| OLD | NEW |
|---|
| (Empty) | |
| 1 /* |
| 2 * Copyright (c) 2014 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 <arm_neon.h> |
| 12 #include "./vpx_config.h" |
| 13 |
| 14 static INLINE void vp8_loop_filter_neon( |
| 15 uint8x16_t qblimit, // flimit |
| 16 uint8x16_t qlimit, // limit |
| 17 uint8x16_t qthresh, // thresh |
| 18 uint8x16_t q3, // p3 |
| 19 uint8x16_t q4, // p2 |
| 20 uint8x16_t q5, // p1 |
| 21 uint8x16_t q6, // p0 |
| 22 uint8x16_t q7, // q0 |
| 23 uint8x16_t q8, // q1 |
| 24 uint8x16_t q9, // q2 |
| 25 uint8x16_t q10, // q3 |
| 26 uint8x16_t *q5r, // p1 |
| 27 uint8x16_t *q6r, // p0 |
| 28 uint8x16_t *q7r, // q0 |
| 29 uint8x16_t *q8r) { // q1 |
| 30 uint8x16_t q0u8, q1u8, q2u8, q11u8, q12u8, q13u8, q14u8, q15u8; |
| 31 int16x8_t q2s16, q11s16; |
| 32 uint16x8_t q4u16; |
| 33 int8x16_t q1s8, q2s8, q10s8, q11s8, q12s8, q13s8; |
| 34 int8x8_t d2s8, d3s8; |
| 35 |
| 36 q11u8 = vabdq_u8(q3, q4); |
| 37 q12u8 = vabdq_u8(q4, q5); |
| 38 q13u8 = vabdq_u8(q5, q6); |
| 39 q14u8 = vabdq_u8(q8, q7); |
| 40 q3 = vabdq_u8(q9, q8); |
| 41 q4 = vabdq_u8(q10, q9); |
| 42 |
| 43 q11u8 = vmaxq_u8(q11u8, q12u8); |
| 44 q12u8 = vmaxq_u8(q13u8, q14u8); |
| 45 q3 = vmaxq_u8(q3, q4); |
| 46 q15u8 = vmaxq_u8(q11u8, q12u8); |
| 47 |
| 48 q9 = vabdq_u8(q6, q7); |
| 49 |
| 50 // vp8_hevmask |
| 51 q13u8 = vcgtq_u8(q13u8, qthresh); |
| 52 q14u8 = vcgtq_u8(q14u8, qthresh); |
| 53 q15u8 = vmaxq_u8(q15u8, q3); |
| 54 |
| 55 q2u8 = vabdq_u8(q5, q8); |
| 56 q9 = vqaddq_u8(q9, q9); |
| 57 |
| 58 q15u8 = vcgeq_u8(qlimit, q15u8); |
| 59 |
| 60 // vp8_filter() function |
| 61 // convert to signed |
| 62 q10 = vdupq_n_u8(0x80); |
| 63 q8 = veorq_u8(q8, q10); |
| 64 q7 = veorq_u8(q7, q10); |
| 65 q6 = veorq_u8(q6, q10); |
| 66 q5 = veorq_u8(q5, q10); |
| 67 |
| 68 q2u8 = vshrq_n_u8(q2u8, 1); |
| 69 q9 = vqaddq_u8(q9, q2u8); |
| 70 |
| 71 q10 = vdupq_n_u8(3); |
| 72 |
| 73 q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), |
| 74 vget_low_s8(vreinterpretq_s8_u8(q6))); |
| 75 q11s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), |
| 76 vget_high_s8(vreinterpretq_s8_u8(q6))); |
| 77 |
| 78 q9 = vcgeq_u8(qblimit, q9); |
| 79 |
| 80 q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), |
| 81 vreinterpretq_s8_u8(q8)); |
| 82 |
| 83 q14u8 = vorrq_u8(q13u8, q14u8); |
| 84 |
| 85 q4u16 = vmovl_u8(vget_low_u8(q10)); |
| 86 q2s16 = vmulq_s16(q2s16, vreinterpretq_s16_u16(q4u16)); |
| 87 q11s16 = vmulq_s16(q11s16, vreinterpretq_s16_u16(q4u16)); |
| 88 |
| 89 q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q14u8); |
| 90 q15u8 = vandq_u8(q15u8, q9); |
| 91 |
| 92 q1s8 = vreinterpretq_s8_u8(q1u8); |
| 93 q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); |
| 94 q11s16 = vaddw_s8(q11s16, vget_high_s8(q1s8)); |
| 95 |
| 96 q9 = vdupq_n_u8(4); |
| 97 // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) |
| 98 d2s8 = vqmovn_s16(q2s16); |
| 99 d3s8 = vqmovn_s16(q11s16); |
| 100 q1s8 = vcombine_s8(d2s8, d3s8); |
| 101 q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q15u8); |
