File size: 87,134 Bytes
be94e5d |
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 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 |
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
// This file is modified from the ficus (https://github.com/vpisarev/ficus/blob/master/lib/NN/OpConv.fx).
// Here is the original license:
/*
This file is a part of ficus language project.
See ficus/LICENSE for the licensing terms
*/
#include "../../precomp.hpp"
#include "convolution.hpp"
#include "conv_block.simd.hpp"
#include "layers/cpu_kernels/conv_block.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
#include <opencv2/core/utils/logger.hpp>
namespace cv { namespace dnn {
enum { VEC_ALIGN = 32}; // Memory alignment.
void convBlock_F32(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int outLen,
const int convMR, const int convNR);
void convBlockMR1_F32(int np, const float* a, const float* b, float *c, const float bias, bool init_c,
const float minval, const float maxval, bool ifMinMaxAct, const int outLen, const int convNR);
#ifdef CONV_ARM_FP16
// Fast convert float 32 to float16
static inline void _cvt32f16f(const float* src, __fp16* dst, int len)
{
int j = 0;
const int VECSZ = 4;
__fp16* dst_FP16 = (__fp16 *)dst;
if (len > VECSZ * 4)
{
const int VECSZ4 = 4 * VECSZ;
for( ; j + VECSZ4 < len; j += VECSZ4)
{
float32x4_t v0 = vld1q_f32(src + j);
float32x4_t v1 = vld1q_f32(src + j + 4);
float32x4_t v2 = vld1q_f32(src + j + 8);
float32x4_t v3 = vld1q_f32(src + j + 12);
vst1q_f16(dst_FP16 + j, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
vst1q_f16(dst_FP16 + j + 8, vcombine_f16(vcvt_f16_f32(v2), vcvt_f16_f32(v3)));
}
}
for( ; j < len; j += VECSZ )
{
if( j > len - VECSZ )
{
if( j == 0 )
break;
j = len - VECSZ;
}
float16x4_t hv = vcvt_f16_f32(vld1q_f32(src + j));
vst1_f16(dst_FP16 + j, hv);
}
for( ; j < len; j++ )
dst[j] = __fp16(src[j]);
}
#endif
float* FastConv::getWeights()
{
return alignPtr(weightsBuf.data(), VEC_ALIGN);
}
float* FastConv::getWeightsWino()
{
return alignPtr(weightsWinoBuf.data(), VEC_ALIGN);
}
hfloat* FastConv::getWeightsFP16()
{
return alignPtr(weightsBuf_FP16.data(), VEC_ALIGN);
}
hfloat* FastConv::getWeightsWinoFP16()
{
return alignPtr(weightsWinoBuf_FP16.data(), VEC_ALIGN);
}
Ptr<FastConv> initFastConv(
InputArray _weightsMat,
float* srcBias,
int ngroups,
int K, int C,
const std::vector<size_t>& kernel_size,
const std::vector<size_t>& strides,
const std::vector<size_t>& dilations,
const std::vector<size_t>& pads_begin,
const std::vector<size_t>& pads_end,
int conv_dim,
const bool _useFP16,
bool useWinograd)
{
Ptr<FastConv> conv = makePtr<FastConv>();
CV_Assert(ngroups > 0 && K > 0 && C > 0 && K % ngroups == 0);
// Weight shape, [K, C, Dk, Hk, Wk] for Conv3D, [K, C, Hk, Wk] for Conv2D, [K, C, Wk] for Conv1D.
int Dk = conv_dim == CONV_3D ? (int)kernel_size[0] : 1;
int Hk = conv_dim == CONV_1D ? 1 : (int)kernel_size[kernel_size.size() - 2];
int Wk = (int)kernel_size.back();
int karea = Wk*Hk*Dk;
conv->pad_front = conv_dim == CONV_3D ? (int)pads_begin[0] : 0;
conv->pad_top = conv_dim == CONV_1D ? 0 : (int)pads_begin[pads_begin.size() - 2];
conv->pad_left = (int)pads_begin.back();
conv->pad_behind = conv_dim == CONV_3D ? (int)pads_end[0] : 0;
conv->pad_bottom = conv_dim == CONV_1D ? 0 : (int)pads_end[pads_end.size() - 2];
conv->pad_right = (int)pads_end.back();
int stride_d = conv_dim == CONV_3D ? (int)strides[0] : 1;
int stride_h = conv_dim == CONV_1D ? 1 : (int)strides[strides.size() - 2];
int stride_w = (int)strides.back();
int dilation_d = conv_dim == CONV_3D ? (int)dilations[0] : 1;
int dilation_h = conv_dim == CONV_1D ? 1 : (int)dilations[dilations.size() - 2];
int dilation_w = (int)dilations.back();
CV_Assert(Dk > 0 && Hk > 0 && Wk > 0);
CV_Assert(stride_d >= 0 && stride_h >= 0 && stride_w > 0);
CV_Assert(dilation_d > 0 && dilation_h > 0 && dilation_w > 0);
conv->K = K; conv->C = C; conv->Hk = Hk; conv->Wk = Wk, conv->Dk = Dk;
conv->stride_d = stride_d;
conv->stride_h = stride_h;
conv->stride_w = stride_w;
conv->dilation_d = dilation_d;
conv->dilation_h = dilation_h;
conv->dilation_w = dilation_w;
conv->conv_dim = conv_dim;
conv->ngroups = ngroups;
bool ifRunDepthWise = ngroups > 1 && ngroups == K && ngroups == C;
bool ifRunDepthWiseRemain = false; // It's for big padding or big kernel or Conv3D depth-wise convolution.
