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 <immintrin.h> /* AVX2 */ 12 13 static void mb_lpf_horizontal_edge_w_avx2_8(unsigned char *s, int p, 14 const unsigned char *_blimit, const unsigned char *_limit, 15 const unsigned char *_thresh) { 16 __m128i mask, hev, flat, flat2; 17 const __m128i zero = _mm_set1_epi16(0); 18 const __m128i one = _mm_set1_epi8(1); 19 __m128i q7p7, q6p6, q5p5, q4p4, q3p3, q2p2, q1p1, q0p0, p0q0, p1q1; 20 __m128i abs_p1p0; 21 22 const __m128i thresh = _mm_broadcastb_epi8( 23 _mm_cvtsi32_si128((int) _thresh[0])); 24 const __m128i limit = _mm_broadcastb_epi8( 25 _mm_cvtsi32_si128((int) _limit[0])); 26 const __m128i blimit = _mm_broadcastb_epi8( 27 _mm_cvtsi32_si128((int) _blimit[0])); 28 29 q4p4 = _mm_loadl_epi64((__m128i *) (s - 5 * p)); 30 q4p4 = _mm_castps_si128( 31 _mm_loadh_pi(_mm_castsi128_ps(q4p4), (__m64 *) (s + 4 * p))); 32 q3p3 = _mm_loadl_epi64((__m128i *) (s - 4 * p)); 33 q3p3 = _mm_castps_si128( 34 _mm_loadh_pi(_mm_castsi128_ps(q3p3), (__m64 *) (s + 3 * p))); 35 q2p2 = _mm_loadl_epi64((__m128i *) (s - 3 * p)); 36 q2p2 = _mm_castps_si128( 37 _mm_loadh_pi(_mm_castsi128_ps(q2p2), (__m64 *) (s + 2 * p))); 38 q1p1 = _mm_loadl_epi64((__m128i *) (s - 2 * p)); 39 q1p1 = _mm_castps_si128( 40 _mm_loadh_pi(_mm_castsi128_ps(q1p1), (__m64 *) (s + 1 * p))); 41 p1q1 = _mm_shuffle_epi32(q1p1, 78); 42 q0p0 = _mm_loadl_epi64((__m128i *) (s - 1 * p)); 43 q0p0 = _mm_castps_si128( 44 _mm_loadh_pi(_mm_castsi128_ps(q0p0), (__m64 *) (s - 0 * p))); 45 p0q0 = _mm_shuffle_epi32(q0p0, 78); 46 47 { 48 __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work; 49 abs_p1p0 = _mm_or_si128(_mm_subs_epu8(q1p1, q0p0), 50 _mm_subs_epu8(q0p0, q1p1)); 51 abs_q1q0 = _mm_srli_si128(abs_p1p0, 8); 52 fe = _mm_set1_epi8(0xfe); 53 ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); 54 abs_p0q0 = _mm_or_si128(_mm_subs_epu8(q0p0, p0q0), 55 _mm_subs_epu8(p0q0, q0p0)); 56 abs_p1q1 = _mm_or_si128(_mm_subs_epu8(q1p1, p1q1), 57 _mm_subs_epu8(p1q1, q1p1)); 58 flat = _mm_max_epu8(abs_p1p0, abs_q1q0); 59 hev = _mm_subs_epu8(flat, thresh); 60 hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); 61 62 abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0); 63 abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); 64 mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); 65 mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); 66 // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; 67 mask = _mm_max_epu8(abs_p1p0, mask); 68 // mask |= (abs(p1 - p0) > limit) * -1; 69 // mask |= (abs(q1 - q0) > limit) * -1; 70 71 work = _mm_max_epu8( 72 _mm_or_si128(_mm_subs_epu8(q2p2, q1p1), 73 _mm_subs_epu8(q1p1, q2p2)), 74 _mm_or_si128(_mm_subs_epu8(q3p3, q2p2), 75 _mm_subs_epu8(q2p2, q3p3))); 76 mask = _mm_max_epu8(work, mask); 77 mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8)); 78 mask = _mm_subs_epu8(mask, limit); 79 mask = _mm_cmpeq_epi8(mask, zero); 80 } 81 82 // lp filter 83 { 84 const __m128i t4 = _mm_set1_epi8(4); 85 const __m128i t3 = _mm_set1_epi8(3); 86 const __m128i t80 = _mm_set1_epi8(0x80); 87 const __m128i t1 = _mm_set1_epi16(0x1); 88 __m128i qs1ps1 = _mm_xor_si128(q1p1, t80); 89 __m128i qs0ps0 = _mm_xor_si128(q0p0, t80); 90 __m128i qs0 = _mm_xor_si128(p0q0, t80); 91 __m128i qs1 = _mm_xor_si128(p1q1, t80); 92 __m128i filt; 93 __m128i work_a; 94 __m128i filter1, filter2; 95 __m128i flat2_q6p6, flat2_q5p5, flat2_q4p4, flat2_q3p3, flat2_q2p2; 96 __m128i flat2_q1p1, flat2_q0p0, flat_q2p2, flat_q1p1, flat_q0p0; 97 98 filt = _mm_and_si128(_mm_subs_epi8(qs1ps1, qs1), hev); 99 work_a = _mm_subs_epi8(qs0, qs0ps0); 100 filt = _mm_adds_epi8(filt, work_a); 101 filt = _mm_adds_epi8(filt, work_a); 102 filt = _mm_adds_epi8(filt, work_a); 103 /* (vp9_filter + 3 * (qs0 - ps0)) & mask */ 104 filt = _mm_and_si128(filt, mask); 105 106 filter1 = _mm_adds_epi8(filt, t4); 107 filter2 = _mm_adds_epi8(filt, t3); 108 109 filter1 = _mm_unpacklo_epi8(zero, filter1); 110 filter1 = _mm_srai_epi16(filter1, 0xB); 111 filter2 = _mm_unpacklo_epi8(zero, filter2); 112 filter2 = _mm_srai_epi16(filter2, 0xB); 113 114 /* Filter1 >> 3 */ 115 filt = _mm_packs_epi16(filter2, _mm_subs_epi16(zero, filter1)); 116 qs0ps0 = _mm_xor_si128(_mm_adds_epi8(qs0ps0, filt), t80); 117 118 /* filt >> 1 */ 119 filt = _mm_adds_epi16(filter1, t1); 120 filt = _mm_srai_epi16(filt, 1); 121 filt = _mm_andnot_si128( 122 _mm_srai_epi16(_mm_unpacklo_epi8(zero, hev), 0x8), filt); 123 filt = _mm_packs_epi16(filt, _mm_subs_epi16(zero, filt)); 124 qs1ps1 = _mm_xor_si128(_mm_adds_epi8(qs1ps1, filt), t80); 125 // loopfilter done 126 127 { 128 __m128i work; 129 flat = _mm_max_epu8( 130 _mm_or_si128(_mm_subs_epu8(q2p2, q0p0), 131 _mm_subs_epu8(q0p0, q2p2)), 132 _mm_or_si128(_mm_subs_epu8(q3p3, q0p0), 133 _mm_subs_epu8(q0p0, q3p3))); 134 flat = _mm_max_epu8(abs_p1p0, flat); 135 flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8)); 136 flat = _mm_subs_epu8(flat, one); 137 flat = _mm_cmpeq_epi8(flat, zero); 138 flat = _mm_and_si128(flat, mask); 139 140 q5p5 = _mm_loadl_epi64((__m128i *) (s - 6 * p)); 141 q5p5 = _mm_castps_si128( 142 _mm_loadh_pi(_mm_castsi128_ps(q5p5), 143 (__m64 *) (s + 5 * p))); 144 145 q6p6 = _mm_loadl_epi64((__m128i *) (s - 7 * p)); 146 q6p6 = _mm_castps_si128( 147 _mm_loadh_pi(_mm_castsi128_ps(q6p6), 148 (__m64 *) (s + 6 * p))); 149 150 flat2 = _mm_max_epu8( 151 _mm_or_si128(_mm_subs_epu8(q4p4, q0p0), 152 _mm_subs_epu8(q0p0, q4p4)), 153 _mm_or_si128(_mm_subs_epu8(q5p5, q0p0), 154 _mm_subs_epu8(q0p0, q5p5))); 155 156 q7p7 = _mm_loadl_epi64((__m128i *) (s - 8 * p)); 157 q7p7 = _mm_castps_si128( 158 _mm_loadh_pi(_mm_castsi128_ps(q7p7), 159 (__m64 *) (s + 7 * p))); 160 161 work = _mm_max_epu8( 162 _mm_or_si128(_mm_subs_epu8(q6p6, q0p0), 163 _mm_subs_epu8(q0p0, q6p6)), 164 _mm_or_si128(_mm_subs_epu8(q7p7, q0p0), 165 _mm_subs_epu8(q0p0, q7p7))); 166 167 flat2 = _mm_max_epu8(work, flat2); 168 flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8)); 169 flat2 = _mm_subs_epu8(flat2, one); 170 flat2 = _mm_cmpeq_epi8(flat2, zero); 171 flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask 172 } 173 174 // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 175 // flat and wide flat calculations 176 { 177 const __m128i eight = _mm_set1_epi16(8); 178 const __m128i four = _mm_set1_epi16(4); 179 __m128i p7_16, p6_16, p5_16, p4_16, p3_16, p2_16, p1_16, p0_16; 180 __m128i q7_16, q6_16, q5_16, q4_16, q3_16, q2_16, q1_16, q0_16; 181 __m128i pixelFilter_p, pixelFilter_q; 182 __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0; 183 __m128i sum_p7, sum_q7, sum_p3, sum_q3, res_p, res_q; 184 185 p7_16 = _mm_unpacklo_epi8(q7p7, zero); 186 p6_16 = _mm_unpacklo_epi8(q6p6, zero); 187 p5_16 = _mm_unpacklo_epi8(q5p5, zero); 188 p4_16 = _mm_unpacklo_epi8(q4p4, zero); 189 p3_16 = _mm_unpacklo_epi8(q3p3, zero); 190 p2_16 = _mm_unpacklo_epi8(q2p2, zero); 191 p1_16 = _mm_unpacklo_epi8(q1p1, zero); 192 p0_16 = _mm_unpacklo_epi8(q0p0, zero); 193 q0_16 = _mm_unpackhi_epi8(q0p0, zero); 194 q1_16 = _mm_unpackhi_epi8(q1p1, zero); 195 q2_16 = _mm_unpackhi_epi8(q2p2, zero); 196 q3_16 = _mm_unpackhi_epi8(q3p3, zero); 197 q4_16 = _mm_unpackhi_epi8(q4p4, zero); 198 q5_16 = _mm_unpackhi_epi8(q5p5, zero); 199 q6_16 = _mm_unpackhi_epi8(q6p6, zero); 200 q7_16 = _mm_unpackhi_epi8(q7p7, zero); 201 202 pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6_16, p5_16), 203 _mm_add_epi16(p4_16, p3_16)); 204 pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6_16, q5_16), 205 _mm_add_epi16(q4_16, q3_16)); 206 207 pixetFilter_p2p1p0 = _mm_add_epi16(p0_16, 208 _mm_add_epi16(p2_16, p1_16)); 209 pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0); 210 211 pixetFilter_q2q1q0 = _mm_add_epi16(q0_16, 212 _mm_add_epi16(q2_16, q1_16)); 213 pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0); 214 pixelFilter_p = _mm_add_epi16(eight, 215 _mm_add_epi16(pixelFilter_p, pixelFilter_q)); 216 pixetFilter_p2p1p0 = _mm_add_epi16(four, 217 _mm_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0)); 218 res_p = _mm_srli_epi16( 219 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(p7_16, p0_16)), 220 4); 221 res_q = _mm_srli_epi16( 222 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(q7_16, q0_16)), 223 4); 224 flat2_q0p0 = _mm_packus_epi16(res_p, res_q); 225 res_p = _mm_srli_epi16( 226 _mm_add_epi16(pixetFilter_p2p1p0, 227 _mm_add_epi16(p3_16, p0_16)), 3); 228 res_q = _mm_srli_epi16( 229 _mm_add_epi16(pixetFilter_p2p1p0, 230 _mm_add_epi16(q3_16, q0_16)), 3); 231 232 flat_q0p0 = _mm_packus_epi16(res_p, res_q); 233 234 sum_p7 = _mm_add_epi16(p7_16, p7_16); 235 sum_q7 = _mm_add_epi16(q7_16, q7_16); 236 sum_p3 = _mm_add_epi16(p3_16, p3_16); 237 sum_q3 = _mm_add_epi16(q3_16, q3_16); 238 239 pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6_16); 240 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6_16); 241 res_p = _mm_srli_epi16( 242 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p1_16)), 243 4); 244 res_q = _mm_srli_epi16( 245 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q1_16)), 246 4); 247 flat2_q1p1 = _mm_packus_epi16(res_p, res_q); 248 249 pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2_16); 250 pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2_16); 251 res_p = _mm_srli_epi16( 252 _mm_add_epi16(pixetFilter_p2p1p0, 253 _mm_add_epi16(sum_p3, p1_16)), 3); 254 res_q = _mm_srli_epi16( 255 _mm_add_epi16(pixetFilter_q2q1q0, 256 _mm_add_epi16(sum_q3, q1_16)), 3); 257 flat_q1p1 = _mm_packus_epi16(res_p, res_q); 258 259 sum_p7 = _mm_add_epi16(sum_p7, p7_16); 260 sum_q7 = _mm_add_epi16(sum_q7, q7_16); 261 sum_p3 = _mm_add_epi16(sum_p3, p3_16); 262 sum_q3 = _mm_add_epi16(sum_q3, q3_16); 263 264 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5_16); 265 pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5_16); 266 res_p = _mm_srli_epi16( 267 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p2_16)), 268 4); 269 res_q = _mm_srli_epi16( 270 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q2_16)), 271 4); 272 flat2_q2p2 = _mm_packus_epi16(res_p, res_q); 273 274 pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1_16); 275 pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1_16); 276 277 res_p = _mm_srli_epi16( 278 _mm_add_epi16(pixetFilter_p2p1p0, 279 _mm_add_epi16(sum_p3, p2_16)), 3); 280 res_q = _mm_srli_epi16( 281 _mm_add_epi16(pixetFilter_q2q1q0, 282 _mm_add_epi16(sum_q3, q2_16)), 3); 283 flat_q2p2 = _mm_packus_epi16(res_p, res_q); 284 285 sum_p7 = _mm_add_epi16(sum_p7, p7_16); 286 sum_q7 = _mm_add_epi16(sum_q7, q7_16); 287 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4_16); 288 pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4_16); 289 res_p = _mm_srli_epi16( 290 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p3_16)), 291 4); 292 res_q = _mm_srli_epi16( 293 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q3_16)), 294 4); 295 flat2_q3p3 = _mm_packus_epi16(res_p, res_q); 296 297 sum_p7 = _mm_add_epi16(sum_p7, p7_16); 298 sum_q7 = _mm_add_epi16(sum_q7, q7_16); 299 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3_16); 300 pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3_16); 301 res_p = _mm_srli_epi16( 302 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p4_16)), 303 4); 304 res_q = _mm_srli_epi16( 305 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q4_16)), 306 4); 307 flat2_q4p4 = _mm_packus_epi16(res_p, res_q); 308 309 sum_p7 = _mm_add_epi16(sum_p7, p7_16); 310 sum_q7 = _mm_add_epi16(sum_q7, q7_16); 311 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2_16); 312 pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2_16); 313 res_p = _mm_srli_epi16( 314 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p5_16)), 315 4); 316 res_q = _mm_srli_epi16( 317 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q5_16)), 318 4); 319 flat2_q5p5 = _mm_packus_epi16(res_p, res_q); 320 321 sum_p7 = _mm_add_epi16(sum_p7, p7_16); 322 sum_q7 = _mm_add_epi16(sum_q7, q7_16); 323 pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1_16); 324 pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1_16); 325 res_p = _mm_srli_epi16( 326 _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p6_16)), 327 4); 328 res_q = _mm_srli_epi16( 329 _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q6_16)), 330 4); 331 flat2_q6p6 = _mm_packus_epi16(res_p, res_q); 332 } 333 // wide flat 334 // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 335 336 flat = _mm_shuffle_epi32(flat, 68); 337 flat2 = _mm_shuffle_epi32(flat2, 68); 338 339 q2p2 = _mm_andnot_si128(flat, q2p2); 340 flat_q2p2 = _mm_and_si128(flat, flat_q2p2); 