1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include <stdlib.h> 12 #include "loopfilter.h" 13 #include "onyxc_int.h" 14 15 typedef unsigned char uc; 16 17 static signed char vp8_signed_char_clamp(int t) { 18 t = (t < -128 ? -128 : t); 19 t = (t > 127 ? 127 : t); 20 return (signed char)t; 21 } 22 23 /* should we apply any filter at all ( 11111111 yes, 00000000 no) */ 24 static signed char vp8_filter_mask(uc limit, uc blimit, uc p3, uc p2, uc p1, 25 uc p0, uc q0, uc q1, uc q2, uc q3) { 26 signed char mask = 0; 27 mask |= (abs(p3 - p2) > limit); 28 mask |= (abs(p2 - p1) > limit); 29 mask |= (abs(p1 - p0) > limit); 30 mask |= (abs(q1 - q0) > limit); 31 mask |= (abs(q2 - q1) > limit); 32 mask |= (abs(q3 - q2) > limit); 33 mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit); 34 return mask - 1; 35 } 36 37 /* is there high variance internal edge ( 11111111 yes, 00000000 no) */ 38 static signed char vp8_hevmask(uc thresh, uc p1, uc p0, uc q0, uc q1) { 39 signed char hev = 0; 40 hev |= (abs(p1 - p0) > thresh) * -1; 41 hev |= (abs(q1 - q0) > thresh) * -1; 42 return hev; 43 } 44 45 static void vp8_filter(signed char mask, uc hev, uc *op1, uc *op0, uc *oq0, 46 uc *oq1) { 47 signed char ps0, qs0; 48 signed char ps1, qs1; 49 signed char filter_value, Filter1, Filter2; 50 signed char u; 51 52 ps1 = (signed char)*op1 ^ 0x80; 53 ps0 = (signed char)*op0 ^ 0x80; 54 qs0 = (signed char)*oq0 ^ 0x80; 55 qs1 = (signed char)*oq1 ^ 0x80; 56 57 /* add outer taps if we have high edge variance */ 58 filter_value = vp8_signed_char_clamp(ps1 - qs1); 59 filter_value &= hev; 60 61 /* inner taps */ 62 filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0)); 63 filter_value &= mask; 64 65 /* save bottom 3 bits so that we round one side +4 and the other +3 66 * if it equals 4 we'll set it to adjust by -1 to account for the fact 67 * we'd round it by 3 the other way 68 */ 69 Filter1 = vp8_signed_char_clamp(filter_value + 4); 70 Filter2 = vp8_signed_char_clamp(filter_value + 3); 71 Filter1 >>= 3; 72 Filter2 >>= 3; 73 u = vp8_signed_char_clamp(qs0 - Filter1); 74 *oq0 = u ^ 0x80; 75 u = vp8_signed_char_clamp(ps0 + Filter2); 76 *op0 = u ^ 0x80; 77 filter_value = Filter1; 78 79 /* outer tap adjustments */ 80 filter_value += 1; 81 filter_value >>= 1; 82 filter_value &= ~hev; 83 84 u = vp8_signed_char_clamp(qs1 - filter_value); 85 *oq1 = u ^ 0x80; 86 u = vp8_signed_char_clamp(ps1 + filter_value); 87 *op1 = u ^ 0x80; 88 } 89 void vp8_loop_filter_horizontal_edge_c(unsigned char *s, int p, /* pitch */ 90 const unsigned char *blimit, 91 const unsigned char *limit, 92 const unsigned char *thresh, int count) { 93 int hev = 0; /* high edge variance */ 94 signed char mask = 0; 95 int i = 0; 96 97 /* loop filter designed to work using chars so that we can make maximum use 98 * of 8 bit simd instructions. 