if (ifRunDepthWise)
{
if (conv_dim == CONV_1D)
{
ifRunDepthWise &= Hk == 1 && Wk == 3 && (stride_w == 1 || (stride_w == 2 && dilation_w == 1))
&& max(stride_w, dilation_w) >= conv->pad_left && conv->pad_left <= 1;
}
else if (conv_dim == CONV_2D)
{
ifRunDepthWise &= Hk == 3 && Wk == 3 && ((stride_w == 1) || (stride_w == 2 && dilation_w == 1)) &&
max(stride_w, dilation_w) >= conv->pad_left && max(stride_h, dilation_h) >= conv->pad_top
&& conv->pad_left <= 1 && conv->pad_top <= 1;
}
if (!ifRunDepthWise || conv_dim == CONV_3D)
{
ifRunDepthWise = false;
ifRunDepthWiseRemain = true;
}
}
conv->conv_type = ifRunDepthWise && conv_dim != CONV_3D ? CONV_TYPE_DEPTHWISE :
useWinograd && (conv_dim == CONV_2D && (conv->useSIMD128 || conv->useAVX || conv->useAVX2 || conv->useNEON) &&
Hk == 3 && Wk == 3 && dilation_h == 1 && dilation_w == 1 && stride_h == 1 && stride_w == 1) ?
CONV_TYPE_WINOGRAD3X3 :
(ifRunDepthWiseRemain ? CONV_TYPE_DEPTHWISE_REMAIN : CONV_TYPE_GENERIC);
#if !(CV_NEON || CV_SIMD128 || CV_TRY_AVX || CV_TRY_AVX2)
if (conv->conv_type == CONV_TYPE_WINOGRAD3X3) // Disabel Winograd when CV_NEON, CV_SIMD128 ,CV_TRY_AVX and CV_TRY_AVX2 are not available.
conv->conv_type = CONV_TYPE_GENERIC;
#endif
Mat weightsMat = _weightsMat.getMat();
auto wShape = shape(weightsMat);
const size_t wstep = weightsMat.step1();
conv->useFP16 = false;
#ifdef CONV_ARM_FP16
if (_useFP16 && (conv->conv_type == CONV_TYPE_GENERIC || conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN
|| conv->conv_type == CONV_TYPE_WINOGRAD3X3))
conv->useFP16 = true;
// Runtime FP16 check.
if (conv->useFP16 && !checkHardwareSupport(CPU_NEON_FP16))
{
conv->useFP16 = false;
CV_LOG_ONCE_WARNING(NULL, "DNN: the CPU does not support the instruction set required by FP16, fallback to FP32.");
}
#endif
float *srcWeights = (float *)weightsMat.data;
if (conv->conv_type == CONV_TYPE_DEPTHWISE || conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN)
{
// Handle the Conv1D, Conv2D and Conv3D depth-wise.
// for depth-wise convolutions on NCHW data we just preserve the weights in KCHW layout,
// but add some padding to make the weights array layout more SIMD-friendly
int ksize = karea;
// TODO: simplify the following code with std::copy.
// this code aims to let memory fit with vector size.