341 q2p2 = _mm_or_si128(q2p2, flat_q2p2); 342 343 qs1ps1 = _mm_andnot_si128(flat, qs1ps1); 344 flat_q1p1 = _mm_and_si128(flat, flat_q1p1); 345 q1p1 = _mm_or_si128(qs1ps1, flat_q1p1); 346 347 qs0ps0 = _mm_andnot_si128(flat, qs0ps0); 348 flat_q0p0 = _mm_and_si128(flat, flat_q0p0); 349 q0p0 = _mm_or_si128(qs0ps0, flat_q0p0); 350 351 q6p6 = _mm_andnot_si128(flat2, q6p6); 352 flat2_q6p6 = _mm_and_si128(flat2, flat2_q6p6); 353 q6p6 = _mm_or_si128(q6p6, flat2_q6p6); 354 _mm_storel_epi64((__m128i *) (s - 7 * p), q6p6); 355 _mm_storeh_pi((__m64 *) (s + 6 * p), _mm_castsi128_ps(q6p6)); 356 357 q5p5 = _mm_andnot_si128(flat2, q5p5); 358 flat2_q5p5 = _mm_and_si128(flat2, flat2_q5p5); 359 q5p5 = _mm_or_si128(q5p5, flat2_q5p5); 360 _mm_storel_epi64((__m128i *) (s - 6 * p), q5p5); 361 _mm_storeh_pi((__m64 *) (s + 5 * p), _mm_castsi128_ps(q5p5)); 362 363 q4p4 = _mm_andnot_si128(flat2, q4p4); 364 flat2_q4p4 = _mm_and_si128(flat2, flat2_q4p4); 365 q4p4 = _mm_or_si128(q4p4, flat2_q4p4); 366 _mm_storel_epi64((__m128i *) (s - 5 * p), q4p4); 367 _mm_storeh_pi((__m64 *) (s + 4 * p), _mm_castsi128_ps(q4p4)); 368 369 q3p3 = _mm_andnot_si128(flat2, q3p3); 370 flat2_q3p3 = _mm_and_si128(flat2, flat2_q3p3); 371 q3p3 = _mm_or_si128(q3p3, flat2_q3p3); 372 _mm_storel_epi64((__m128i *) (s - 4 * p), q3p3); 373 _mm_storeh_pi((__m64 *) (s + 3 * p), _mm_castsi128_ps(q3p3)); 374 375 q2p2 = _mm_andnot_si128(flat2, q2p2); 376 flat2_q2p2 = _mm_and_si128(flat2, flat2_q2p2); 377 q2p2 = _mm_or_si128(q2p2, flat2_q2p2); 378 _mm_storel_epi64((__m128i *) (s - 3 * p), q2p2); 379 _mm_storeh_pi((__m64 *) (s + 2 * p), _mm_castsi128_ps(q2p2)); 380 381 q1p1 = _mm_andnot_si128(flat2, q1p1); 382 flat2_q1p1 = _mm_and_si128(flat2, flat2_q1p1); 383 q1p1 = _mm_or_si128(q1p1, flat2_q1p1); 384 _mm_storel_epi64((__m128i *) (s - 2 * p), q1p1); 385 _mm_storeh_pi((__m64 *) (s + 1 * p), _mm_castsi128_ps(q1p1)); 386 387 q0p0 = _mm_andnot_si128(flat2, q0p0); 388 flat2_q0p0 = _mm_and_si128(flat2, flat2_q0p0); 389 q0p0 = _mm_or_si128(q0p0, flat2_q0p0); 390 _mm_storel_epi64((__m128i *) (s - 1 * p), q0p0); 391 _mm_storeh_pi((__m64 *) (s - 0 * p), _mm_castsi128_ps(q0p0)); 392 } 393 } 394 395 static void mb_lpf_horizontal_edge_w_avx2_16(unsigned char *s, int p, 396 const unsigned char *_blimit, const unsigned char *_limit, 397 const unsigned char *_thresh) { 398 __m128i mask, hev, flat, flat2; 399 const __m128i zero = _mm_set1_epi16(0); 400 const __m128i one = _mm_set1_epi8(1); 401 __m128i p7, p6, p5; 402 __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4; 403 __m128i q5, q6, q7; 404 405 const __m128i thresh = _mm_broadcastb_epi8( 406 _mm_cvtsi32_si128((int) _thresh[0])); 407 const __m128i limit = _mm_broadcastb_epi8( 408 _mm_cvtsi32_si128((int) _limit[0])); 409 const __m128i blimit = _mm_broadcastb_epi8( 410 _mm_cvtsi32_si128((int) _blimit[0])); 411 412 p4 = _mm_loadu_si128((__m128i *) (s - 5 * p)); 413 p3 = _mm_loadu_si128((__m128i *) (s - 4 * p)); 414 p2 = _mm_loadu_si128((__m128i *) (s - 3 * p)); 415 p1 = _mm_loadu_si128((__m128i *) (s - 2 * p)); 416 p0 = _mm_loadu_si128((__m128i *) (s - 1 * p)); 417 q0 = _mm_loadu_si128((__m128i *) (s - 0 * p)); 418 q1 = _mm_loadu_si128((__m128i *) (s + 1 * p)); 419 q2 = _mm_loadu_si128((__m128i *) (s + 2 * p)); 420 q3 = _mm_loadu_si128((__m128i *) (s + 3 * p)); 421 q4 = _mm_loadu_si128((__m128i *) (s + 4 * p)); 422 423 { 424 const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0), 425 _mm_subs_epu8(p0, p1)); 426 const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0), 427 _mm_subs_epu8(q0, q1)); 428 const __m128i fe = _mm_set1_epi8(0xfe); 429 const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0); 430 __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0), 431 _mm_subs_epu8(q0, p0)); 432 __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1), 433 _mm_subs_epu8(q1, p1)); 434 __m128i work; 435 flat = _mm_max_epu8(abs_p1p0, abs_q1q0); 436 hev = _mm_subs_epu8(flat, thresh); 437 hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff); 438 439 abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0); 440 abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1); 441 mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit); 