_mix2_wd(0, 1, 48, bpp); \
258 init_lpf_funcs_mix2_wd(1, 0, 84, bpp); \
259 init_lpf_funcs_mix2_wd(1, 1, 88, bpp)
261 init_lpf_funcs_8(BPP);
262 init_lpf_funcs_16(BPP);
263 init_lpf_funcs_mix2(BPP);
267 av_cold void INIT_FUNC(VP9DSPContext *dsp)
269 vp9dsp_mc_init_aarch64(dsp);
270 vp9dsp_loopfilter_init_aarch64(dsp);
271 vp9dsp_itxfm_init_aarch64(dsp);