int padded_ksize = ((ksize + VEC_ALIGN-1) / VEC_ALIGN) * VEC_ALIGN;
int nweights = C * padded_ksize;
#ifdef CONV_ARM_FP16
if (conv->useFP16)
{
conv->weightsBuf_FP16.resize(nweights + VEC_ALIGN);
auto weightsPtr_FP16 = (__fp16*)conv->getWeightsFP16();
parallel_for_(Range(0, C), [&](const Range& r0){
for(int c = r0.start; c < r0.end; c++)
_cvt32f16f(srcWeights + c*wstep, weightsPtr_FP16 + c*padded_ksize, ksize);
});
}
else
#endif
{
conv->weightsBuf.resize(nweights + VEC_ALIGN);
auto weightsPtr = conv->getWeights();
parallel_for_(Range(0, C), [&](const Range& r0) {
for(int c = r0.start; c < r0.end; c++)
memcpy(weightsPtr + c*padded_ksize, srcWeights + c*wstep, ksize*sizeof(weightsPtr[0]));
});
}
}
else if(conv->conv_type == CONV_TYPE_WINOGRAD3X3) // winograd
{
static const float ktm[8][3] = {
{1.0f, 0.0f, 0.0f},
{-2.0f / 9, -2.0f / 9, -2.0f / 9},
{-2.0f / 9, 2.0f / 9, -2.0f / 9},
{1.0f / 90, 1.0f / 45, 2.0f / 45},
{1.0f / 90, -1.0f / 45, 2.0f / 45},
{32.f/45, 16.f/45, 8.f/45},
{32.f/45, -16.f/45, 8.f/45},
{0.0f, 0.0f, 1.0f}
};
const int CONV_WINO_KBLOCK = 4;
#if CV_TRY_AVX || CV_TRY_AVX2
const int CONV_WINO_ATOM_F32 = (conv->useAVX || conv->useAVX2) ? 8 : 4;
#else
const int CONV_WINO_ATOM_F32 = 4;
#endif
const int CONV_WINO_NATOMS_F32 = CONV_WINO_AREA / CONV_WINO_ATOM_F32; // for AVX2, it is 8, otherwise, it's 16.
#ifdef CONV_ARM_FP16
// FP 16
const int CONV_WINO_ATOM_F16 = CONV_WINO_ATOM_F32 * 2;
const int CONV_WINO_NATOMS_F16 = CONV_WINO_AREA / CONV_WINO_ATOM_F16;
#endif
// the weights are packed as 6-dim tensor:
// ngroups * ceil((K/ngroups)/KBLOCK) * (W*W/ATOM_SIZE) * (C/ngroups) * KBLOCK * ATOM_SIZE,
// where W is the size of Winograd-transformed kernel (8x8),
// ATOM_SIZE is number of lanes in SIMD register (4 for NEON and FP32),
// KBLOCK is some platform-dependent constant dependent on the number of SIMD registers.
int ksize = CONV_WINO_KSIZE * CONV_WINO_KSIZE;
int Cg = C/ngroups;
int Kg = K/ngroups;
int Kg_nblocks = (Kg + CONV_WINO_KBLOCK - 1)/CONV_WINO_KBLOCK;
size_t nweights = ngroups*Kg_nblocks*Cg*CONV_WINO_KBLOCK*CONV_WINO_AREA;
float* wptrWino = nullptr;
#ifdef CONV_ARM_FP16
__fp16* wptrWino_FP16 = nullptr;
if (conv->useFP16)
{
conv->weightsWinoBuf_FP16.resize(nweights + VEC_ALIGN);
wptrWino_FP16 = (__fp16*)conv->getWeightsWinoFP16();
}
else
#endif
{
conv->weightsWinoBuf.resize(nweights + VEC_ALIGN);
wptrWino = conv->getWeightsWino();
}
parallel_for_(Range(0, K), [&](const Range& r0){
float kernelTm[CONV_WINO_AREA];
for (int k = r0.start; k < r0.end; k++)
{
int g = k / Kg;
int k_ = k - g*Kg;
int ki = k_ / CONV_WINO_KBLOCK;
int dk = k_ - ki*CONV_WINO_KBLOCK;
for (int c = 0; c < Cg; c++)
{
// wstep = Hk*Wk*Cg
const float *kernel0 = srcWeights + k * wstep + c * ksize;
// transform kernel, transposed
const float *k0 = kernel0;
const float *k1 = kernel0 + 3;
const float *k2 = kernel0 + 6;
// h
float tmp[8][3];
for (int i = 0; i < 8; i++)
{
tmp[i][0] = k0[0] * ktm[i][0] + k0[1] * ktm[i][1] + k0[2] * ktm[i][2];
tmp[i][1] = k1[0] * ktm[i][0] + k1[1] * ktm[i][1] + k1[2] * ktm[i][2];
tmp[i][2] = k2[0] * ktm[i][0] + k2[1] * ktm[i][1] + k2[2] * ktm[i][2];
}
// v
for (int j = 0; j < 8; j++)
{
float *tmpp = &tmp[j][0];
for (int i = 0; i < 8; i++)
kernelTm[j * 8 + i] = tmpp[0] * ktm[i][0] + tmpp[1] * ktm[i][1] + tmpp[2] * ktm[i][2];
}
// repack the data.