442 mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff); 443 // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1; 444 mask = _mm_max_epu8(flat, mask); 445 // mask |= (abs(p1 - p0) > limit) * -1; 446 // mask |= (abs(q1 - q0) > limit) * -1; 447 work = _mm_max_epu8( 448 _mm_or_si128(_mm_subs_epu8(p2, p1), _mm_subs_epu8(p1, p2)), 449 _mm_or_si128(_mm_subs_epu8(p3, p2), _mm_subs_epu8(p2, p3))); 450 mask = _mm_max_epu8(work, mask); 451 work = _mm_max_epu8( 452 _mm_or_si128(_mm_subs_epu8(q2, q1), _mm_subs_epu8(q1, q2)), 453 _mm_or_si128(_mm_subs_epu8(q3, q2), _mm_subs_epu8(q2, q3))); 454 mask = _mm_max_epu8(work, mask); 455 mask = _mm_subs_epu8(mask, limit); 456 mask = _mm_cmpeq_epi8(mask, zero); 457 } 458 459 // lp filter 460 { 461 const __m128i t4 = _mm_set1_epi8(4); 462 const __m128i t3 = _mm_set1_epi8(3); 463 const __m128i t80 = _mm_set1_epi8(0x80); 464 const __m128i te0 = _mm_set1_epi8(0xe0); 465 const __m128i t1f = _mm_set1_epi8(0x1f); 466 const __m128i t1 = _mm_set1_epi8(0x1); 467 const __m128i t7f = _mm_set1_epi8(0x7f); 468 469 __m128i ps1 = _mm_xor_si128(p1, t80); 470 __m128i ps0 = _mm_xor_si128(p0, t80); 471 __m128i qs0 = _mm_xor_si128(q0, t80); 472 __m128i qs1 = _mm_xor_si128(q1, t80); 473 __m128i filt; 474 __m128i work_a; 475 __m128i filter1, filter2; 476 __m128i flat2_p6, flat2_p5, flat2_p4, flat2_p3, flat2_p2, flat2_p1, 477 flat2_p0, flat2_q0, flat2_q1, flat2_q2, flat2_q3, flat2_q4, 478 flat2_q5, flat2_q6, flat_p2, flat_p1, flat_p0, flat_q0, flat_q1, 479 flat_q2; 480 481 filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev); 482 work_a = _mm_subs_epi8(qs0, ps0); 483 filt = _mm_adds_epi8(filt, work_a); 484 filt = _mm_adds_epi8(filt, work_a); 485 filt = _mm_adds_epi8(filt, work_a); 486 /* (vp9_filter + 3 * (qs0 - ps0)) & mask */ 487 filt = _mm_and_si128(filt, mask); 488 489 filter1 = _mm_adds_epi8(filt, t4); 490 filter2 = _mm_adds_epi8(filt, t3); 491 492 /* Filter1 >> 3 */ 493 work_a = _mm_cmpgt_epi8(zero, filter1); 494 filter1 = _mm_srli_epi16(filter1, 3); 495 work_a = _mm_and_si128(work_a, te0); 496 filter1 = _mm_and_si128(filter1, t1f); 497 filter1 = _mm_or_si128(filter1, work_a); 498 qs0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80); 499 500 /* Filter2 >> 3 */ 501 work_a = _mm_cmpgt_epi8(zero, filter2); 502 filter2 = _mm_srli_epi16(filter2, 3); 503 work_a = _mm_and_si128(work_a, te0); 504 filter2 = _mm_and_si128(filter2, t1f); 505 filter2 = _mm_or_si128(filter2, work_a); 506 ps0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80); 507 508 /* filt >> 1 */ 509 filt = _mm_adds_epi8(filter1, t1); 510 work_a = _mm_cmpgt_epi8(zero, filt); 511 filt = _mm_srli_epi16(filt, 1); 512 work_a = _mm_and_si128(work_a, t80); 513 filt = _mm_and_si128(filt, t7f); 514 filt = _mm_or_si128(filt, work_a); 515 filt = _mm_andnot_si128(hev, filt); 516 ps1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80); 517 qs1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80); 518 // loopfilter done 519 520 { 521 __m128i work; 522 work = _mm_max_epu8( 523 _mm_or_si128(_mm_subs_epu8(p2, p0), _mm_subs_epu8(p0, p2)), 524 _mm_or_si128(_mm_subs_epu8(q2, q0), _mm_subs_epu8(q0, q2))); 525 flat = _mm_max_epu8(work, flat); 526 work = _mm_max_epu8( 527 _mm_or_si128(_mm_subs_epu8(p3, p0), _mm_subs_epu8(p0, p3)), 528 _mm_or_si128(_mm_subs_epu8(q3, q0), _mm_subs_epu8(q0, q3))); 529 flat = _mm_max_epu8(work, flat); 530 work = _mm_max_epu8( 531 _mm_or_si128(_mm_subs_epu8(p4, p0), _mm_subs_epu8(p0, p4)), 532 _mm_or_si128(_mm_subs_epu8(q4, q0), _mm_subs_epu8(q0, q4))); 533 flat = _mm_subs_epu8(flat, one); 534 flat = _mm_cmpeq_epi8(flat, zero); 535 flat = _mm_and_si128(flat, mask); 536 537 p5 = _mm_loadu_si128((__m128i *) (s - 6 * p)); 538 q5 = _mm_loadu_si128((__m128i *) (s + 5 * p)); 539 flat2 = _mm_max_epu8( 540 _mm_or_si128(_mm_subs_epu8(p5, p0), _mm_subs_epu8(p0, p5)), 541 _mm_or_si128(_mm_subs_epu8(q5, q0), _mm_subs_epu8(q0, q5))); 542 543 flat2 = _mm_max_epu8(work, flat2); 544 p6 = _mm_loadu_si128((__m128i *) (s - 7 * p)); 545 q6 = _mm_loadu_si128((__m128i *) (s + 6 * p)); 546 work = _mm_max_epu8( 547 _mm_or_si128(_mm_subs_epu8(p6, p0), _mm_subs_epu8(p0, p6)), 548 _mm_or_si128(_mm_subs_epu8(q6, q0), _mm_subs_epu8(q0, q6))); 549 550 flat2 = _mm_max_epu8(work, flat2); 551 552 p7 = _mm_loadu_si128((__m128i *) (s - 8 * p)); 553 q7 = _mm_loadu_si128((__m128i *) (s + 7 * p)); 554 work = _mm_max_epu8( 555 _mm_or_si128(_mm_subs_epu8(p7, p0), _mm_subs_epu8(p0, p7)), 556 _mm_or_si128(_mm_subs_epu8(q7, q0), _mm_subs_epu8(q0, q7))); 557 558 flat2 = _mm_max_epu8(work, flat2); 559 flat2 = _mm_subs_epu8(flat2, one); 560 flat2 = _mm_cmpeq_epi8(flat2, zero); 561 flat2 = _mm_and_si128(flat2, flat); // flat2 & flat & mask 562 } 563 564 // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 565 // flat and wide flat calculations 566 { 567 const __m256i eight = _mm256_set1_epi16(8); 568 const __m256i four = _mm256_set1_epi16(4); 569 __m256i p256_7, q256_7, p256_6, q256_6, p256_5, q256_5, p256_4, 570 q256_4, p256_3, q256_3, p256_2, q256_2, p256_1, q256_1, 571 p256_0, q256_0; 572 __m256i pixelFilter_p, pixelFilter_q, pixetFilter_p2p1p0, 573 pixetFilter_q2q1q0, sum_p7, sum_q7, sum_p3, sum_q3, res_p, 574 res_q; 575 576 p256_7 = _mm256_cvtepu8_epi16(p7); 577 p256_6 = _mm256_cvtepu8_epi16(p6); 578 p256_5 = _mm256_cvtepu8_epi16(p5); 579 p256_4 = _mm256_cvtepu8_epi16(p4); 580 p256_3 = _mm256_cvtepu8_epi16(p3); 581 p256_2 = _mm256_cvtepu8_epi16(p2); 582 p256_1 = _mm256_cvtepu8_epi16(p1); 583 p256_0 = _mm256_cvtepu8_epi16(p0); 584 q256_0 = _mm256_cvtepu8_epi16(q0); 585 q256_1 = _mm256_cvtepu8_epi16(q1); 586 q256_2 = _mm256_cvtepu8_epi16(q2); 587 q256_3 = _mm256_cvtepu8_epi16(q3); 588 q256_4 = _mm256_cvtepu8_epi16(q4); 589 q256_5 = _mm256_cvtepu8_epi16(q5); 590 q256_6 = _mm256_cvtepu8_epi16(q6); 591 q256_7 = _mm256_cvtepu8_epi16(q7); 592 593 pixelFilter_p = _mm256_add_epi16(_mm256_add_epi16(p256_6, p256_5), 594 _mm256_add_epi16(p256_4, p256_3)); 595 pixelFilter_q = _mm256_add_epi16(_mm256_add_epi16(q256_6, q256_5), 596 _mm256_add_epi16(q256_4, q256_3)); 597 598 pixetFilter_p2p1p0 = _mm256_add_epi16(p256_0, 599 _mm256_add_epi16(p256_2, p256_1)); 600 pixelFilter_p = _mm256_add_epi16(pixelFilter_p, pixetFilter_p2p1p0); 601 602 pixetFilter_q2q1q0 = _mm256_add_epi16(q256_0, 603 _mm256_add_epi16(q256_2, q256_1)); 604 pixelFilter_q = _mm256_add_epi16(pixelFilter_q, pixetFilter_q2q1q0); 605 606 pixelFilter_p = _mm256_add_epi16(eight, 607 _mm256_add_epi16(pixelFilter_p, pixelFilter_q)); 608 609 pixetFilter_p2p1p0 = _mm256_add_epi16(four, 610 _mm256_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0)); 611 612 res_p = _mm256_srli_epi16( 613 _mm256_add_epi16(pixelFilter_p, 614 _mm256_add_epi16(p256_7, p256_0)), 4); 615 616 flat2_p0 = _mm256_castsi256_si128( 617 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 618 168)); 619 620 res_q = _mm256_srli_epi16( 621 _mm256_add_epi16(pixelFilter_p, 622 _mm256_add_epi16(q256_7, q256_0)), 4); 623 624 flat2_q0 = _mm256_castsi256_si128( 625 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 626 168)); 627 628 res_p = _mm256_srli_epi16( 629 _mm256_add_epi16(pixetFilter_p2p1p0, 630 _mm256_add_epi16(p256_3, p256_0)), 3); 631 632 flat_p0 = _mm256_castsi256_si128( 633 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 634 168)); 635 636 res_q = _mm256_srli_epi16( 637 _mm256_add_epi16(pixetFilter_p2p1p0, 638 _mm256_add_epi16(q256_3, q256_0)), 3); 639 640 flat_q0 = _mm256_castsi256_si128( 641 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 642 168)); 643 644 sum_p7 = _mm256_add_epi16(p256_7, p256_7); 645 646 sum_q7 = _mm256_add_epi16(q256_7, q256_7); 647 648 sum_p3 = _mm256_add_epi16(p256_3, p256_3); 649 650 sum_q3 = _mm256_add_epi16(q256_3, q256_3); 651 652 pixelFilter_q = _mm256_sub_epi16(pixelFilter_p, p256_6); 653 654 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_6); 655 656 res_p = _mm256_srli_epi16( 657 _mm256_add_epi16(pixelFilter_p, 658 _mm256_add_epi16(sum_p7, p256_1)), 4); 659 660 flat2_p1 = _mm256_castsi256_si128( 661 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 662 168)); 663 664 res_q = _mm256_srli_epi16( 665 _mm256_add_epi16(pixelFilter_q, 666 _mm256_add_epi16(sum_q7, q256_1)), 4); 667 668 flat2_q1 = _mm256_castsi256_si128( 669 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 670 168)); 671 672 pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_p2p1p0, p256_2); 