99 */ 100 do { 101 mask = vp8_filter_mask(limit[0], blimit[0], s[-4 * p], s[-3 * p], s[-2 * p], 102 s[-1 * p], s[0 * p], s[1 * p], s[2 * p], s[3 * p]); 103 104 hev = vp8_hevmask(thresh[0], s[-2 * p], s[-1 * p], s[0 * p], s[1 * p]); 105 106 vp8_filter(mask, hev, s - 2 * p, s - 1 * p, s, s + 1 * p); 107 108 ++s; 109 } while (++i < count * 8); 110 } 111 112 void vp8_loop_filter_vertical_edge_c(unsigned char *s, int p, 113 const unsigned char *blimit, 114 const unsigned char *limit, 115 const unsigned char *thresh, int count) { 116 int hev = 0; /* high edge variance */ 117 signed char mask = 0; 118 int i = 0; 119 120 /* loop filter designed to work using chars so that we can make maximum use 121 * of 8 bit simd instructions. 122 */ 123 do { 124 mask = vp8_filter_mask(limit[0], blimit[0], s[-4], s[-3], s[-2], s[-1], 125 s[0], s[1], s[2], s[3]); 126 127 hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); 128 129 vp8_filter(mask, hev, s - 2, s - 1, s, s + 1); 130 131 s += p; 132 } while (++i < count * 8); 133 } 134 135 static void vp8_mbfilter(signed char mask, uc hev, uc *op2, uc *op1, uc *op0, 136 uc *oq0, uc *oq1, uc *oq2) { 137 signed char s, u; 138 signed char filter_value, Filter1, Filter2; 139 signed char ps2 = (signed char)*op2 ^ 0x80; 140 signed char ps1 = (signed char)*op1 ^ 0x80; 141 signed char ps0 = (signed char)*op0 ^ 0x80; 142 signed char qs0 = (signed char)*oq0 ^ 0x80; 143 signed char qs1 = (signed char)*oq1 ^ 0x80; 144 signed char qs2 = (signed char)*oq2 ^ 0x80; 145 146 /* add outer taps if we have high edge variance */ 147 filter_value = vp8_signed_char_clamp(ps1 - qs1); 148 filter_value = vp8_signed_char_clamp(filter_value + 3 * (qs0 - ps0)); 149 filter_value &= mask; 150 151 Filter2 = filter_value; 152 Filter2 &= hev; 153 154 /* save bottom 3 bits so that we round one side +4 and the other +3 */ 155 Filter1 = vp8_signed_char_clamp(Filter2 + 4); 156 Filter2 = vp8_signed_char_clamp(Filter2 + 3); 157 Filter1 >>= 3; 158 Filter2 >>= 3; 159 qs0 = vp8_signed_char_clamp(qs0 - Filter1); 160 ps0 = vp8_signed_char_clamp(ps0 + Filter2); 161 162 /* only apply wider filter if not high edge variance */ 163 filter_value &= ~hev; 164 Filter2 = filter_value; 165 166 /* roughly 3/7th difference across boundary */ 167 u = vp8_signed_char_clamp((63 + Filter2 * 27) >> 7); 168 s = vp8_signed_char_clamp(qs0 - u); 169 *oq0 = s ^ 0x80; 170 s = vp8_signed_char_clamp(ps0 + u); 171 *op0 = s ^ 0x80; 172 173 /* roughly 2/7th difference across boundary */ 174 u = vp8_signed_char_clamp((63 + Filter2 * 18) >> 7); 175 s = vp8_signed_char_clamp(qs1 - u); 176 *oq1 = s ^ 0x80; 177 s = vp8_signed_char_clamp(ps1 + u); 178 *op1 = s ^ 0x80; 179 180 /* roughly 1/7th difference across boundary */ 181 u = vp8_signed_char_clamp((63 + Filter2 * 9) >> 7); 182 s = vp8_signed_char_clamp(qs2 - u); 183 *oq2 = s ^ 0x80; 184 s = vp8_signed_char_clamp(ps2 + u); 185 *op2 = s ^ 0x80; 186 } 187 188 void vp8_mbloop_filter_horizontal_edge_c(unsigned char *s, int p, 189 const unsigned char *blimit, 190 const unsigned char *limit, 191 const unsigned char *thresh, 192 int count) { 193 signed char hev = 0; /* high edge variance */ 194 signed char mask = 0; 195 int i = 0; 196 197 /* loop filter designed to work using chars so that we can make maximum use 198 * of 8 bit simd instructions. 