| 102 q1s8 = vreinterpretq_s8_u8(q1u8); |
| 103 |
| 104 q2s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q10)); |
| 105 q1s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q9)); |
| 106 q2s8 = vshrq_n_s8(q2s8, 3); |
| 107 q1s8 = vshrq_n_s8(q1s8, 3); |
| 108 |
| 109 q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q2s8); |
| 110 q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q1s8); |
| 111 |
| 112 q1s8 = vrshrq_n_s8(q1s8, 1); |
| 113 q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); |
| 114 |
| 115 q13s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q1s8); |
| 116 q12s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q1s8); |
| 117 |
| 118 q0u8 = vdupq_n_u8(0x80); |
| 119 *q8r = veorq_u8(vreinterpretq_u8_s8(q12s8), q0u8); |
| 120 *q7r = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8); |
| 121 *q6r = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8); |
| 122 *q5r = veorq_u8(vreinterpretq_u8_s8(q13s8), q0u8); |
| 123 return; |
| 124 } |
| 125 |
| 126 void vp8_loop_filter_horizontal_edge_y_neon( |
| 127 unsigned char *src, |
| 128 int pitch, |
| 129 unsigned char blimit, |
| 130 unsigned char limit, |
| 131 unsigned char thresh) { |
| 132 uint8x16_t qblimit, qlimit, qthresh, q3, q4; |
| 133 uint8x16_t q5, q6, q7, q8, q9, q10; |
| 134 |
| 135 qblimit = vdupq_n_u8(blimit); |
| 136 qlimit = vdupq_n_u8(limit); |
| 137 qthresh = vdupq_n_u8(thresh); |
| 138 src -= (pitch << 2); |
| 139 |
| 140 q3 = vld1q_u8(src); |
| 141 src += pitch; |
| 142 q4 = vld1q_u8(src); |
| 143 src += pitch; |
| 144 q5 = vld1q_u8(src); |
| 145 src += pitch; |
| 146 q6 = vld1q_u8(src); |
| 147 src += pitch; |
| 148 q7 = vld1q_u8(src); |
| 149 src += pitch; |
| 150 q8 = vld1q_u8(src); |
| 151 src += pitch; |
| 152 q9 = vld1q_u8(src); |
| 153 src += pitch; |
| 154 q10 = vld1q_u8(src); |
| 155 |
| 156 vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, |
| 157 q5, q6, q7, q8, q9, q10, |
| 158 &q5, &q6, &q7, &q8); |
| 159 |
| 160 src -= (pitch * 5); |
| 161 vst1q_u8(src, q5); |
| 162 src += pitch; |
| 163 vst1q_u8(src, q6); |
| 164 src += pitch; |
| 165 vst1q_u8(src, q7); |
| 166 src += pitch; |
| 167 vst1q_u8(src, q8); |
| 168 return; |
| 169 } |
| 170 |
| 171 void vp8_loop_filter_horizontal_edge_uv_neon( |
| 172 unsigned char *u, |
| 173 int pitch, |
| 174 unsigned char blimit, |
| 175 unsigned char limit, |
| 176 unsigned char thresh, |
| 177 unsigned char *v) { |
| 178 uint8x16_t qblimit, qlimit, qthresh, q3, q4; |
| 179 uint8x16_t q5, q6, q7, q8, q9, q10; |
| 180 uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; |
| 181 uint8x8_t d15, d16, d17, d18, d19, d20, d21; |
| 182 |
| 183 qblimit = vdupq_n_u8(blimit); |
| 184 qlimit = vdupq_n_u8(limit); |
| 185 qthresh = vdupq_n_u8(thresh); |
| 186 |
| 187 u -= (pitch << 2); |
| 188 v -= (pitch << 2); |
| 189 |
| 190 d6 = vld1_u8(u); |
| 191 u += pitch; |
| 192 d7 = vld1_u8(v); |
| 193 v += pitch; |
| 194 d8 = vld1_u8(u); |
| 195 u += pitch; |
| 196 d9 = vld1_u8(v); |
| 197 v += pitch; |
| 198 d10 = vld1_u8(u); |