673 674 pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_2); 675 676 res_p = _mm256_srli_epi16( 677 _mm256_add_epi16(pixetFilter_p2p1p0, 678 _mm256_add_epi16(sum_p3, p256_1)), 3); 679 680 flat_p1 = _mm256_castsi256_si128( 681 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 682 168)); 683 684 res_q = _mm256_srli_epi16( 685 _mm256_add_epi16(pixetFilter_q2q1q0, 686 _mm256_add_epi16(sum_q3, q256_1)), 3); 687 688 flat_q1 = _mm256_castsi256_si128( 689 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 690 168)); 691 692 sum_p7 = _mm256_add_epi16(sum_p7, p256_7); 693 694 sum_q7 = _mm256_add_epi16(sum_q7, q256_7); 695 696 sum_p3 = _mm256_add_epi16(sum_p3, p256_3); 697 698 sum_q3 = _mm256_add_epi16(sum_q3, q256_3); 699 700 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_5); 701 702 pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_5); 703 704 res_p = _mm256_srli_epi16( 705 _mm256_add_epi16(pixelFilter_p, 706 _mm256_add_epi16(sum_p7, p256_2)), 4); 707 708 flat2_p2 = _mm256_castsi256_si128( 709 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 710 168)); 711 712 res_q = _mm256_srli_epi16( 713 _mm256_add_epi16(pixelFilter_q, 714 _mm256_add_epi16(sum_q7, q256_2)), 4); 715 716 flat2_q2 = _mm256_castsi256_si128( 717 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 718 168)); 719 720 pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_1); 721 722 pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_q2q1q0, p256_1); 723 724 res_p = _mm256_srli_epi16( 725 _mm256_add_epi16(pixetFilter_p2p1p0, 726 _mm256_add_epi16(sum_p3, p256_2)), 3); 727 728 flat_p2 = _mm256_castsi256_si128( 729 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 730 168)); 731 732 res_q = _mm256_srli_epi16( 733 _mm256_add_epi16(pixetFilter_q2q1q0, 734 _mm256_add_epi16(sum_q3, q256_2)), 3); 735 736 flat_q2 = _mm256_castsi256_si128( 737 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 738 168)); 739 740 sum_p7 = _mm256_add_epi16(sum_p7, p256_7); 741 742 sum_q7 = _mm256_add_epi16(sum_q7, q256_7); 743 744 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_4); 745 746 pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_4); 747 748 res_p = _mm256_srli_epi16( 749 _mm256_add_epi16(pixelFilter_p, 750 _mm256_add_epi16(sum_p7, p256_3)), 4); 751 752 flat2_p3 = _mm256_castsi256_si128( 753 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 754 168)); 755 756 res_q = _mm256_srli_epi16( 757 _mm256_add_epi16(pixelFilter_q, 758 _mm256_add_epi16(sum_q7, q256_3)), 4); 759 760 flat2_q3 = _mm256_castsi256_si128( 761 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 762 168)); 763 764 sum_p7 = _mm256_add_epi16(sum_p7, p256_7); 765 766 sum_q7 = _mm256_add_epi16(sum_q7, q256_7); 767 768 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_3); 769 770 pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_3); 771 772 res_p = _mm256_srli_epi16( 773 _mm256_add_epi16(pixelFilter_p, 774 _mm256_add_epi16(sum_p7, p256_4)), 4); 775 776 flat2_p4 = _mm256_castsi256_si128( 777 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 778 168)); 779 780 res_q = _mm256_srli_epi16( 781 _mm256_add_epi16(pixelFilter_q, 782 _mm256_add_epi16(sum_q7, q256_4)), 4); 783 784 flat2_q4 = _mm256_castsi256_si128( 785 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 786 168)); 787 788 sum_p7 = _mm256_add_epi16(sum_p7, p256_7); 789 790 sum_q7 = _mm256_add_epi16(sum_q7, q256_7); 791 792 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_2); 793 794 pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_2); 795 796 res_p = _mm256_srli_epi16( 797 _mm256_add_epi16(pixelFilter_p, 798 _mm256_add_epi16(sum_p7, p256_5)), 4); 799 800 flat2_p5 = _mm256_castsi256_si128( 801 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 802 168)); 803 804 res_q = _mm256_srli_epi16( 805 _mm256_add_epi16(pixelFilter_q, 806 _mm256_add_epi16(sum_q7, q256_5)), 4); 807 808 flat2_q5 = _mm256_castsi256_si128( 809 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 810 168)); 811 812 sum_p7 = _mm256_add_epi16(sum_p7, p256_7); 813 814 sum_q7 = _mm256_add_epi16(sum_q7, q256_7); 815 816 pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_1); 817 