Skip to content

P
Paddle-Lite

PaddlePaddle / Paddle-Lite

通知 331
Star 4
Fork 1
P
Paddle-Lite
切换分支/标签
查找文件 普通视图历史永久链接Permalink
image_convert.cc 37.1 KB
Newer Older
H
[LITE][ARM]add cv image process (#2402)  
HappyAngel 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 
// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "lite/utils/cv/image_convert.h"
#include <arm_neon.h>
#include <math.h>
#include <string.h>
namespace paddle {
namespace lite {
namespace utils {
namespace cv {
void nv_to_bgr(
    const uint8_t* src, uint8_t* dst, int srcw, int srch, int x_num, int y_num);

void nv_to_bgra(
    const uint8_t* src, uint8_t* dst, int srcw, int srch, int x_num, int y_num);

void nv21_to_bgr(const uint8_t* src, uint8_t* dst, int srcw, int srch);
void nv21_to_bgra(const uint8_t* src, uint8_t* dst, int srcw, int srch);
void nv12_to_bgr(const uint8_t* src, uint8_t* dst, int srcw, int srch);
void nv12_to_bgra(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgr rgb to gray
void hwc3_to_hwc1(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// gray to bgr rgb
void hwc1_to_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgr to bgra or rgb to rgba
void hwc3_to_hwc4(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgra to bgr or rgba to rgb
void hwc4_to_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgr to rgb or rgb to bgr
void hwc3_trans(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgra to rgba or rgba to bgra
void hwc4_trans(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgra to rgb or rgba to bgr
void hwc4_trans_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch);
// bgr to rgba or rgb to bgra
void hwc3_trans_hwc4(const uint8_t* src, uint8_t* dst, int srcw, int srch);

/*
  * image color convert
  * support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
  * BGR(RGB)and BGRA(RGBA) transform,
  * BGR(RGB)and RGB(BGR) transform,
  * BGR(RGB)and RGBA(BGRA) transform,
  * BGR(RGB)and GRAY transform,
  * param src: input image data
  * param dst: output image data
  * param srcFormat: input image image format support: GRAY, NV12(NV21),
 * BGR(RGB) and BGRA(RGBA)
  * param dstFormat: output image image format, support GRAY, BGR(RGB) and
 * BGRA(RGBA)
*/
void ImageConvert::choose(const uint8_t* src,
                          uint8_t* dst,
                          ImageFormat srcFormat,
                          ImageFormat dstFormat,
                          int srcw,
                          int srch) {
  if (srcFormat == dstFormat) {
    // copy
    int size = srcw * srch;
    if (srcFormat == NV12 || srcFormat == NV21) {
      size = srcw * (ceil(1.5 * srch));
    } else if (srcFormat == BGR || srcFormat == RGB) {
      size = 3 * srcw * srch;
    } else if (srcFormat == BGRA || srcFormat == RGBA) {
      size = 4 * srcw * srch;
    }
    memcpy(dst, src, sizeof(uint8_t) * size);
    return;
  } else {
    if (srcFormat == NV12 && (dstFormat == BGR || dstFormat == RGB)) {
      impl_ = nv12_to_bgr;
    } else if (srcFormat == NV21 && (dstFormat == BGR || dstFormat == RGB)) {
      impl_ = nv21_to_bgr;
    } else if (srcFormat == NV12 && (dstFormat == BGRA || dstFormat == RGBA)) {
      impl_ = nv12_to_bgra;
    } else if (srcFormat == NV21 && (dstFormat == BGRA || dstFormat == RGBA)) {
      impl_ = nv21_to_bgra;
    } else if ((srcFormat == RGBA && dstFormat == RGB) ||
               (srcFormat == BGRA && dstFormat == BGR)) {
      impl_ = hwc4_to_hwc3;
    } else if ((srcFormat == RGB && dstFormat == RGBA) ||
               (srcFormat == BGR && dstFormat == BGRA)) {
      impl_ = hwc3_to_hwc4;
    } else if ((srcFormat == RGB && dstFormat == BGR) ||
               (srcFormat == BGR && dstFormat == RGB)) {
      impl_ = hwc3_trans;
    } else if ((srcFormat == RGBA && dstFormat == BGRA) ||
               (srcFormat == BGRA && dstFormat == RGBA)) {
      impl_ = hwc4_trans;
    } else if ((srcFormat == RGB && dstFormat == GRAY) ||
               (srcFormat == BGR && dstFormat == GRAY)) {
      impl_ = hwc3_to_hwc1;
    } else if ((srcFormat == GRAY && dstFormat == RGB) ||
               (srcFormat == GRAY && dstFormat == BGR)) {
      impl_ = hwc1_to_hwc3;
    } else if ((srcFormat == RGBA && dstFormat == BGR) ||
               (srcFormat == BGRA && dstFormat == RGB)) {
      impl_ = hwc4_trans_hwc3;
    } else if ((srcFormat == RGB && dstFormat == BGRA) ||
               (srcFormat == BGR && dstFormat == RGBA)) {
      impl_ = hwc3_trans_hwc4;
    } else {
      printf("srcFormat: %d, dstFormat: %d does not support! \n",
             srcFormat,
             dstFormat);
    }
  }
  impl_(src, dst, srcw, srch);
}
/*
nv21(yvu)  to BGR: stroe hwc dsth * dstw = srch * (srcw)
y_w = srcw, y_h = srch uv_w = srcw uv_h = 1/2 * srch
R = Y + 1.402*(V-128);
G = Y - 0.34414*(U-128) - 0.71414*(V-128);
B = Y + 1.772*(U-128);
浮点乘法用 7位精度处理（即a*b = ((a << 7)*b )>>7）
ra = 1.402 *128 = 179.456 = 179
ga = 0.34414 * 64 = 44.3721 = 44
gb = 0.71414 * 64 = 91.40992 = 91
ba = 1.772 * 62 = 226.816 = 227
nv12bgr, nv21tobgr
*/
void nv_to_bgr(const uint8_t* src,
               uint8_t* dst,
               int srcw,
               int srch,
               int x_num,
               int y_num) {
  // nv21 x = 0, y = 1
  // nv12 x = 1, y = 0
  int y_h = srch;
  int wout = srcw * 3;
  const uint8_t* y = src;
  const uint8_t* vu = src + y_h * srcw;

  int16x8_t bias = vdupq_n_s16(128);
  int16x8_t ga = vdupq_n_s16(44);
  int16x8_t ra = vdupq_n_s16(179);
  int16x8_t ba = vdupq_n_s16(227);
  int16x8_t gb = vdupq_n_s16(91);
  int16x8_t zero = vdupq_n_s16(0);
  int16x8_t max = vdupq_n_s16(255);