| 199 u += pitch; |
| 200 d11 = vld1_u8(v); |
| 201 v += pitch; |
| 202 d12 = vld1_u8(u); |
| 203 u += pitch; |
| 204 d13 = vld1_u8(v); |
| 205 v += pitch; |
| 206 d14 = vld1_u8(u); |
| 207 u += pitch; |
| 208 d15 = vld1_u8(v); |
| 209 v += pitch; |
| 210 d16 = vld1_u8(u); |
| 211 u += pitch; |
| 212 d17 = vld1_u8(v); |
| 213 v += pitch; |
| 214 d18 = vld1_u8(u); |
| 215 u += pitch; |
| 216 d19 = vld1_u8(v); |
| 217 v += pitch; |
| 218 d20 = vld1_u8(u); |
| 219 d21 = vld1_u8(v); |
| 220 |
| 221 q3 = vcombine_u8(d6, d7); |
| 222 q4 = vcombine_u8(d8, d9); |
| 223 q5 = vcombine_u8(d10, d11); |
| 224 q6 = vcombine_u8(d12, d13); |
| 225 q7 = vcombine_u8(d14, d15); |
| 226 q8 = vcombine_u8(d16, d17); |
| 227 q9 = vcombine_u8(d18, d19); |
| 228 q10 = vcombine_u8(d20, d21); |
| 229 |
| 230 vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, |
| 231 q5, q6, q7, q8, q9, q10, |
| 232 &q5, &q6, &q7, &q8); |
| 233 |
| 234 u -= (pitch * 5); |
| 235 vst1_u8(u, vget_low_u8(q5)); |
| 236 u += pitch; |
| 237 vst1_u8(u, vget_low_u8(q6)); |
| 238 u += pitch; |
| 239 vst1_u8(u, vget_low_u8(q7)); |
| 240 u += pitch; |
| 241 vst1_u8(u, vget_low_u8(q8)); |
| 242 |
| 243 v -= (pitch * 5); |
| 244 vst1_u8(v, vget_high_u8(q5)); |
| 245 v += pitch; |
| 246 vst1_u8(v, vget_high_u8(q6)); |
| 247 v += pitch; |
| 248 vst1_u8(v, vget_high_u8(q7)); |
| 249 v += pitch; |
| 250 vst1_u8(v, vget_high_u8(q8)); |
| 251 return; |
| 252 } |
| 253 |
| 254 static INLINE void write_4x8(unsigned char *dst, int pitch, |
| 255 const uint8x8x4_t result) { |
| 256 #if (__GNUC__ == 4 && (__GNUC_MINOR__ >= 7)) |
| 257 vst4_lane_u8(dst, result, 0); |
| 258 dst += pitch; |
| 259 vst4_lane_u8(dst, result, 1); |
| 260 dst += pitch; |
| 261 vst4_lane_u8(dst, result, 2); |
| 262 dst += pitch; |
| 263 vst4_lane_u8(dst, result, 3); |
| 264 dst += pitch; |
| 265 vst4_lane_u8(dst, result, 4); |
| 266 dst += pitch; |
| 267 vst4_lane_u8(dst, result, 5); |
| 268 dst += pitch; |
| 269 vst4_lane_u8(dst, result, 6); |
| 270 dst += pitch; |
| 271 vst4_lane_u8(dst, result, 7); |
| 272 #else |
| 273 /* |
| 274 * uint8x8x4_t result |
| 275 00 01 02 03 | 04 05 06 07 |
| 276 10 11 12 13 | 14 15 16 17 |
| 277 20 21 22 23 | 24 25 26 27 |
| 278 30 31 32 33 | 34 35 36 37 |
| 279 --- |
| 280 * after vtrn_u16 |
| 281 00 01 20 21 | 04 05 24 25 |
| 282 02 03 22 23 | 06 07 26 27 |
| 283 10 11 30 31 | 14 15 34 35 |
| 284 12 13 32 33 | 16 17 36 37 |
| 285 --- |
| 286 * after vtrn_u8 |
| 287 00 10 20 30 | 04 14 24 34 |
| 288 01 11 21 31 | 05 15 25 35 |
| 289 02 12 22 32 | 06 16 26 36 |
| 290 03 13 23 33 | 07 17 27 37 |
| 291 */ |
| 292 const uint16x4x2_t r02_u16 = vtrn_u16(vreinterpret_u16_u8(result.val[0]), |
| 293 vreinterpret_u16_u8(result.val[2])); |
| 294 const uint16x4x2_t r13_u16 = vtrn_u16(vreinterpret_u16_u8(result.val[1]), |
| 295 vreinterpret_u16_u8(result.val[3])); |
| 296 const uint8x8x2_t r01_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[0]), |
| 297 vreinterpret_u8_u16(r13_u16.val[0])); |
