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 "./vpx_config.h" 12#include "vp9/common/vp9_common.h" 13#include "vp9/common/vp9_loopfilter.h" 14#include "vp9/common/vp9_onyxc_int.h" 15 16static INLINE int8_t signed_char_clamp(int t) { 17 return (int8_t)clamp(t, -128, 127); 18} 19 20// should we apply any filter at all: 11111111 yes, 00000000 no 21static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, 22 uint8_t p3, uint8_t p2, 23 uint8_t p1, uint8_t p0, 24 uint8_t q0, uint8_t q1, 25 uint8_t q2, uint8_t q3) { 26 int8_t mask = 0; 27 mask |= (abs(p3 - p2) > limit) * -1; 28 mask |= (abs(p2 - p1) > limit) * -1; 29 mask |= (abs(p1 - p0) > limit) * -1; 30 mask |= (abs(q1 - q0) > limit) * -1; 31 mask |= (abs(q2 - q1) > limit) * -1; 32 mask |= (abs(q3 - q2) > limit) * -1; 33 mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; 34 return ~mask; 35} 36 37static INLINE int8_t flat_mask4(uint8_t thresh, 38 uint8_t p3, uint8_t p2, 39 uint8_t p1, uint8_t p0, 40 uint8_t q0, uint8_t q1, 41 uint8_t q2, uint8_t q3) { 42 int8_t mask = 0; 43 mask |= (abs(p1 - p0) > thresh) * -1; 44 mask |= (abs(q1 - q0) > thresh) * -1; 45 mask |= (abs(p2 - p0) > thresh) * -1; 46 mask |= (abs(q2 - q0) > thresh) * -1; 47 mask |= (abs(p3 - p0) > thresh) * -1; 48 mask |= (abs(q3 - q0) > thresh) * -1; 49 return ~mask; 50} 51 52static INLINE int8_t flat_mask5(uint8_t thresh, 53 uint8_t p4, uint8_t p3, 54 uint8_t p2, uint8_t p1, 55 uint8_t p0, uint8_t q0, 56 uint8_t q1, uint8_t q2, 57 uint8_t q3, uint8_t q4) { 58 int8_t mask = ~flat_mask4(thresh, p3, p2, p1, p0, q0, q1, q2, q3); 59 mask |= (abs(p4 - p0) > thresh) * -1; 60 mask |= (abs(q4 - q0) > thresh) * -1; 61 return ~mask; 62} 63 64// is there high edge variance internal edge: 11111111 yes, 00000000 no 65static INLINE int8_t hev_mask(uint8_t thresh, uint8_t p1, uint8_t p0, 66 uint8_t q0, uint8_t q1) { 67 int8_t hev = 0; 68 hev |= (abs(p1 - p0) > thresh) * -1; 69 hev |= (abs(q1 - q0) > thresh) * -1; 70 return hev; 71} 72 73static INLINE void filter4(int8_t mask, uint8_t thresh, uint8_t *op1, 74 uint8_t *op0, uint8_t *oq0, uint8_t *oq1) { 75 int8_t filter1, filter2; 76 77 const int8_t ps1 = (int8_t) *op1 ^ 0x80; 78 const int8_t ps0 = (int8_t) *op0 ^ 0x80; 79 const int8_t qs0 = (int8_t) *oq0 ^ 0x80; 80 const int8_t qs1 = (int8_t) *oq1 ^ 0x80; 81 const uint8_t hev = hev_mask(thresh, *op1, *op0, *oq0, *oq1); 82 83 // add outer taps if we have high edge variance 84 int8_t filter = signed_char_clamp(ps1 - qs1) & hev; 85 86 // inner taps 87 filter = signed_char_clamp(filter + 3 * (qs0 - ps0)) & mask; 88 89 // save bottom 3 bits so that we round one side +4 and the other +3 90 // if it equals 4 we'll set to adjust by -1 to account for the fact 91 // we'd round 3 the other way 92 filter1 = signed_char_clamp(filter + 4) >> 3; 93 filter2 = signed_char_clamp(filter + 3) >> 3; 94 95 *oq0 = signed_char_clamp(qs0 - filter1) ^ 0x80; 96 *op0 = signed_char_clamp(ps0 + filter2) ^ 0x80; 97 98 // outer tap adjustments 99 filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev; 100 101 *oq1 = signed_char_clamp(qs1 - filter) ^ 0x80; 102 *op1 = signed_char_clamp(ps1 + filter) ^ 0x80; 103} 104 105void vp9_lpf_horizontal_4_c(uint8_t *s, int p /* pitch */, 106 const uint8_t *blimit, const uint8_t *limit, 107 const uint8_t *thresh, int count) { 108 int i; 109 110 // loop filter designed to work using chars so that we can make maximum use 111 // of 8 bit simd instructions. 112 for (i = 0; i < 8 * count; ++i) { 113 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; 114 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; 115 const int8_t mask = filter_mask(*limit, *blimit, 116 p3, p2, p1, p0, q0, q1, q2, q3); 117 filter4(mask, *thresh, s - 2 * p, s - 1 * p, s, s + 1 * p); 118 ++s; 119 } 120} 121 122void vp9_lpf_horizontal_4_dual_c(uint8_t *s, int p, const uint8_t *blimit0, 123 const uint8_t *limit0, const uint8_t *thresh0, 124 const uint8_t *blimit1, const uint8_t *limit1, 125 const uint8_t *thresh1) { 126 vp9_lpf_horizontal_4_c(s, p, blimit0, limit0, thresh0, 1); 127 vp9_lpf_horizontal_4_c(s + 8, p, blimit1, limit1, thresh1, 1); 128} 129 130void vp9_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit, 131 const uint8_t *limit, const uint8_t *thresh, 132 int count) { 133 int i; 134 135 // loop filter designed to work using chars so that we can make maximum use 136 // of 8 bit simd instructions. 