  uint8_t* zerobuf = new uint8_t[srcw];
  uint8_t* writebuf = new uint8_t[wout];
  memset(zerobuf, 0, sizeof(uint8_t) * srcw);

  int i = 0;
#pragma omp parallel for
  for (i = 0; i < y_h; i += 2) {
    const uint8_t* ptr_y1 = y + i * srcw;
    const uint8_t* ptr_y2 = ptr_y1 + srcw;
    const uint8_t* ptr_vu = vu + (i / 2) * srcw;
    uint8_t* ptr_bgr1 = dst + i * wout;
    uint8_t* ptr_bgr2 = ptr_bgr1 + wout;
    if (i + 2 > y_h) {
      ptr_y2 = zerobuf;
      ptr_bgr2 = writebuf;
    }
    int j = 0;
    for (; j < srcw - 15; j += 16) {
      uint8x8x2_t y1 = vld2_u8(ptr_y1);  // d8 = y0y2y4y6...y14 d9 =
                                         // y1y3y5...y15
      uint8x8x2_t vu =
          vld2_u8(ptr_vu);  // d0 = v0v1v2v3v4v5...v7 d1 = u0u1u2...u7

      uint8x8x2_t y2 = vld2_u8(ptr_y2);

      uint16x8_t v = vmovl_u8(vu.val[x_num]);
      uint16x8_t u = vmovl_u8(vu.val[y_num]);
      int16x8_t v_s = vreinterpretq_s16_u16(v);
      int16x8_t u_s = vreinterpretq_s16_u16(u);
      int16x8_t v_bias = vsubq_s16(v_s, bias);
      int16x8_t u_bias = vsubq_s16(u_s, bias);

      // G = Y - 0.34414*(U-128) - 0.71414*(V-128);
      int16x8_t g0 = vmulq_s16(ga, u_bias);
      // R = Y + 1.402*(V-128);
      int16x8_t r0 = vmulq_s16(ra, v_bias);
      // B = Y + 1.772*(U-128);
      int16x8_t b0 = vmulq_s16(ba, u_bias);

      g0 = vmlaq_s16(g0, gb, v_bias);

      int16x8_t y1_0_8 = vreinterpretq_s16_u16(vmovl_u8(y1.val[0]));
      int16x8_t y1_1_8 = vreinterpretq_s16_u16(vmovl_u8(y1.val[1]));

      int16x8_t y2_0_8 = vreinterpretq_s16_u16(vmovl_u8(y2.val[0]));
      int16x8_t y2_1_8 = vreinterpretq_s16_u16(vmovl_u8(y2.val[1]));

      int16x8_t r0_bias = vshrq_n_s16(r0, 7);  // r0 / 128
      int16x8_t b0_bias = vshrq_n_s16(b0, 7);
      int16x8_t g0_bias = vshrq_n_s16(g0, 7);

      int16x8_t r0_1 = vaddq_s16(y1_0_8, r0_bias);
      int16x8_t b0_1 = vaddq_s16(y1_0_8, b0_bias);
      int16x8_t g0_1 = vsubq_s16(y1_0_8, g0_bias);  // g0_1 = y1_0_8 - g0_1

      int16x8_t r0_2 = vaddq_s16(y1_1_8, r0_bias);
      int16x8_t b0_2 = vaddq_s16(y1_1_8, b0_bias);
      int16x8_t g0_2 = vsubq_s16(y1_1_8, g0_bias);

      r0_1 = vmaxq_s16(r0_1, zero);
      b0_1 = vmaxq_s16(b0_1, zero);
      g0_1 = vmaxq_s16(g0_1, zero);

      r0_2 = vmaxq_s16(r0_2, zero);
      b0_2 = vmaxq_s16(b0_2, zero);
      g0_2 = vmaxq_s16(g0_2, zero);

      r0_1 = vminq_s16(r0_1, max);
      b0_1 = vminq_s16(b0_1, max);
      g0_1 = vminq_s16(g0_1, max);

      r0_2 = vminq_s16(r0_2, max);
      b0_2 = vminq_s16(b0_2, max);
      g0_2 = vminq_s16(g0_2, max);

      uint8x8_t r00 = vreinterpret_u8_s8(vmovn_s16(r0_1));
      uint8x8_t b00 = vreinterpret_u8_s8(vmovn_s16(b0_1));
      uint8x8_t g00 = vreinterpret_u8_s8(vmovn_s16(g0_1));

      uint8x8_t r01 = vreinterpret_u8_s8(vmovn_s16(r0_2));
      uint8x8_t b01 = vreinterpret_u8_s8(vmovn_s16(b0_2));
      uint8x8_t g01 = vreinterpret_u8_s8(vmovn_s16(g0_2));

      int16x8_t r1_1 = vaddq_s16(y2_0_8, r0_bias);
      int16x8_t b1_1 = vaddq_s16(y2_0_8, b0_bias);
      int16x8_t g1_1 = vsubq_s16(y2_0_8, g0_bias);  // g0_1 = y1_0_8 - g0_1

      int16x8_t r1_2 = vaddq_s16(y2_1_8, r0_bias);
      int16x8_t b1_2 = vaddq_s16(y2_1_8, b0_bias);
      int16x8_t g1_2 = vsubq_s16(y2_1_8, g0_bias);

      uint8x8x2_t r00_0 = vtrn_u8(r00, r01);  // 014589  236710
      uint8x8x2_t b00_0 = vtrn_u8(b00, b01);
      uint8x8x2_t g00_0 = vtrn_u8(g00, g01);

      r1_1 = vmaxq_s16(r1_1, zero);
      b1_1 = vmaxq_s16(b1_1, zero);
      g1_1 = vmaxq_s16(g1_1, zero);

      r1_2 = vmaxq_s16(r1_2, zero);
      b1_2 = vmaxq_s16(b1_2, zero);
      g1_2 = vmaxq_s16(g1_2, zero);

      uint16x4_t r0_16 = vreinterpret_u16_u8(r00_0.val[0]);
      uint16x4_t r1_16 = vreinterpret_u16_u8(r00_0.val[1]);

      uint16x4_t b0_16 = vreinterpret_u16_u8(b00_0.val[0]);
      uint16x4_t b1_16 = vreinterpret_u16_u8(b00_0.val[1]);

      uint16x4_t g0_16 = vreinterpret_u16_u8(g00_0.val[0]);
      uint16x4_t g1_16 = vreinterpret_u16_u8(g00_0.val[1]);

      uint16x4x2_t r00_1 = vtrn_u16(r0_16, r1_16);  // 012389 456710
      uint16x4x2_t b00_1 = vtrn_u16(b0_16, b1_16);
      uint16x4x2_t g00_1 = vtrn_u16(g0_16, g1_16);

      r1_1 = vminq_s16(r1_1, max);
      b1_1 = vminq_s16(b1_1, max);
      g1_1 = vminq_s16(g1_1, max);