| 298 const uint8x8x2_t r23_u8 = vtrn_u8(vreinterpret_u8_u16(r02_u16.val[1]), |
| 299 vreinterpret_u8_u16(r13_u16.val[1])); |
| 300 const uint32x2_t x_0_4 = vreinterpret_u32_u8(r01_u8.val[0]); |
| 301 const uint32x2_t x_1_5 = vreinterpret_u32_u8(r01_u8.val[1]); |
| 302 const uint32x2_t x_2_6 = vreinterpret_u32_u8(r23_u8.val[0]); |
| 303 const uint32x2_t x_3_7 = vreinterpret_u32_u8(r23_u8.val[1]); |
| 304 vst1_lane_u32((uint32_t *)dst, x_0_4, 0); |
| 305 dst += pitch; |
| 306 vst1_lane_u32((uint32_t *)dst, x_1_5, 0); |
| 307 dst += pitch; |
| 308 vst1_lane_u32((uint32_t *)dst, x_2_6, 0); |
| 309 dst += pitch; |
| 310 vst1_lane_u32((uint32_t *)dst, x_3_7, 0); |
| 311 dst += pitch; |
| 312 vst1_lane_u32((uint32_t *)dst, x_0_4, 1); |
| 313 dst += pitch; |
| 314 vst1_lane_u32((uint32_t *)dst, x_1_5, 1); |
| 315 dst += pitch; |
| 316 vst1_lane_u32((uint32_t *)dst, x_2_6, 1); |
| 317 dst += pitch; |
| 318 vst1_lane_u32((uint32_t *)dst, x_3_7, 1); |
| 319 #endif |
| 320 } |
| 321 |
| 322 void vp8_loop_filter_vertical_edge_y_neon( |
| 323 unsigned char *src, |
| 324 int pitch, |
| 325 unsigned char blimit, |
| 326 unsigned char limit, |
| 327 unsigned char thresh) { |
| 328 unsigned char *s, *d; |
| 329 uint8x16_t qblimit, qlimit, qthresh, q3, q4; |
| 330 uint8x16_t q5, q6, q7, q8, q9, q10; |
| 331 uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; |
| 332 uint8x8_t d15, d16, d17, d18, d19, d20, d21; |
| 333 uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; |
| 334 uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; |
| 335 uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; |
| 336 uint8x8x4_t q4ResultH, q4ResultL; |
| 337 |
| 338 qblimit = vdupq_n_u8(blimit); |
| 339 qlimit = vdupq_n_u8(limit); |
| 340 qthresh = vdupq_n_u8(thresh); |
| 341 |
| 342 s = src - 4; |
| 343 d6 = vld1_u8(s); |
| 344 s += pitch; |
| 345 d8 = vld1_u8(s); |
| 346 s += pitch; |
| 347 d10 = vld1_u8(s); |
| 348 s += pitch; |
| 349 d12 = vld1_u8(s); |
| 350 s += pitch; |
| 351 d14 = vld1_u8(s); |
| 352 s += pitch; |
| 353 d16 = vld1_u8(s); |
| 354 s += pitch; |
| 355 d18 = vld1_u8(s); |
| 356 s += pitch; |
| 357 d20 = vld1_u8(s); |
| 358 s += pitch; |
| 359 d7 = vld1_u8(s); |
| 360 s += pitch; |
| 361 d9 = vld1_u8(s); |
| 362 s += pitch; |
| 363 d11 = vld1_u8(s); |
| 364 s += pitch; |
| 365 d13 = vld1_u8(s); |
| 366 s += pitch; |
| 367 d15 = vld1_u8(s); |
| 368 s += pitch; |
| 369 d17 = vld1_u8(s); |
| 370 s += pitch; |
| 371 d19 = vld1_u8(s); |
| 372 s += pitch; |
| 373 d21 = vld1_u8(s); |
| 374 |
| 375 q3 = vcombine_u8(d6, d7); |
| 376 q4 = vcombine_u8(d8, d9); |
| 377 q5 = vcombine_u8(d10, d11); |
| 378 q6 = vcombine_u8(d12, d13); |
| 379 q7 = vcombine_u8(d14, d15); |
| 380 q8 = vcombine_u8(d16, d17); |
| 381 q9 = vcombine_u8(d18, d19); |
| 382 q10 = vcombine_u8(d20, d21); |
| 383 |
| 384 q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); |
| 385 q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); |
| 386 q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); |
| 387 q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); |
