vp9dsp_init_arm.c (10176B)
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/mem_internal.h" 26 #include "libavutil/arm/cpu.h" 27 #include "libavcodec/vp9dsp.h" 28 #include "vp9dsp_init.h" 29 30 #define declare_fpel(type, sz) \ 31 void ff_vp9_##type##sz##_neon(uint8_t *dst, ptrdiff_t dst_stride, \ 32 const uint8_t *src, ptrdiff_t src_stride, \ 33 int h, int mx, int my) 34 35 #define declare_copy_avg(sz) \ 36 declare_fpel(copy, sz); \ 37 declare_fpel(avg , sz) 38 39 #define decl_mc_func(op, filter, dir, sz) \ 40 void ff_vp9_##op##_##filter##sz##_##dir##_neon(uint8_t *dst, ptrdiff_t dst_stride, \ 41 const uint8_t *src, ptrdiff_t src_stride, \ 42 int h, int mx, int my) 43 44 #define define_8tap_2d_fn(op, filter, sz) \ 45 static void op##_##filter##sz##_hv_neon(uint8_t *dst, ptrdiff_t dst_stride, \ 46 const uint8_t *src, ptrdiff_t src_stride, \ 47 int h, int mx, int my) \ 48 { \ 49 LOCAL_ALIGNED_16(uint8_t, temp, [((1 + (sz < 64)) * sz + 8) * sz]); \ 50 /* We only need h + 7 lines, but the horizontal filter assumes an \ 51 * even number of rows, so filter h + 8 lines here. */ \ 52 ff_vp9_put_##filter##sz##_h_neon(temp, sz, \ 53 src - 3 * src_stride, src_stride, \ 54 h + 8, mx, 0); \ 55 ff_vp9_##op##_##filter##sz##_v_neon(dst, dst_stride, \ 56 temp + 3 * sz, sz, \ 57 h, 0, my); \ 58 } 59 60 #define decl_filter_funcs(op, dir, sz) \ 61 decl_mc_func(op, regular, dir, sz); \ 62 decl_mc_func(op, sharp, dir, sz); \ 63 decl_mc_func(op, smooth, dir, sz) 64 65 #define decl_mc_funcs(sz) \ 66 decl_filter_funcs(put, h, sz); \ 67 decl_filter_funcs(avg, h, sz); \ 68 decl_filter_funcs(put, v, sz); \ 69 decl_filter_funcs(avg, v, sz); \ 70 decl_filter_funcs(put, hv, sz); \ 71 decl_filter_funcs(avg, hv, sz) 72 73 declare_copy_avg(64); 74 declare_copy_avg(32); 75 declare_copy_avg(16); 76 declare_copy_avg(8); 77 declare_copy_avg(4); 78 79 decl_mc_funcs(64); 80 decl_mc_funcs(32); 81 decl_mc_funcs(16); 82 decl_mc_funcs(8); 83 decl_mc_funcs(4); 84 85 #define define_8tap_2d_funcs(sz) \ 86 define_8tap_2d_fn(put, regular, sz) \ 87 define_8tap_2d_fn(put, sharp, sz) \ 88 define_8tap_2d_fn(put, smooth, sz) \ 89 define_8tap_2d_fn(avg, regular, sz) \ 90 define_8tap_2d_fn(avg, sharp, sz) \ 91 define_8tap_2d_fn(avg, smooth, sz) 92 93 define_8tap_2d_funcs(64) 94 define_8tap_2d_funcs(32) 95 define_8tap_2d_funcs(16) 96 define_8tap_2d_funcs(8) 97 define_8tap_2d_funcs(4) 98 99 100 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) \ 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##_neon 110 111 #define init_copy_avg(idx, sz) \ 112 init_fpel(idx, 0, sz, copy); \ 113 init_fpel(idx, 1, sz, avg) 114 115 #define init_mc_func(idx1, idx2, op, filter, fname, dir, mx, my, sz, pfx) \ 116 dsp->mc[idx1][filter][idx2][mx][my] = pfx##op##_##fname##sz##_##dir##_neon 117 118 #define init_mc_funcs(idx, dir, mx, my, sz, pfx) \ 119 init_mc_func(idx, 0, put, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \ 120 init_mc_func(idx, 0, put, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \ 121 init_mc_func(idx, 0, put, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx); \ 122 init_mc_func(idx, 1, avg, FILTER_8TAP_REGULAR, regular, dir, mx, my, sz, pfx); \ 123 init_mc_func(idx, 1, avg, FILTER_8TAP_SHARP, sharp, dir, mx, my, sz, pfx); \ 124 init_mc_func(idx, 1, avg, FILTER_8TAP_SMOOTH, smooth, dir, mx, my, sz, pfx) 125 126 #define init_mc_funcs_dirs(idx, sz) \ 127 init_mc_funcs(idx, h, 1, 0, sz, ff_vp9_); \ 128 init_mc_funcs(idx, v, 0, 1, sz, ff_vp9_); \ 129 init_mc_funcs(idx, hv, 1, 1, sz,) 130 131 init_copy_avg(0, 64); 132 init_copy_avg(1, 32); 133 init_copy_avg(2, 16); 134 init_copy_avg(3, 8); 135 init_copy_avg(4, 4); 136 137 init_mc_funcs_dirs(0, 64); 138 init_mc_funcs_dirs(1, 32); 139 init_mc_funcs_dirs(2, 16); 140 init_mc_funcs_dirs(3, 8); 141 init_mc_funcs_dirs(4, 4); 142 } 143 } 144 145 #define define_itxfm(type_a, type_b, sz) \ 146 