      r1_2 = vminq_s16(r1_2, max);
      b1_2 = vminq_s16(b1_2, max);
      g1_2 = vminq_s16(g1_2, max);

      uint32x2_t r0_32 = vreinterpret_u32_u16(r00_1.val[0]);
      uint32x2_t r1_32 = vreinterpret_u32_u16(r00_1.val[1]);

      uint32x2_t b0_32 = vreinterpret_u32_u16(b00_1.val[0]);
      uint32x2_t b1_32 = vreinterpret_u32_u16(b00_1.val[1]);

      uint32x2_t g0_32 = vreinterpret_u32_u16(g00_1.val[0]);
      uint32x2_t g1_32 = vreinterpret_u32_u16(g00_1.val[1]);

      uint32x2x2_t r00_2 = vtrn_u32(r0_32, r1_32);  // 01234567 8910
      uint32x2x2_t b00_2 = vtrn_u32(b0_32, b1_32);
      uint32x2x2_t g00_2 = vtrn_u32(g0_32, g1_32);

      r00 = vreinterpret_u8_s8(vmovn_s16(r1_1));
      b00 = vreinterpret_u8_s8(vmovn_s16(b1_1));
      g00 = vreinterpret_u8_s8(vmovn_s16(g1_1));

      r01 = vreinterpret_u8_s8(vmovn_s16(r1_2));
      b01 = vreinterpret_u8_s8(vmovn_s16(b1_2));
      g01 = vreinterpret_u8_s8(vmovn_s16(g1_2));

      uint8x8_t r0_8 = vreinterpret_u8_u32(r00_2.val[0]);
      uint8x8_t b0_8 = vreinterpret_u8_u32(b00_2.val[0]);
      uint8x8_t g0_8 = vreinterpret_u8_u32(g00_2.val[0]);

      uint8x8_t r1_8 = vreinterpret_u8_u32(r00_2.val[1]);
      uint8x8_t b1_8 = vreinterpret_u8_u32(b00_2.val[1]);
      uint8x8_t g1_8 = vreinterpret_u8_u32(g00_2.val[1]);

      uint8x8x3_t v_bgr;
      v_bgr.val[0] = b0_8;
      v_bgr.val[1] = g0_8;
      v_bgr.val[2] = r0_8;

      r00_0 = vtrn_u8(r00, r01);  // 014589  236710
      b00_0 = vtrn_u8(b00, b01);
      g00_0 = vtrn_u8(g00, g01);

      vst3_u8(ptr_bgr1, v_bgr);

      r0_16 = vreinterpret_u16_u8(r00_0.val[0]);
      r1_16 = vreinterpret_u16_u8(r00_0.val[1]);

      b0_16 = vreinterpret_u16_u8(b00_0.val[0]);
      b1_16 = vreinterpret_u16_u8(b00_0.val[1]);

      g0_16 = vreinterpret_u16_u8(g00_0.val[0]);
      g1_16 = vreinterpret_u16_u8(g00_0.val[1]);

      ptr_bgr1 += 24;
      uint8x8x3_t v_bgr1;
      v_bgr1.val[0] = b1_8;
      v_bgr1.val[1] = g1_8;
      v_bgr1.val[2] = r1_8;

      r00_1 = vtrn_u16(r0_16, r1_16);  // 012389 456710
      b00_1 = vtrn_u16(b0_16, b1_16);
      g00_1 = vtrn_u16(g0_16, g1_16);

      vst3_u8(ptr_bgr1, v_bgr1);

      r0_32 = vreinterpret_u32_u16(r00_1.val[0]);
      r1_32 = vreinterpret_u32_u16(r00_1.val[1]);

      b0_32 = vreinterpret_u32_u16(b00_1.val[0]);
      b1_32 = vreinterpret_u32_u16(b00_1.val[1]);

      g0_32 = vreinterpret_u32_u16(g00_1.val[0]);
      g1_32 = vreinterpret_u32_u16(g00_1.val[1]);

      ptr_bgr1 += 24;

      r00_2 = vtrn_u32(r0_32, r1_32);  // 01234567 8910
      b00_2 = vtrn_u32(b0_32, b1_32);
      g00_2 = vtrn_u32(g0_32, g1_32);

      ptr_vu += 16;
      ptr_y1 += 16;
      ptr_y2 += 16;

      r0_8 = vreinterpret_u8_u32(r00_2.val[0]);
      b0_8 = vreinterpret_u8_u32(b00_2.val[0]);
      g0_8 = vreinterpret_u8_u32(g00_2.val[0]);

      r1_8 = vreinterpret_u8_u32(r00_2.val[1]);
      b1_8 = vreinterpret_u8_u32(b00_2.val[1]);
      g1_8 = vreinterpret_u8_u32(g00_2.val[1]);

      v_bgr.val[0] = b0_8;
      v_bgr.val[1] = g0_8;
      v_bgr.val[2] = r0_8;

      v_bgr1.val[0] = b1_8;
      v_bgr1.val[1] = g1_8;
      v_bgr1.val[2] = r1_8;

      vst3_u8(ptr_bgr2, v_bgr);
      vst3_u8(ptr_bgr2 + 24, v_bgr1);

      ptr_bgr2 += 48;
    }
    // two data
    for (; j < srcw; j += 2) {
      uint8_t _y0 = ptr_y1[0];
      uint8_t _y1 = ptr_y1[1];
      uint8_t _v = ptr_vu[x_num];
      uint8_t _u = ptr_vu[y_num];
      uint8_t _y0_1 = ptr_y2[0];
      uint8_t _y1_1 = ptr_y2[1];

      int ra = floor((179 * (_v - 128)) >> 7);
      int ga = floor((44 * (_u - 128) + 91 * (_v - 128)) >> 7);
      int ba = floor((227 * (_u - 128)) >> 7);

      int r = _y0 + ra;
      int g = _y0 - ga;
      int b = _y0 + ba;

      int r1 = _y1 + ra;
      int g1 = _y1 - ga;
      int b1 = _y1 + ba;

      r = r < 0 ? 0 : (r > 255) ? 255 : r;
      g = g < 0 ? 0 : (g > 255) ? 255 : g;
      b = b < 0 ? 0 : (b > 255) ? 255 : b;

      r1 = r1 < 0 ? 0 : (r1 > 255) ? 255 : r1;
      g1 = g1 < 0 ? 0 : (g1 > 255) ? 255 : g1;
      b1 = b1 < 0 ? 0 : (b1 > 255) ? 255 : b1;