| 388 |
| 389 q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), |
| 390 vreinterpretq_u16_u32(q2tmp2.val[0])); |
| 391 q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), |
| 392 vreinterpretq_u16_u32(q2tmp3.val[0])); |
| 393 q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), |
| 394 vreinterpretq_u16_u32(q2tmp2.val[1])); |
| 395 q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), |
| 396 vreinterpretq_u16_u32(q2tmp3.val[1])); |
| 397 |
| 398 q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), |
| 399 vreinterpretq_u8_u16(q2tmp5.val[0])); |
| 400 q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), |
| 401 vreinterpretq_u8_u16(q2tmp5.val[1])); |
| 402 q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), |
| 403 vreinterpretq_u8_u16(q2tmp7.val[0])); |
| 404 q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), |
| 405 vreinterpretq_u8_u16(q2tmp7.val[1])); |
| 406 |
| 407 q3 = q2tmp8.val[0]; |
| 408 q4 = q2tmp8.val[1]; |
| 409 q5 = q2tmp9.val[0]; |
| 410 q6 = q2tmp9.val[1]; |
| 411 q7 = q2tmp10.val[0]; |
| 412 q8 = q2tmp10.val[1]; |
| 413 q9 = q2tmp11.val[0]; |
| 414 q10 = q2tmp11.val[1]; |
| 415 |
| 416 vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, |
| 417 q5, q6, q7, q8, q9, q10, |
| 418 &q5, &q6, &q7, &q8); |
| 419 |
| 420 q4ResultL.val[0] = vget_low_u8(q5); // d10 |
| 421 q4ResultL.val[1] = vget_low_u8(q6); // d12 |
| 422 q4ResultL.val[2] = vget_low_u8(q7); // d14 |
| 423 q4ResultL.val[3] = vget_low_u8(q8); // d16 |
| 424 q4ResultH.val[0] = vget_high_u8(q5); // d11 |
| 425 q4ResultH.val[1] = vget_high_u8(q6); // d13 |
| 426 q4ResultH.val[2] = vget_high_u8(q7); // d15 |
| 427 q4ResultH.val[3] = vget_high_u8(q8); // d17 |
| 428 |
| 429 d = src - 2; |
| 430 write_4x8(d, pitch, q4ResultL); |
| 431 d += pitch * 8; |
| 432 write_4x8(d, pitch, q4ResultH); |
| 433 } |
| 434 |
| 435 void vp8_loop_filter_vertical_edge_uv_neon( |
| 436 unsigned char *u, |
| 437 int pitch, |
| 438 unsigned char blimit, |
| 439 unsigned char limit, |
| 440 unsigned char thresh, |
| 441 unsigned char *v) { |
| 442 unsigned char *us, *ud; |
| 443 unsigned char *vs, *vd; |
| 444 uint8x16_t qblimit, qlimit, qthresh, q3, q4; |
| 445 uint8x16_t q5, q6, q7, q8, q9, q10; |
| 446 uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14; |
| 447 uint8x8_t d15, d16, d17, d18, d19, d20, d21; |
| 448 uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3; |
| 449 uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7; |
| 450 uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11; |
| 451 uint8x8x4_t q4ResultH, q4ResultL; |
| 452 |
| 453 qblimit = vdupq_n_u8(blimit); |
| 454 qlimit = vdupq_n_u8(limit); |
| 455 qthresh = vdupq_n_u8(thresh); |
| 456 |
| 457 us = u - 4; |
| 458 d6 = vld1_u8(us); |
| 459 us += pitch; |
| 460 d8 = vld1_u8(us); |
| 461 us += pitch; |
| 462 d10 = vld1_u8(us); |
| 463 us += pitch; |
| 464 d12 = vld1_u8(us); |
| 465 us += pitch; |
| 466 d14 = vld1_u8(us); |
| 467 us += pitch; |
| 468 d16 = vld1_u8(us); |
| 469 us += pitch; |