818 pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_1); 819 820 res_p = _mm256_srli_epi16( 821 _mm256_add_epi16(pixelFilter_p, 822 _mm256_add_epi16(sum_p7, p256_6)), 4); 823 824 flat2_p6 = _mm256_castsi256_si128( 825 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p), 826 168)); 827 828 res_q = _mm256_srli_epi16( 829 _mm256_add_epi16(pixelFilter_q, 830 _mm256_add_epi16(sum_q7, q256_6)), 4); 831 832 flat2_q6 = _mm256_castsi256_si128( 833 _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q), 834 168)); 835 } 836 837 // wide flat 838 // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 839 840 p2 = _mm_andnot_si128(flat, p2); 841 flat_p2 = _mm_and_si128(flat, flat_p2); 842 p2 = _mm_or_si128(flat_p2, p2); 843 844 p1 = _mm_andnot_si128(flat, ps1); 845 flat_p1 = _mm_and_si128(flat, flat_p1); 846 p1 = _mm_or_si128(flat_p1, p1); 847 848 p0 = _mm_andnot_si128(flat, ps0); 849 flat_p0 = _mm_and_si128(flat, flat_p0); 850 p0 = _mm_or_si128(flat_p0, p0); 851 852 q0 = _mm_andnot_si128(flat, qs0); 853 flat_q0 = _mm_and_si128(flat, flat_q0); 854 q0 = _mm_or_si128(flat_q0, q0); 855 856 q1 = _mm_andnot_si128(flat, qs1); 857 flat_q1 = _mm_and_si128(flat, flat_q1); 858 q1 = _mm_or_si128(flat_q1, q1); 859 860 q2 = _mm_andnot_si128(flat, q2); 861 flat_q2 = _mm_and_si128(flat, flat_q2); 862 q2 = _mm_or_si128(flat_q2, q2); 863 864 p6 = _mm_andnot_si128(flat2, p6); 865 flat2_p6 = _mm_and_si128(flat2, flat2_p6); 866 p6 = _mm_or_si128(flat2_p6, p6); 867 _mm_storeu_si128((__m128i *) (s - 7 * p), p6); 868 869 p5 = _mm_andnot_si128(flat2, p5); 870 flat2_p5 = _mm_and_si128(flat2, flat2_p5); 871 p5 = _mm_or_si128(flat2_p5, p5); 872 _mm_storeu_si128((__m128i *) (s - 6 * p), p5); 873 874 p4 = _mm_andnot_si128(flat2, p4); 875 flat2_p4 = _mm_and_si128(flat2, flat2_p4); 876 p4 = _mm_or_si128(flat2_p4, p4); 877 _mm_storeu_si128((__m128i *) (s - 5 * p), p4); 878 879 p3 = _mm_andnot_si128(flat2, p3); 880 flat2_p3 = _mm_and_si128(flat2, flat2_p3); 881 p3 = _mm_or_si128(flat2_p3, p3); 882 _mm_storeu_si128((__m128i *) (s - 4 * p), p3); 883 884 p2 = _mm_andnot_si128(flat2, p2); 885 flat2_p2 = _mm_and_si128(flat2, flat2_p2); 886 p2 = _mm_or_si128(flat2_p2, p2); 887 _mm_storeu_si128((__m128i *) (s - 3 * p), p2); 888 889 p1 = _mm_andnot_si128(flat2, p1); 890 flat2_p1 = _mm_and_si128(flat2, flat2_p1); 891 p1 = _mm_or_si128(flat2_p1, p1); 892 _mm_storeu_si128((__m128i *) (s - 2 * p), p1); 893 894 p0 = _mm_andnot_si128(flat2, p0); 895 flat2_p0 = _mm_and_si128(flat2, flat2_p0); 896 p0 = _mm_or_si128(flat2_p0, p0); 897 _mm_storeu_si128((__m128i *) (s - 1 * p), p0); 898 899 q0 = _mm_andnot_si128(flat2, q0); 900 flat2_q0 = _mm_and_si128(flat2, flat2_q0); 901 q0 = _mm_or_si128(flat2_q0, q0); 902 _mm_storeu_si128((__m128i *) (s - 0 * p), q0); 903 904 q1 = _mm_andnot_si128(flat2, q1); 905 flat2_q1 = _mm_and_si128(flat2, flat2_q1); 906 q1 = _mm_or_si128(flat2_q1, q1); 907 _mm_storeu_si128((__m128i *) (s + 1 * p), q1); 908 909 q2 = _mm_andnot_si128(flat2, q2); 910 flat2_q2 = _mm_and_si128(flat2, flat2_q2); 911 q2 = _mm_or_si128(flat2_q2, q2); 912 _mm_storeu_si128((__m128i *) (s + 2 * p), q2); 913 914 q3 = _mm_andnot_si128(flat2, q3); 915 flat2_q3 = _mm_and_si128(flat2, flat2_q3); 916 q3 = _mm_or_si128(flat2_q3, q3); 917 _mm_storeu_si128((__m128i *) (s + 3 * p), q3); 918 919 q4 = _mm_andnot_si128(flat2, q4); 920 flat2_q4 = _mm_and_si128(flat2, flat2_q4); 921 q4 = _mm_or_si128(flat2_q4, q4); 922 _mm_storeu_si128((__m128i *) (s + 4 * p), q4); 923 924 q5 = _mm_andnot_si128(flat2, q5); 925 flat2_q5 = _mm_and_si128(flat2, flat2_q5); 926 q5 = _mm_or_si128(flat2_q5, q5); 927 _mm_storeu_si128((__m128i *) (s + 5 * p), q5); 928 929 q6 = _mm_andnot_si128(flat2, q6); 930 flat2_q6 = _mm_and_si128(flat2, flat2_q6); 931 q6 = _mm_or_si128(flat2_q6, q6); 932 _mm_storeu_si128((__m128i *) (s + 6 * p), q6); 933 } 934 } 935 936 void vp9_lpf_horizontal_16_avx2(unsigned char *s, int p, 937 const unsigned char *_blimit, const unsigned char *_limit, 938 const unsigned char *_thresh, int count) { 939 if (count == 1) 940 mb_lpf_horizontal_edge_w_avx2_8(s, p, _blimit, _limit, _thresh); 941 else 942 mb_lpf_horizontal_edge_w_avx2_16(s, p, _blimit, _limit, _thresh); 943 } 944