      *ptr_bgr1++ = b;
      *ptr_bgr1++ = g;
      *ptr_bgr1++ = r;

      int r2 = _y0_1 + ra;
      int g2 = _y0_1 - ga;
      int b2 = _y0_1 + ba;

      int r3 = _y1_1 + ra;
      int g3 = _y1_1 - ga;
      int b3 = _y1_1 + ba;

      r2 = r2 < 0 ? 0 : (r2 > 255) ? 255 : r2;
      g2 = g2 < 0 ? 0 : (g2 > 255) ? 255 : g2;
      b2 = b2 < 0 ? 0 : (b2 > 255) ? 255 : b2;

      r3 = r3 < 0 ? 0 : (r3 > 255) ? 255 : r3;
      g3 = g3 < 0 ? 0 : (g3 > 255) ? 255 : g3;
      b3 = b3 < 0 ? 0 : (b3 > 255) ? 255 : b3;

      *ptr_bgr1++ = b1;
      *ptr_bgr1++ = g1;
      *ptr_bgr1++ = r1;

      *ptr_bgr2++ = b2;
      *ptr_bgr2++ = g2;
      *ptr_bgr2++ = r2;

      ptr_y1 += 2;
      ptr_y2 += 2;
      ptr_vu += 2;

      *ptr_bgr2++ = b3;
      *ptr_bgr2++ = g3;
      *ptr_bgr2++ = r3;
    }
  }
  delete[] zerobuf;
  delete[] writebuf;
}
// nv12bgra, nv21tobgra
void nv_to_bgra(const uint8_t* src,
                uint8_t* dst,
                int srcw,
                int srch,
                int x_num,
                int y_num) {
  // nv21 x = 0, y = 1
  // nv12 x = 1, y = 0
  int y_h = srch;
  int vu_h = 1 / 2 * srch;
  const uint8_t* y = src;
  const uint8_t* vu = src + y_h * srcw;
  int wout = srcw * 4;

  uint8_t* zerobuf = new uint8_t[srcw];
  uint8_t* writebuf = new uint8_t[wout];
  memset(zerobuf, 0, sizeof(uint8_t) * srcw);

  int16x8_t bias = vdupq_n_s16(128);
  int16x8_t ga = vdupq_n_s16(44);
  int16x8_t ra = vdupq_n_s16(179);
  int16x8_t ba = vdupq_n_s16(227);
  int16x8_t gb = vdupq_n_s16(91);
  int16x8_t zero = vdupq_n_s16(0);
  int16x8_t max = vdupq_n_s16(255);
  uint8x8_t a_8 = vdup_n_u8(255);
#pragma omp parallel for
  for (int i = 0; i < y_h; i += 2) {
    const uint8_t* ptr_y1 = y + i * srcw;
    const uint8_t* ptr_y2 = ptr_y1 + srcw;
    const uint8_t* ptr_vu = vu + (i / 2) * srcw;
    uint8_t* ptr_bgr1 = dst + i * wout;
    uint8_t* ptr_bgr2 = ptr_bgr1 + wout;
    if (i + 2 > y_h) {
      ptr_y2 = zerobuf;
      ptr_bgr2 = writebuf;
    }
    int j = 0;
    for (; j < srcw - 15; j += 16) {
      uint8x8x2_t y1 = vld2_u8(ptr_y1);  // d8 = y0y2y4y6...y14 d9 =
                                         // y1y3y5...y15
      uint8x8x2_t vu =
          vld2_u8(ptr_vu);  // d0 = v0v1v2v3v4v5...v7 d1 = u0u1u2...u7

      uint8x8x2_t y2 = vld2_u8(ptr_y2);

      uint16x8_t v = vmovl_u8(vu.val[x_num]);
      uint16x8_t u = vmovl_u8(vu.val[y_num]);
      int16x8_t v_s = vreinterpretq_s16_u16(v);
      int16x8_t u_s = vreinterpretq_s16_u16(u);
      int16x8_t v_bias = vsubq_s16(v_s, bias);
      int16x8_t u_bias = vsubq_s16(u_s, bias);

      // G = Y - 0.34414*(U-128) - 0.71414*(V-128);
      int16x8_t g0 = vmulq_s16(ga, u_bias);
      // R = Y + 1.402*(V-128);
      int16x8_t r0 = vmulq_s16(ra, v_bias);
      // B = Y + 1.772*(U-128);
      int16x8_t b0 = vmulq_s16(ba, u_bias);

      g0 = vmlaq_s16(g0, gb, v_bias);

      int16x8_t y1_0_8 = vreinterpretq_s16_u16(vmovl_u8(y1.val[0]));
      int16x8_t y1_1_8 = vreinterpretq_s16_u16(vmovl_u8(y1.val[1]));

      int16x8_t y2_0_8 = vreinterpretq_s16_u16(vmovl_u8(y2.val[0]));
      int16x8_t y2_1_8 = vreinterpretq_s16_u16(vmovl_u8(y2.val[1]));

      int16x8_t r0_bias = vshrq_n_s16(r0, 7);  // r0 / 128
      int16x8_t b0_bias = vshrq_n_s16(b0, 7);
      int16x8_t g0_bias = vshrq_n_s16(g0, 7);

      int16x8_t r0_1 = vaddq_s16(y1_0_8, r0_bias);
      int16x8_t b0_1 = vaddq_s16(y1_0_8, b0_bias);
      int16x8_t g0_1 = vsubq_s16(y1_0_8, g0_bias);  // g0_1 = y1_0_8 - g0_1

      int16x8_t r0_2 = vaddq_s16(y1_1_8, r0_bias);
      int16x8_t b0_2 = vaddq_s16(y1_1_8, b0_bias);
      int16x8_t g0_2 = vsubq_s16(y1_1_8, g0_bias);

      r0_1 = vmaxq_s16(r0_1, zero);
      b0_1 = vmaxq_s16(b0_1, zero);
      g0_1 = vmaxq_s16(g0_1, zero);

      r0_2 = vmaxq_s16(r0_2, zero);
      b0_2 = vmaxq_s16(b0_2, zero);
      g0_2 = vmaxq_s16(g0_2, zero);

      r0_1 = vminq_s16(r0_1, max);
      b0_1 = vminq_s16(b0_1, max);
      g0_1 = vminq_s16(g0_1, max);

      r0_2 = vminq_s16(r0_2, max);
      b0_2 = vminq_s16(b0_2, max);
      g0_2 = vminq_s16(g0_2, max);

      uint8x8_t r00 = vreinterpret_u8_s8(vmovn_s16(r0_1));
      uint8x8_t b00 = vreinterpret_u8_s8(vmovn_s16(b0_1));
      uint8x8_t g00 = vreinterpret_u8_s8(vmovn_s16(g0_1));

      uint8x8_t r01 = vreinterpret_u8_s8(vmovn_s16(r0_2));
      uint8x8_t b01 = vreinterpret_u8_s8(vmovn_s16(b0_2));
      uint8x8_t g01 = vreinterpret_u8_s8(vmovn_s16(g0_2));