| 470 d18 = vld1_u8(us); |
| 471 us += pitch; |
| 472 d20 = vld1_u8(us); |
| 473 |
| 474 vs = v - 4; |
| 475 d7 = vld1_u8(vs); |
| 476 vs += pitch; |
| 477 d9 = vld1_u8(vs); |
| 478 vs += pitch; |
| 479 d11 = vld1_u8(vs); |
| 480 vs += pitch; |
| 481 d13 = vld1_u8(vs); |
| 482 vs += pitch; |
| 483 d15 = vld1_u8(vs); |
| 484 vs += pitch; |
| 485 d17 = vld1_u8(vs); |
| 486 vs += pitch; |
| 487 d19 = vld1_u8(vs); |
| 488 vs += pitch; |
| 489 d21 = vld1_u8(vs); |
| 490 |
| 491 q3 = vcombine_u8(d6, d7); |
| 492 q4 = vcombine_u8(d8, d9); |
| 493 q5 = vcombine_u8(d10, d11); |
| 494 q6 = vcombine_u8(d12, d13); |
| 495 q7 = vcombine_u8(d14, d15); |
| 496 q8 = vcombine_u8(d16, d17); |
| 497 q9 = vcombine_u8(d18, d19); |
| 498 q10 = vcombine_u8(d20, d21); |
| 499 |
| 500 q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7)); |
| 501 q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8)); |
| 502 q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9)); |
| 503 q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10)); |
| 504 |
| 505 q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]), |
| 506 vreinterpretq_u16_u32(q2tmp2.val[0])); |
| 507 q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]), |
| 508 vreinterpretq_u16_u32(q2tmp3.val[0])); |
| 509 q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]), |
| 510 vreinterpretq_u16_u32(q2tmp2.val[1])); |
| 511 q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]), |
| 512 vreinterpretq_u16_u32(q2tmp3.val[1])); |
| 513 |
| 514 q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]), |
| 515 vreinterpretq_u8_u16(q2tmp5.val[0])); |
| 516 q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]), |
| 517 vreinterpretq_u8_u16(q2tmp5.val[1])); |
| 518 q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]), |
| 519 vreinterpretq_u8_u16(q2tmp7.val[0])); |
| 520 q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]), |
| 521 vreinterpretq_u8_u16(q2tmp7.val[1])); |
| 522 |
| 523 q3 = q2tmp8.val[0]; |
| 524 q4 = q2tmp8.val[1]; |
| 525 q5 = q2tmp9.val[0]; |
| 526 q6 = q2tmp9.val[1]; |
| 527 q7 = q2tmp10.val[0]; |
| 528 q8 = q2tmp10.val[1]; |
| 529 q9 = q2tmp11.val[0]; |
| 530 q10 = q2tmp11.val[1]; |
| 531 |
| 532 vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4, |
| 533 q5, q6, q7, q8, q9, q10, |
| 534 &q5, &q6, &q7, &q8); |
| 535 |
| 536 q4ResultL.val[0] = vget_low_u8(q5); // d10 |
| 537 q4ResultL.val[1] = vget_low_u8(q6); // d12 |
| 538 q4ResultL.val[2] = vget_low_u8(q7); // d14 |
| 539 q4ResultL.val[3] = vget_low_u8(q8); // d16 |
| 540 ud = u - 2; |
| 541 write_4x8(ud, pitch, q4ResultL); |
| 542 |
| 543 q4ResultH.val[0] = vget_high_u8(q5); // d11 |
| 544 q4ResultH.val[1] = vget_high_u8(q6); // d13 |
| 545 q4ResultH.val[2] = vget_high_u8(q7); // d15 |
| 546 q4ResultH.val[3] = vget_high_u8(q8); // d17 |
| 547 vd = v - 2; |
| 548 write_4x8(vd, pitch, q4ResultH); |
| 549 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 554673004:
libvpx: Pull from upstream (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/