137 for (i = 0; i < 8 * count; ++i) { 138 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; 139 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; 140 const int8_t mask = filter_mask(*limit, *blimit, 141 p3, p2, p1, p0, q0, q1, q2, q3); 142 filter4(mask, *thresh, s - 2, s - 1, s, s + 1); 143 s += pitch; 144 } 145} 146 147void vp9_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, 148 const uint8_t *limit0, const uint8_t *thresh0, 149 const uint8_t *blimit1, const uint8_t *limit1, 150 const uint8_t *thresh1) { 151 vp9_lpf_vertical_4_c(s, pitch, blimit0, limit0, thresh0, 1); 152 vp9_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, 153 thresh1, 1); 154} 155 156static INLINE void filter8(int8_t mask, uint8_t thresh, uint8_t flat, 157 uint8_t *op3, uint8_t *op2, 158 uint8_t *op1, uint8_t *op0, 159 uint8_t *oq0, uint8_t *oq1, 160 uint8_t *oq2, uint8_t *oq3) { 161 if (flat && mask) { 162 const uint8_t p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; 163 const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3; 164 165 // 7-tap filter [1, 1, 1, 2, 1, 1, 1] 166 *op2 = ROUND_POWER_OF_TWO(p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0, 3); 167 *op1 = ROUND_POWER_OF_TWO(p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1, 3); 168 *op0 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2, 3); 169 *oq0 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3, 3); 170 *oq1 = ROUND_POWER_OF_TWO(p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3, 3); 171 *oq2 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3, 3); 172 } else { 173 filter4(mask, thresh, op1, op0, oq0, oq1); 174 } 175} 176 177void vp9_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit, 178 const uint8_t *limit, const uint8_t *thresh, 179 int count) { 180 int i; 181 182 // loop filter designed to work using chars so that we can make maximum use 183 // of 8 bit simd instructions. 184 for (i = 0; i < 8 * count; ++i) { 185 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; 186 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; 187 188 const int8_t mask = filter_mask(*limit, *blimit, 189 p3, p2, p1, p0, q0, q1, q2, q3); 190 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); 191 filter8(mask, *thresh, flat, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, 192 s, s + 1 * p, s + 2 * p, s + 3 * p); 193 ++s; 194 } 195} 196 197void vp9_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0, 198 const uint8_t *limit0, const uint8_t *thresh0, 199 const uint8_t *blimit1, const uint8_t *limit1, 200 const uint8_t *thresh1) { 201 vp9_lpf_horizontal_8_c(s, p, blimit0, limit0, thresh0, 1); 202 vp9_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, 1); 203} 204 205void vp9_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit, 206 const uint8_t *limit, const uint8_t *thresh, 207 int count) { 208 int i; 209 210 for (i = 0; i < 8 * count; ++i) { 211 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; 212 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; 213 const int8_t mask = filter_mask(*limit, *blimit, 214 p3, p2, p1, p0, q0, q1, q2, q3); 215 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); 216 filter8(mask, *thresh, flat, s - 4, s - 3, s - 2, s - 1, 217 s, s + 1, s + 2, s + 3); 218 s += pitch; 219 } 220} 221 222void vp9_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0, 223 const uint8_t *limit0, const uint8_t *thresh0, 224 const uint8_t *blimit1, const uint8_t *limit1, 225 const uint8_t *thresh1) { 226 vp9_lpf_vertical_8_c(s, pitch, blimit0, limit0, thresh0, 1); 227 vp9_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, 228 thresh1, 1); 229} 230 231static INLINE void filter16(int8_t mask, uint8_t thresh, 232 uint8_t flat, uint8_t flat2, 233 uint8_t *op7, uint8_t *op6, 234 uint8_t *op5, uint8_t *op4, 235 uint8_t *op3, uint8_t *op2, 236 uint8_t *op1, uint8_t *op0, 237 uint8_t *oq0, uint8_t *oq1, 238 uint8_t *oq2, uint8_t *oq3, 239 uint8_t *oq4, uint8_t *oq5, 240 uint8_t *oq6, uint8_t *oq7) { 241 if (flat2 && flat && mask) { 242 const uint8_t p7 = *op7, p6 = *op6, p5 = *op5, p4 = *op4, 243 p3 = *op3, p2 = *op2, p1 = *op1, p0 = *op0; 244 245 const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, 246 q4 = *oq4, q5 = *oq5, q6 = *oq6, q7 = *oq7; 247 248 // 15-tap filter [1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1] 249 *op6 = ROUND_POWER_OF_TWO(p7 * 7 + p6 * 2 + p5 + p4 + p3 + p2 + p1 + p0 + 250 q0, 4); 251 *op5 = ROUND_POWER_OF_TWO(p7 * 6 + p6 + p5 * 2 + p4 + p3 + p2 + p1 + p0 + 252 q0 + q1, 4); 253 *op4 = ROUND_POWER_OF_TWO(p7 * 5 + p6 + p5 + p4 * 2 + p3 + p2 + p1 + p0 + 254 q0 + q1 + q2, 4); 255 *op3 = ROUND_POWER_OF_TWO(p7 * 4 + p6 + p5 + p4 + p3 * 2 + p2 + p1 + p0 + 256 q0 + q1 + q2 + q3, 4); 257 *op2 = ROUND_POWER_OF_TWO(p7 * 3 + p6 + p5 + p4 + p3 + p2 * 2 + p1 + p0 + 258 q0 + q1 + q2 + q3 + q4, 4); 259 *op1 = ROUND_POWER_OF_TWO(p7 * 2 + p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 + 260 q0 + q1 + q2 + q3 + q4 + q5, 4); 261 *op0 = ROUND_POWER_OF_TWO(p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 + 262 q0 + q1 + q2 + q3 + q4 + q5 + q6, 4); 263 *oq0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 + p0 + 264 q0 * 2 + q1 + q2 + q3 + q4 + q5 + q6 + q7, 4); 265 *oq1 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 + 266 q0 + q1 * 2 + q2 + q3 + q4 + q5 + q6 + q7 * 2, 4); 267 *oq2 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + 268 q0 + q1 + q2 * 2 + q3 + q4 + q5 + q6 + q7 * 3, 4); 269 *oq3 = ROUND_POWER_OF_TWO(p3 + p2 + p1 + p0 + 270 q0 + q1 + q2 + q3 * 2 + q4 + q5 + q6 + q7 * 4, 4); 271 *oq4 = ROUND_POWER_OF_TWO(p2 + p1 + p0 + 272 q0 + q1 + q2 + q3 + q4 * 2 + q5 + q6 + q7 * 5, 4); 273 *oq5 = ROUND_POWER_OF_TWO(p1 + p0 + 274 q0 + q1 + q2 + q3 + q4 + q5 * 2 + q6 + q7 * 6, 4); 275 *oq6 = ROUND_POWER_OF_TWO(p0 + 276 q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 + q7 * 7, 4); 277 } else { 278 filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3); 279 } 280} 281 282void vp9_lpf_horizontal_16_c(uint8_t *s, int p, const uint8_t *blimit, 283 const uint8_t *limit, const uint8_t *thresh, 284 int count) { 285 int i; 286 287 // loop filter designed to work using chars so that we can make maximum use 288 // of 8 bit simd instructions. 289 for (i = 0; i < 8 * count; ++i) { 290 const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p]; 291 const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p]; 292 const int8_t mask = filter_mask(*limit, *blimit, 293 p3, p2, p1, p0, q0, q1, q2, q3); 294 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); 295 const int8_t flat2 = flat_mask5(1, 296 s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, 297 q0, s[4 * p], s[5 * p], s[6 * p], s[7 * p]); 298 299 filter16(mask, *thresh, flat, flat2, 300 s - 8 * p, s - 7 * p, s - 6 * p, s - 5 * p, 301 s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, 302 s, s + 1 * p, s + 2 * p, s + 3 * p, 303 s + 4 * p, s + 5 * p, s + 6 * p, s + 7 * p); 304 ++s; 305 } 306} 307 308static void mb_lpf_vertical_edge_w(uint8_t *s, int p, 309 const uint8_t *blimit, 310 const uint8_t *limit, 311 const uint8_t *thresh, 312 int count) { 313 int i; 314 315 for (i = 0; i < count; ++i) { 316 const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1]; 317 const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3]; 318 const int8_t mask = filter_mask(*limit, *blimit, 319 p3, p2, p1, p0, q0, q1, q2, q3); 320 const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3); 321 const int8_t flat2 = flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0, 322 q0, s[4], s[5], s[6], s[7]); 323 324 filter16(mask, *thresh, flat, flat2, 325 s - 8, s - 7, s - 6, s - 5, s - 4, s - 3, s - 2, s - 1, 326 s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6, s + 7); 327 s += p; 328 } 329} 330 331void vp9_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit, 332 const uint8_t *limit, const uint8_t *thresh) { 333 mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8); 334} 335 336void vp9_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit, 337 const uint8_t *limit, const uint8_t *thresh) { 338 mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16); 339} 340