      int16x8_t r1_1 = vaddq_s16(y2_0_8, r0_bias);
      int16x8_t b1_1 = vaddq_s16(y2_0_8, b0_bias);
      int16x8_t g1_1 = vsubq_s16(y2_0_8, g0_bias);  // g0_1 = y1_0_8 - g0_1

      int16x8_t r1_2 = vaddq_s16(y2_1_8, r0_bias);
      int16x8_t b1_2 = vaddq_s16(y2_1_8, b0_bias);
      int16x8_t g1_2 = vsubq_s16(y2_1_8, g0_bias);

      uint8x8x2_t r00_0 = vtrn_u8(r00, r01);  // 014589  236710
      uint8x8x2_t b00_0 = vtrn_u8(b00, b01);
      uint8x8x2_t g00_0 = vtrn_u8(g00, g01);

      r1_1 = vmaxq_s16(r1_1, zero);
      b1_1 = vmaxq_s16(b1_1, zero);
      g1_1 = vmaxq_s16(g1_1, zero);

      r1_2 = vmaxq_s16(r1_2, zero);
      b1_2 = vmaxq_s16(b1_2, zero);
      g1_2 = vmaxq_s16(g1_2, zero);

      uint16x4_t r0_16 = vreinterpret_u16_u8(r00_0.val[0]);
      uint16x4_t r1_16 = vreinterpret_u16_u8(r00_0.val[1]);

      uint16x4_t b0_16 = vreinterpret_u16_u8(b00_0.val[0]);
      uint16x4_t b1_16 = vreinterpret_u16_u8(b00_0.val[1]);

      uint16x4_t g0_16 = vreinterpret_u16_u8(g00_0.val[0]);
      uint16x4_t g1_16 = vreinterpret_u16_u8(g00_0.val[1]);

      uint16x4x2_t r00_1 = vtrn_u16(r0_16, r1_16);  // 012389 456710
      uint16x4x2_t b00_1 = vtrn_u16(b0_16, b1_16);
      uint16x4x2_t g00_1 = vtrn_u16(g0_16, g1_16);

      r1_1 = vminq_s16(r1_1, max);
      b1_1 = vminq_s16(b1_1, max);
      g1_1 = vminq_s16(g1_1, max);

      r1_2 = vminq_s16(r1_2, max);
      b1_2 = vminq_s16(b1_2, max);
      g1_2 = vminq_s16(g1_2, max);

      uint32x2_t r0_32 = vreinterpret_u32_u16(r00_1.val[0]);
      uint32x2_t r1_32 = vreinterpret_u32_u16(r00_1.val[1]);

      uint32x2_t b0_32 = vreinterpret_u32_u16(b00_1.val[0]);
      uint32x2_t b1_32 = vreinterpret_u32_u16(b00_1.val[1]);

      uint32x2_t g0_32 = vreinterpret_u32_u16(g00_1.val[0]);
      uint32x2_t g1_32 = vreinterpret_u32_u16(g00_1.val[1]);

      uint32x2x2_t r00_2 = vtrn_u32(r0_32, r1_32);  // 01234567 8910
      uint32x2x2_t b00_2 = vtrn_u32(b0_32, b1_32);
      uint32x2x2_t g00_2 = vtrn_u32(g0_32, g1_32);

      r00 = vreinterpret_u8_s8(vmovn_s16(r1_1));
      b00 = vreinterpret_u8_s8(vmovn_s16(b1_1));
      g00 = vreinterpret_u8_s8(vmovn_s16(g1_1));

      r01 = vreinterpret_u8_s8(vmovn_s16(r1_2));
      b01 = vreinterpret_u8_s8(vmovn_s16(b1_2));
      g01 = vreinterpret_u8_s8(vmovn_s16(g1_2));

      uint8x8_t r0_8 = vreinterpret_u8_u32(r00_2.val[0]);
      uint8x8_t b0_8 = vreinterpret_u8_u32(b00_2.val[0]);
      uint8x8_t g0_8 = vreinterpret_u8_u32(g00_2.val[0]);

      uint8x8_t r1_8 = vreinterpret_u8_u32(r00_2.val[1]);
      uint8x8_t b1_8 = vreinterpret_u8_u32(b00_2.val[1]);
      uint8x8_t g1_8 = vreinterpret_u8_u32(g00_2.val[1]);

      uint8x8x4_t v_bgr;
      v_bgr.val[0] = b0_8;
      v_bgr.val[1] = g0_8;
      v_bgr.val[2] = r0_8;
      v_bgr.val[3] = a_8;

      r00_0 = vtrn_u8(r00, r01);  // 014589  236710
      b00_0 = vtrn_u8(b00, b01);
      g00_0 = vtrn_u8(g00, g01);

      // ptr_bgr3 += 8;
      // ptr_bgr1 += 8;
      // ptr_bgr2 += 8;
      // vst3_u8(ptr_bgr1, v_bgr);
      vst4_u8(ptr_bgr1, v_bgr);

      r0_16 = vreinterpret_u16_u8(r00_0.val[0]);
      r1_16 = vreinterpret_u16_u8(r00_0.val[1]);

      b0_16 = vreinterpret_u16_u8(b00_0.val[0]);
      b1_16 = vreinterpret_u16_u8(b00_0.val[1]);

      g0_16 = vreinterpret_u16_u8(g00_0.val[0]);
      g1_16 = vreinterpret_u16_u8(g00_0.val[1]);

      ptr_bgr1 += 32;
      // uint8x8x3_t v_bgr1;
      uint8x8x4_t v_bgr1;
      v_bgr1.val[0] = b1_8;
      v_bgr1.val[1] = g1_8;
      v_bgr1.val[2] = r1_8;
      v_bgr1.val[3] = a_8;

      r00_1 = vtrn_u16(r0_16, r1_16);  // 012389 456710
      b00_1 = vtrn_u16(b0_16, b1_16);
      g00_1 = vtrn_u16(g0_16, g1_16);

      // vst3_u8(ptr_bgr1, v_bgr1);
      vst4_u8(ptr_bgr1, v_bgr1);

      r0_32 = vreinterpret_u32_u16(r00_1.val[0]);
      r1_32 = vreinterpret_u32_u16(r00_1.val[1]);

      b0_32 = vreinterpret_u32_u16(b00_1.val[0]);
      b1_32 = vreinterpret_u32_u16(b00_1.val[1]);

      g0_32 = vreinterpret_u32_u16(g00_1.val[0]);
      g1_32 = vreinterpret_u32_u16(g00_1.val[1]);