void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_neon(uint8_t *_dst, \ 147 ptrdiff_t stride, \ 148 int16_t *_block, int eob) 149 150 #define define_itxfm_funcs(sz) \ 151 define_itxfm(idct, idct, sz); \ 152 define_itxfm(iadst, idct, sz); \ 153 define_itxfm(idct, iadst, sz); \ 154 define_itxfm(iadst, iadst, sz) 155 156 define_itxfm_funcs(4); 157 define_itxfm_funcs(8); 158 define_itxfm_funcs(16); 159 define_itxfm(idct, idct, 32); 160 define_itxfm(iwht, iwht, 4); 161 162 163 static av_cold void vp9dsp_itxfm_init_arm(VP9DSPContext *dsp) 164 { 165 int cpu_flags = av_get_cpu_flags(); 166 167 if (have_neon(cpu_flags)) { 168 #define init_itxfm(tx, sz) \ 169 dsp->itxfm_add[tx][DCT_DCT] = ff_vp9_idct_idct_##sz##_add_neon; \ 170 dsp->itxfm_add[tx][DCT_ADST] = ff_vp9_iadst_idct_##sz##_add_neon; \ 171 dsp->itxfm_add[tx][ADST_DCT] = ff_vp9_idct_iadst_##sz##_add_neon; \ 172 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_neon 173 174 #define init_idct(tx, nm) \ 175 dsp->itxfm_add[tx][DCT_DCT] = \ 176 dsp->itxfm_add[tx][ADST_DCT] = \ 177 dsp->itxfm_add[tx][DCT_ADST] = \ 178 dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_neon 179 180 init_itxfm(TX_4X4, 4x4); 181 init_itxfm(TX_8X8, 8x8); 182 init_itxfm(TX_16X16, 16x16); 183 init_idct(TX_32X32, idct_idct_32x32); 184 init_idct(4, iwht_iwht_4x4); 185 } 186 } 187 188 #define define_loop_filter(dir, wd, size) \ 189 void ff_vp9_loop_filter_##dir##_##wd##_##size##_neon(uint8_t *dst, ptrdiff_t stride, int E, int I, int H) 190 191 #define define_loop_filters(wd, size) \ 192 define_loop_filter(h, wd, size); \ 193 define_loop_filter(v, wd, size) 194 195 define_loop_filters(4, 8); 196 define_loop_filters(8, 8); 197 define_loop_filters(16, 8); 198 define_loop_filters(16, 16); 199 200 define_loop_filters(44, 16); 201 202 #define lf_mix_fn(dir, wd1, wd2, stridea) \ 203 static void loop_filter_##dir##_##wd1##wd2##_16_neon(uint8_t *dst, \ 204 ptrdiff_t stride, \ 205 int E, int I, int H) \ 206 { \ 207 ff_vp9_loop_filter_##dir##_##wd1##_8_neon(dst, stride, E & 0xff, I & 0xff, H & 0xff); \ 208 ff_vp9_loop_filter_##dir##_##wd2##_8_neon(dst + 8 * stridea, stride, E >> 8, I >> 8, H >> 8); \ 209 } 210 211 #define lf_mix_fns(wd1, wd2) \ 212 lf_mix_fn(h, wd1, wd2, stride) \ 213 lf_mix_fn(v, wd1, wd2, sizeof(uint8_t)) 214 215 lf_mix_fns(4, 8) 216 lf_mix_fns(8, 4) 217 lf_mix_fns(8, 8) 218 219 static av_cold void vp9dsp_loopfilter_init_arm(VP9DSPContext *dsp) 220 { 221 int cpu_flags = av_get_cpu_flags(); 222 223 if (have_neon(cpu_flags)) { 224 dsp->loop_filter_8[0][1] = ff_vp9_loop_filter_v_4_8_neon; 225 dsp->loop_filter_8[0][0] = ff_vp9_loop_filter_h_4_8_neon; 226 dsp->loop_filter_8[1][1] = ff_vp9_loop_filter_v_8_8_neon; 227 dsp->loop_filter_8[1][0] = ff_vp9_loop_filter_h_8_8_neon; 228 dsp->loop_filter_8[2][1] = ff_vp9_loop_filter_v_16_8_neon; 229 dsp->loop_filter_8[2][0] = ff_vp9_loop_filter_h_16_8_neon; 230 231 dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_neon; 232 dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_neon; 233 234 dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_neon; 235 dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_neon; 236 dsp->loop_filter_mix2[0][1][0] = loop_filter_h_48_16_neon; 237 dsp->loop_filter_mix2[0][1][1] = loop_filter_v_48_16_neon; 238 dsp->loop_filter_mix2[1][0][0] = loop_filter_h_84_16_neon; 239 dsp->loop_filter_mix2[1][0][1] = loop_filter_v_84_16_neon; 240 dsp->loop_filter_mix2[1][1][0] = loop_filter_h_88_16_neon; 241 dsp->loop_filter_mix2[1][1][1] = loop_filter_v_88_16_neon; 242 } 243 } 244 245 av_cold void ff_vp9dsp_init_arm(VP9DSPContext *dsp, int bpp) 246 { 247 if (bpp == 10) { 248 ff_vp9dsp_init_10bpp_arm(dsp); 249 return; 250 } else if (bpp == 12) { 251 ff_vp9dsp_init_12bpp_arm(dsp); 252 return; 253 } else if (bpp != 8) 254 return; 255 256 vp9dsp_mc_init_arm(dsp); 257 vp9dsp_loopfilter_init_arm(dsp); 258 vp9dsp_itxfm_init_arm(dsp); 259 }