199 */ 200 do { 201 mask = vp8_filter_mask(limit[0], blimit[0], s[-4 * p], s[-3 * p], s[-2 * p], 202 s[-1 * p], s[0 * p], s[1 * p], s[2 * p], s[3 * p]); 203 204 hev = vp8_hevmask(thresh[0], s[-2 * p], s[-1 * p], s[0 * p], s[1 * p]); 205 206 vp8_mbfilter(mask, hev, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p, 207 s + 2 * p); 208 209 ++s; 210 } while (++i < count * 8); 211 } 212 213 void vp8_mbloop_filter_vertical_edge_c(unsigned char *s, int p, 214 const unsigned char *blimit, 215 const unsigned char *limit, 216 const unsigned char *thresh, int count) { 217 signed char hev = 0; /* high edge variance */ 218 signed char mask = 0; 219 int i = 0; 220 221 do { 222 mask = vp8_filter_mask(limit[0], blimit[0], s[-4], s[-3], s[-2], s[-1], 223 s[0], s[1], s[2], s[3]); 224 225 hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); 226 227 vp8_mbfilter(mask, hev, s - 3, s - 2, s - 1, s, s + 1, s + 2); 228 229 s += p; 230 } while (++i < count * 8); 231 } 232 233 /* should we apply any filter at all ( 11111111 yes, 00000000 no) */ 234 static signed char vp8_simple_filter_mask(uc blimit, uc p1, uc p0, uc q0, 235 uc q1) { 236 /* Why does this cause problems for win32? 237 * error C2143: syntax error : missing ';' before 'type' 238 * (void) limit; 239 */ 240 signed char mask = (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 <= blimit) * -1; 241 return mask; 242 } 243 244 static void vp8_simple_filter(signed char mask, uc *op1, uc *op0, uc *oq0, 245 uc *oq1) { 246 signed char filter_value, Filter1, Filter2; 247 signed char p1 = (signed char)*op1 ^ 0x80; 248 signed char p0 = (signed char)*op0 ^ 0x80; 249 signed char q0 = (signed char)*oq0 ^ 0x80; 250 signed char q1 = (signed char)*oq1 ^ 0x80; 251 signed char u; 252 253 filter_value = vp8_signed_char_clamp(p1 - q1); 254 filter_value = vp8_signed_char_clamp(filter_value + 3 * (q0 - p0)); 255 filter_value &= mask; 256 257 /* save bottom 3 bits so that we round one side +4 and the other +3 */ 258 Filter1 = vp8_signed_char_clamp(filter_value + 4); 259 Filter1 >>= 3; 260 u = vp8_signed_char_clamp(q0 - Filter1); 261 *oq0 = u ^ 0x80; 262 263 Filter2 = vp8_signed_char_clamp(filter_value + 3); 264 Filter2 >>= 3; 265 u = vp8_signed_char_clamp(p0 + Filter2); 266 *op0 = u ^ 0x80; 267 } 268 269 void vp8_loop_filter_simple_horizontal_edge_c(unsigned char *s, int p, 270 const unsigned char *blimit) { 271 signed char mask = 0; 272 int i = 0; 273 274 do { 275 mask = vp8_simple_filter_mask(blimit[0], s[-2 * p], s[-1 * p], s[0 * p], 276 s[1 * p]); 277 vp8_simple_filter(mask, s - 2 * p, s - 1 * p, s, s + 1 * p); 278 ++s; 279 } while (++i < 16); 280 } 281 282 void vp8_loop_filter_simple_vertical_edge_c(unsigned char *s, int p, 283 const unsigned char *blimit) { 284 signed char mask = 0; 285 int i = 0; 286 287 do { 288 mask = vp8_simple_filter_mask(blimit[0], s[-2], s[-1], s[0], s[1]); 289 vp8_simple_filter(mask, s - 2, s - 1, s, s + 1); 290 s += p; 291 } while (++i < 16); 292 } 293 294 /* Horizontal MB filtering */ 295 void vp8_loop_filter_mbh_c(unsigned char *y_ptr, unsigned char *u_ptr, 296 unsigned char *v_ptr, int y_stride, int uv_stride, 297 loop_filter_info *lfi) { 298 vp8_mbloop_filter_horizontal_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, 299 lfi->hev_thr, 2); 300 301 if (u_ptr) { 302 vp8_mbloop_filter_horizontal_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, 303 lfi->hev_thr, 1); 304 } 305 306 if (v_ptr) { 307 vp8_mbloop_filter_horizontal_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, 308 lfi->hev_thr, 1); 309 } 310 } 311 312 /* Vertical MB Filtering */ 313 void vp8_loop_filter_mbv_c(unsigned char *y_ptr, unsigned char *u_ptr, 314 unsigned char *v_ptr, int y_stride, int uv_stride, 315 loop_filter_info *lfi) { 316 vp8_mbloop_filter_vertical_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, 317 lfi->hev_thr, 2); 318 319 if (u_ptr) { 320 vp8_mbloop_filter_vertical_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, 321 lfi->hev_thr, 1); 322 } 323 324 if (v_ptr) { 325 vp8_mbloop_filter_vertical_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, 326 lfi->hev_thr, 1); 327 } 328 } 329 330 /* Horizontal B Filtering */ 331 void vp8_loop_filter_bh_c(unsigned char *y_ptr, unsigned char *u_ptr, 332 unsigned char *v_ptr, int y_stride, int uv_stride, 333 loop_filter_info *lfi) { 334 vp8_loop_filter_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, lfi->blim, 335 lfi->lim, lfi->hev_thr, 2); 336 vp8_loop_filter_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, lfi->blim, 337 lfi->lim, lfi->hev_thr, 2); 338 vp8_loop_filter_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, lfi->blim, 339 lfi->lim, lfi->hev_thr, 2); 340 341 if (u_ptr) { 342 vp8_loop_filter_horizontal_edge_c(u_ptr + 4 * uv_stride, uv_stride, 343 lfi->blim, lfi->lim, lfi->hev_thr, 1); 344 } 345 346 if (v_ptr) { 347 vp8_loop_filter_horizontal_edge_c(v_ptr + 4 * uv_stride, uv_stride, 348 lfi->blim, lfi->lim, lfi->hev_thr, 1); 349 } 350 } 351 352 void vp8_loop_filter_bhs_c(unsigned char *y_ptr, int y_stride, 353 const unsigned char *blimit) { 354 vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, 355 blimit); 356 vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, 357 blimit); 358 vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, 359 blimit); 360 } 361 362 /* Vertical B Filtering */ 363 void vp8_loop_filter_bv_c(unsigned char *y_ptr, unsigned char *u_ptr, 364 unsigned char *v_ptr, int y_stride, int uv_stride, 365 loop_filter_info *lfi) { 366 vp8_loop_filter_vertical_edge_c(y_ptr + 4, y_stride, lfi->blim, lfi->lim, 367 lfi->hev_thr, 2); 368 vp8_loop_filter_vertical_edge_c(y_ptr + 8, y_stride, lfi->blim, lfi->lim, 369 lfi->hev_thr, 2); 370 vp8_loop_filter_vertical_edge_c(y_ptr + 12, y_stride, lfi->blim, lfi->lim, 371 lfi->hev_thr, 2); 372 373 if (u_ptr) { 374 vp8_loop_filter_vertical_edge_c(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, 375 lfi->hev_thr, 1); 376 } 377 378 if (v_ptr) { 379 vp8_loop_filter_vertical_edge_c(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, 380 lfi->hev_thr, 1); 381 } 382 } 383 384 void vp8_loop_filter_bvs_c(unsigned char *y_ptr, int y_stride, 385 const unsigned char *blimit) { 386 vp8_loop_filter_simple_vertical_edge_c(y_ptr + 4, y_stride, blimit); 387 vp8_loop_filter_simple_vertical_edge_c(y_ptr + 8, y_stride, blimit); 388 vp8_loop_filter_simple_vertical_edge_c(y_ptr + 12, y_stride, blimit); 389 } 390