      // ptr_bgr1 += 24;
      ptr_bgr1 += 32;

      r00_2 = vtrn_u32(r0_32, r1_32);  // 01234567 8910
      b00_2 = vtrn_u32(b0_32, b1_32);
      g00_2 = vtrn_u32(g0_32, g1_32);

      ptr_vu += 16;
      ptr_y1 += 16;
      ptr_y2 += 16;

      r0_8 = vreinterpret_u8_u32(r00_2.val[0]);
      b0_8 = vreinterpret_u8_u32(b00_2.val[0]);
      g0_8 = vreinterpret_u8_u32(g00_2.val[0]);

      r1_8 = vreinterpret_u8_u32(r00_2.val[1]);
      b1_8 = vreinterpret_u8_u32(b00_2.val[1]);
      g1_8 = vreinterpret_u8_u32(g00_2.val[1]);

      v_bgr.val[0] = b0_8;
      v_bgr.val[1] = g0_8;
      v_bgr.val[2] = r0_8;

      v_bgr1.val[0] = b1_8;
      v_bgr1.val[1] = g1_8;
      v_bgr1.val[2] = r1_8;

      // vst3_u8(ptr_bgr2, v_bgr);
      // vst3_u8(ptr_bgr2 + 24, v_bgr1);
      vst4_u8(ptr_bgr2, v_bgr);
      vst4_u8(ptr_bgr2 + 32, v_bgr1);

      ptr_bgr2 += 64;
    }
    // two data
    for (; j < srcw; j += 2) {
      uint8_t _y0 = ptr_y1[0];
      uint8_t _y1 = ptr_y1[1];
      uint8_t _v = ptr_vu[x_num];
      uint8_t _u = ptr_vu[y_num];
      uint8_t _y0_1 = ptr_y2[0];
      uint8_t _y1_1 = ptr_y2[1];

      int ra = floor((179 * (_v - 128)) >> 7);
      int ga = floor((44 * (_u - 128) + 91 * (_v - 128)) >> 7);
      int ba = floor((227 * (_u - 128)) >> 7);

      int r = _y0 + ra;
      int g = _y0 - ga;
      int b = _y0 + ba;

      int r1 = _y1 + ra;
      int g1 = _y1 - ga;
      int b1 = _y1 + ba;

      r = r < 0 ? 0 : (r > 255) ? 255 : r;
      g = g < 0 ? 0 : (g > 255) ? 255 : g;
      b = b < 0 ? 0 : (b > 255) ? 255 : b;

      r1 = r1 < 0 ? 0 : (r1 > 255) ? 255 : r1;
      g1 = g1 < 0 ? 0 : (g1 > 255) ? 255 : g1;
      b1 = b1 < 0 ? 0 : (b1 > 255) ? 255 : b1;

      // *ptr_bgr1++ = b;
      // *ptr_bgr2++ = g;
      // *ptr_bgr3++ = r;
      *ptr_bgr1++ = b;
      *ptr_bgr1++ = g;
      *ptr_bgr1++ = r;
      *ptr_bgr1++ = 255;

      int r2 = _y0_1 + ra;
      int g2 = _y0_1 - ga;
      int b2 = _y0_1 + ba;

      int r3 = _y1_1 + ra;
      int g3 = _y1_1 - ga;
      int b3 = _y1_1 + ba;

      r2 = r2 < 0 ? 0 : (r2 > 255) ? 255 : r2;
      g2 = g2 < 0 ? 0 : (g2 > 255) ? 255 : g2;
      b2 = b2 < 0 ? 0 : (b2 > 255) ? 255 : b2;

      r3 = r3 < 0 ? 0 : (r3 > 255) ? 255 : r3;
      g3 = g3 < 0 ? 0 : (g3 > 255) ? 255 : g3;
      b3 = b3 < 0 ? 0 : (b3 > 255) ? 255 : b3;

      *ptr_bgr1++ = b1;
      *ptr_bgr1++ = g1;
      *ptr_bgr1++ = r1;
      *ptr_bgr1++ = 255;

      *ptr_bgr2++ = b2;
      *ptr_bgr2++ = g2;
      *ptr_bgr2++ = r2;
      *ptr_bgr2++ = 255;

      ptr_y1 += 2;
      ptr_y2 += 2;
      ptr_vu += 2;

      *ptr_bgr2++ = b3;
      *ptr_bgr2++ = g3;
      *ptr_bgr2++ = r3;
      *ptr_bgr2++ = 255;
    }
  }
  delete[] zerobuf;
  delete[] writebuf;
}
void nv21_to_bgr(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  nv_to_bgr(src, dst, srcw, srch, 0, 1);
}

// nv12(yuv)  to BGR:store hwc dsth * dstw = srch * srcw y_w = srcw, y_h = srch
// uv_w = srcw uv_h = 1/2 * srch
void nv12_to_bgr(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  // exchange vu forward
  nv_to_bgr(src, dst, srcw, srch, 1, 0);
}
// nv21(yvu)  to BGRA: stroe hwc dsth * dstw = srch * (srcw) y_w = srcw, y_h =
// srch uv_w = srcw uv_h = 1/2 * srch
void nv21_to_bgra(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  nv_to_bgra(src, dst, srcw, srch, 0, 1);
}
// nv12(yuv)  to BGRA:store hwc dsth * dstw = srch * srcw y_w = srcw, y_h = srch
// uv_w = srcw uv_h = 1/2 * srch
void nv12_to_bgra(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  nv_to_bgra(src, dst, srcw, srch, 1, 0);
}
/*
采用CV_BGR2GRAY,转换公式Gray = 0.1140*B + 0.5870*G + 0.2989*R
采用CV_RGB2GRAY,转换公式Gray = 0.1140*R + 0.5870*G + 0.2989*B
b = 0.114 *128 = 14.529 = 15
g = 0.587 * 128 = 75.136 = 75
r = 0.2989 * 127 = 38.2592 = 38
Gray = (15*B + 75*G + 38*R)/128
bgr2gray, rgb2gray
*/
void hwc3_to_hwc1(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  uint8_t b = 15;
  uint8_t g = 75;
  uint8_t r = 38;

  uint8x8_t vb = vdup_n_u8(b);
  uint8x8_t vg = vdup_n_u8(g);
  uint8x8_t vr = vdup_n_u8(r);
#ifdef __aarch64__
#else
  uint8_t vb_array[8] = {b, b, b, b, b, b, b, b};
  uint8_t vg_array[8] = {g, g, g, g, g, g, g, g};
  uint8_t vr_array[8] = {r, r, r, r, r, r, r, r};
#endif
  int cnt_pro = srcw >> 3;
  int remain_pro = srcw % 8;
  int win = srcw * 3;
  int i = 0;
#pragma omp parallel for
  for (i = 0; i < srch - 3; i += 4) {
    int j = 0;
    const uint8_t* inptr0 = src + i * win;
    const uint8_t* inptr1 = inptr0 + win;
    const uint8_t* inptr2 = inptr1 + win;
    const uint8_t* inptr3 = inptr2 + win;
    uint8_t* outr0 = dst + i * srcw;
    uint8_t* outr1 = outr0 + srcw;
    uint8_t* outr2 = outr1 + srcw;
    uint8_t* outr3 = outr2 + srcw;

    int cnt = cnt_pro;
    if (cnt > 0) {
#ifdef __aarch64__
      asm volatile(
          "prfm   pldl1keep, [%[inptr0]]                \n"
          "prfm   pldl1keep, [%[inptr0], #128]   \n"
          "prfm   pldl1keep, [%[inptr1]]        \n"
          "prfm   pldl1keep, [%[inptr1], #128]   \n"
          "prfm   pldl1keep, [%[inptr2]]                \n"
          "prfm   pldl1keep, [%[inptr2], #128]   \n"
          "prfm   pldl1keep, [%[inptr3]]                \n"
          "prfm   pldl1keep, [%[inptr3], #128]   \n"
          "1: \n"
          "ld3 {v0.8b - v2.8b}, [%[inptr0]], #24 \n"   // d8 = y0y3y6y9.. d9 =
                                                       // y1y4y7...
          "ld3 {v3.8b - v5.8b}, [%[inptr1]], #24 \n"   // d8 = y0y3y6y9.. d9 =
                                                       // y1y4y7...
          "ld3 {v6.8b - v8.8b}, [%[inptr2]], #24 \n"   // d8 = y0y3y6y9.. d9 =
                                                       // y1y4y7...
          "ld3 {v9.8b - v11.8b}, [%[inptr3]], #24 \n"  // d8 = y0y3y6y9.. d9 =
                                                       // y1y4y7...
          // mul b
          "umull v12.8h, v0.8b, %w[vb].8b \n"  // v0 * vb
          "umull v13.8h, v3.8b, %w[vb].8b \n"  // v0 * vb
          "umull v14.8h, v6.8b, %w[vb].8b \n"  // v0 * vb
          "umull v15.8h, v9.8b, %w[vb].8b \n"  // v0 * vb
          // mul g
          "umull v16.8h, v1.8b, %w[vg].8b \n"   // v0 * vb
          "umull v17.8h, v4.8b, %w[vg].8b \n"   // v0 * vb
          "umull v18.8h, v7.8b, %w[vg].8b \n"   // v0 * vb
          "umull v19.8h, v10.8b, %w[vg].8b \n"  // v0 * vb
          // mul r
          "umlal v12.8h, v2.8b, %w[vr].8b \n"   // v0 * vb
          "umlal v13.8h, v5.8b, %w[vr].8b \n"   // v0 * vb
          "umlal v14.8h, v8.8b, %w[vr].8b \n"   // v0 * vb
          "umlal v15.8h, v11.8b, %w[vr].8b \n"  // v0 * vb
          // 16->32
          "uaddl v0.4s, v16.4h, v12.4h \n"
          "uaddl2 v1.4s, v16.8h, v12.8h \n"
          "uaddl v2.4s, v17.4h, v13.4h \n"
          "uaddl2 v3.4s, v17.8h, v13.8h \n"
          "uaddl v4.4s, v18.4h, v14.4h \n"
          "uaddl2 v5.4s, v18.8h, v14.8h \n"
          "uaddl v6.4s, v19.4h, v15.4h \n"
          "uaddl2 v7.4s, v19.8h, v15.8h \n"
          // 32->16 v0 >> 7
          "shrn v12.4h, v0.4s, #7 \n"
          "shrn2 v12.8h, v1.4s, #7 \n"
          "shrn v13.4h, v2.4s, #7 \n"
          "shrn2 v13.8h, v3.4s, #7 \n"
          "shrn v14.4h, v4.4s, #7 \n"
          "shrn2 v14.8h, v5.4s, #7 \n"
          "shrn v15.4h, v6.4s, #7 \n"
          "shrn2 v15.8h, v7.4s, #7 \n"
          // 16->8
          "xtn v0.8b, v12.8h \n"
          "xtn v1.8b, v13.8h \n"
          "xtn v2.8b, v14.8h \n"
          "xtn v3.8b, v15.8h \n"
          "subs %w[cnt], %w[cnt], #1 \n"
          "st1 {v0.8b}, [%[outr0]], #8 \n"
          "st1 {v1.8b}, [%[outr1]], #8 \n"
          "st1 {v2.8b}, [%[outr2]], #8 \n"
          "st1 {v3.8b}, [%[outr3]], #8 \n"
          "bne 1b \n"
          : [inptr0] "+r"(inptr0),
            [inptr1] "+r"(inptr1),
            [inptr2] "+r"(inptr2),
            [inptr3] "+r"(inptr3),
            [outr0] "+r"(outr0),
            [outr1] "+r"(outr1),
            [outr2] "+r"(outr2),
            [outr3] "+r"(outr3),
            [cnt] "+r"(cnt)
          : [vb] "w"(vb), [vg] "w"(vg), [vr] "w"(vr)
          : "cc",
            "memory",
            "v0",
            "v1",
            "v2",
            "v3",
            "v4",
            "v5",
            "v6",
            "v7",
            "v8",
            "v9",
            "v10",
            "v11",
            "v12",
            "v13",
            "v14",
            "v15",
            "v16",
            "v17",
            "v18",
            "v19",
            "v20");
#else
      asm volatile(
          "pld [%[inptr0]]                         @ preload a, 64byte\n"
          "pld [%[inptr0], #128]                         @ preload a, 64byte\n"
          "pld [%[inptr1]]            @ preload a, 64byte\n"
          "pld [%[inptr1], #128]                         @ preload a, 64byte\n"
          "pld [%[inptr2]]            @ preload a, 64byte\n"
          "pld [%[inptr2], #128]                         @ preload a, 64byte\n"
          "pld [%[inptr3]]            @ preload a, 64byte\n"
          "pld [%[inptr3], #128]                         @ preload a, 64byte\n"
          "vld1.8 d0, [%[vb]] \n"
          "vld1.8 d1, [%[vg]] \n"
          "vld1.8 d2, [%[vr]] \n"
          "1: \n"
          "vld3.8 {d3, d4, d5}, [%[inptr0]]! \n"
          "vld3.8 {d6, d7, d8}, [%[inptr1]]! \n"
          "vld3.8 {d9, d10, d11}, [%[inptr2]]! \n"
          "vld3.8 {d12, d13, d14}, [%[inptr3]]! \n"
          // vb
          "vmull.u8 q8, d3, d0 \n"
          "vmull.u8 q9, d6, d0 \n"
          "vmull.u8 q10, d9, d0 \n"
          "vmull.u8 q11, d12, d0 \n"
          // vg
          "vmull.u8 q12, d4, d1 \n"
          "vmull.u8 q13, d7, d1 \n"
          "vmull.u8 q14, d10, d1 \n"
          "vmull.u8 q15, d13, d1 \n"
          // vr
          "vmlal.u8 q8, d5, d2 \n"
          "vmlal.u8 q9, d6, d2 \n"
          "vmlal.u8 q10, d11, d2 \n"
          "vmlal.u8 q11, d14, d2 \n"
          // 16->32
          "vaddl.u16 q2, d24, d16 \n"
          "vaddl.u16 q3, d25, d17 \n"
          "vaddl.u16 q4, d26, d18 \n"
          "vaddl.u16 q5, d27, d19 \n"
          "vaddl.u16 q6, d28, d20 \n"
          "vaddl.u16 q7, d29, d21 \n"
          "vaddl.u16 q8, d30, d22 \n"
          "vaddl.u16 q9, d31, d23 \n"
          // 32->16 q2 >> 7
          "vshrn.u32  d20, q2, #7 \n"
          "vshrn.u32 d21, q3, #7 \n"
          "vshrn.u32 d22, q4, #7 \n"
          "vshrn.u32 d23, q5, #7 \n"
          "vshrn.u32 d24, q6, #7 \n"
          "vshrn.u32 d25, q7, #7 \n"
          "vshrn.u32 d26, q8, #7 \n"
          "vshrn.u32 d27, q8, #7 \n"
          // 16->8
          "vmovn.u16 d4, q10 \n"
          "vmovn.u16 d5, q11 \n"
          "vmovn.u16 d6, q12 \n"
          "vmovn.u16 d7, q13 \n"
          "subs %[cnt], #1 \n"
          // store
          "vst1.8 d4, [%[outr0]]! \n"
          "vst1.8 d5, [%[outr1]]! \n"
          "vst1.8 d6, [%[outr2]]! \n"
          "vst1.8 d7, [%[outr3]]! \n"
          "bne 1b \n"
          : [inptr0] "+r"(inptr0),
            [inptr1] "+r"(inptr1),
            [inptr2] "+r"(inptr2),
            [inptr3] "+r"(inptr3),
            [outr0] "+r"(outr0),
            [outr1] "+r"(outr1),
            [outr2] "+r"(outr2),
            [outr3] "+r"(outr3),
            [cnt] "+r"(cnt)
          : [vb] "r"(vb_array), [vg] "r"(vg_array), [vr] "r"(vr_array)
          : "cc",
            "memory",
            "q0",
            "q1",
            "q2",
            "q3",
            "q4",
            "q5",
            "q6",
            "q7",
            "q8",
            "q9",
            "q10",
            "q11",
            "q12",
            "q13",
            "q14",
            "q15");
#endif
    }
    for (; j < remain_pro; j++) {
      *outr0++ = (inptr0[0] * b + inptr0[1] * g + inptr0[2] * r) >> 7;
      *outr1++ = (inptr1[0] * b + inptr1[1] * g + inptr1[2] * r) >> 7;
      *outr2++ = (inptr2[0] * b + inptr2[1] * g + inptr2[2] * r) >> 7;
      *outr3++ = (inptr3[0] * b + inptr3[1] * g + inptr3[2] * r) >> 7;
      inptr0 += 3;
      inptr1 += 3;
      inptr2 += 3;
      inptr3 += 3;
    }
  }
  for (; i < srch; i++) {
    int j = 0;
    const uint8_t* inptr0 = src + i * win;
    uint8_t* outr0 = dst + i * srcw;
    for (j = 0; j < cnt_pro; j++) {
      uint8x8x3_t y0 = vld3_u8(inptr0);  // d8 = y0y3y6y9.. d9 = y1y4y7...y
      uint16x8_t val0 = vmull_u8(y0.val[0], vb);

      uint16x8_t val0_1 = vmull_u8(y0.val[1], vg);

      val0 = vmlal_u8(val0, y0.val[2], vr);

      uint32x4_t v0_sum0 = vaddl_u16(vget_low_u16(val0_1), vget_low_u16(val0));
      uint32x4_t v0_sum1 =
          vaddl_u16(vget_high_u16(val0_1), vget_high_u16(val0));

      uint16x4_t v0_sum0_16 = vshrn_n_u32(v0_sum0, 7);
      uint16x4_t v0_sum1_16 = vshrn_n_u32(v0_sum1, 7);

      uint16x8_t v0_sum = vcombine_u16(v0_sum0_16, v0_sum1_16);

      uint8x8_t vout0 = vmovn_u16(v0_sum);

      inptr0 += 24;
      vst1_u8(outr0, vout0);
      outr0 += 8;
    }
    for (; j < srcw; j++) {
      *outr0++ = (inptr0[0] * b + inptr0[1] * g + inptr0[2] * r) >> 7;
      inptr0 += 3;
    }
  }
}
/*
采用CV_GRAY2BGR,转换公式B = G = R = Gray
采用CV_GRAY2RGB,转换公式R = G = B = Gray
gray2bgr, gray2rgb
*/
void hwc1_to_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = *src;
      *dst++ = *src;
      *dst++ = *src;
      src++;
    }
  }
}
// bgr2bgra, rgb2rgba
void hwc3_to_hwc4(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = *src++;
      *dst++ = *src++;
      *dst++ = *src++;
      *dst++ = 255;
    }
  }
}
// bgra2bgr, rgba2rgb
void hwc4_to_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = *src++;
      *dst++ = *src++;
      *dst++ = *src++;
      src++;
    }
  }
}
// bgr2rgb, rgb2bgr
void hwc3_trans(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = src[2];  // r
      *dst++ = src[1];  // g
      *dst++ = src[0];  // b
      src += 3;
    }
  }
}
// bgra2rgba, rgba2bgra
void hwc4_trans(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = src[2];  // r
      *dst++ = src[1];  // g
      *dst++ = src[0];  // b
      *dst++ = src[3];  // a
      src += 4;
    }
  }
}
// bgra2rgb, rgba2bgr
void hwc4_trans_hwc3(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = src[2];  // r
      *dst++ = src[1];  // g
      *dst++ = src[0];  // b
      // *dst++ = src[4];//a
      src += 4;
    }
  }
}
// bgr2rgba, rgb2bga
void hwc3_trans_hwc4(const uint8_t* src, uint8_t* dst, int srcw, int srch) {
  for (int i = 0; i < srch; i++) {
    for (int j = 0; j < srcw; j++) {
      *dst++ = src[2];  // r
      *dst++ = src[1];  // g
      *dst++ = src[0];  // b
      *dst++ = 255;     // a
      src += 3;
    }
  }
}
}  // namespace cv
}  // namespace utils
}  // namespace lite
}  // namespace paddle