#ifdef CONV_ARM_FP16
if (conv->useFP16)
{
__fp16* wptr = wptrWino_FP16 + (g*Kg_nblocks + ki) * Cg *CONV_WINO_KBLOCK*CONV_WINO_AREA +
(c*CONV_WINO_KBLOCK + dk)*CONV_WINO_ATOM_F16;
for (int i = 0; i < CONV_WINO_NATOMS_F16; i++,
wptr += Cg * CONV_WINO_KBLOCK * CONV_WINO_ATOM_F16)
{
CV_Assert(wptrWino_FP16 <= wptr && wptr + CONV_WINO_ATOM_F16 <= wptrWino_FP16 + nweights);
for (int j = 0; j < CONV_WINO_ATOM_F16; j++)
{
wptr[j] = (__fp16)kernelTm[i * CONV_WINO_ATOM_F16 + j];
}
}
}
else
#endif
{
float* wptr = wptrWino + (g*Kg_nblocks + ki) * Cg *CONV_WINO_KBLOCK*CONV_WINO_AREA +
(c*CONV_WINO_KBLOCK + dk)*CONV_WINO_ATOM_F32;
for (int i = 0; i < CONV_WINO_NATOMS_F32; i++,
wptr += Cg * CONV_WINO_KBLOCK * CONV_WINO_ATOM_F32)
{
CV_Assert(wptrWino <= wptr && wptr + CONV_WINO_ATOM_F32 <= wptrWino + nweights);
memcpy(wptr, kernelTm + i * CONV_WINO_ATOM_F32, CONV_WINO_ATOM_F32*sizeof (wptr[0]));
}
}
}
}
});
}
else if (conv->conv_type == CONV_TYPE_GENERIC)
{
// The weights are packed as
// ngroups x (ceil((K/ngroups)/CONV_MR)*CONV_MR) x (Cg*Hk*Wk*Dk) x CONV_MR tensor
int Kg = K/ngroups, Cg = max(C/ngroups, 1);
int DkHkWkCg = Dk*Hk*Wk*Cg;
int numStripsMR = (Kg + CONV_MR_FP32 - 1) / CONV_MR_FP32;
int Kg_aligned = numStripsMR * CONV_MR_FP32;
size_t nweights = ngroups*Kg_aligned*DkHkWkCg;
float* weightsPtr = nullptr;
#ifdef CONV_ARM_FP16
int numStripsMR_FP16 = (Kg + CONV_MR_FP16 - 1) / CONV_MR_FP16;
int Kg_aligned_FP16 = numStripsMR_FP16 * CONV_MR_FP16;
size_t nweights_FP16 = ngroups * Kg_aligned_FP16 * DkHkWkCg;
__fp16* weightsPtr_FP16 = nullptr;
if (conv->useFP16)
{
conv->weightsBuf_FP16.resize(nweights_FP16 + VEC_ALIGN);
weightsPtr_FP16 = (__fp16*)conv->getWeightsFP16();
}
else
#endif
{
conv->weightsBuf.resize(nweights + VEC_ALIGN);
weightsPtr = conv->getWeights();
}
// Pack the weight.
#ifdef CONV_ARM_FP16
if (conv->useFP16)
{
parallel_for_(Range(0, ngroups * numStripsMR_FP16), [&](const Range& r0){
for (int gsi = r0.start; gsi < r0.end; gsi++)
{
int g = gsi / numStripsMR_FP16;
int si = gsi - g * numStripsMR_FP16;
int startK = si * CONV_MR_FP16;
CV_Assert(startK < Kg_aligned_FP16);
__fp16* packed_wptr = weightsPtr_FP16 + DkHkWkCg * (startK + g * Kg_aligned_FP16);
int dk = Kg - startK < CONV_MR_FP16 ? Kg - startK : CONV_MR_FP16; // check if we need zero padding.
int k_idx = g*Kg + startK;
for(int hwd = 0; hwd < Hk*Wk*Dk; hwd++)
{
for(int c = 0; c < Cg; c++, packed_wptr += CONV_MR_FP16)
{
const float* wptr = srcWeights + wstep * k_idx + c*Hk*Wk*Dk + hwd;
int k = 0;
for(; k < dk; k++, wptr += wstep)
packed_wptr[k] = (__fp16)(*wptr);
for(; k < CONV_MR_FP16; k++)
packed_wptr[k] = (__fp16)0.f;
}
}
}});
}
else
#endif
{
parallel_for_(Range(0, ngroups * numStripsMR), [&](const Range& r0){
for (int gsi = r0.start; gsi < r0.end; gsi++)
{
int g = gsi / numStripsMR;
int si = gsi - g * numStripsMR;
int startK = si * CONV_MR_FP32;
CV_Assert(startK < Kg_aligned);
float* packed_wptr = weightsPtr + DkHkWkCg * (startK + g * Kg_aligned);
int dk = Kg - startK < CONV_MR_FP32 ? Kg - startK : CONV_MR_FP32; // check if we need zero padding.
int k_idx = g*Kg + startK;
for(int hwd = 0; hwd < Hk*Wk*Dk; hwd++)
{
for(int c = 0; c < Cg; c++, packed_wptr += CONV_MR_FP32)
{
const float* wptr = srcWeights + wstep * k_idx + c*Hk*Wk*Dk + hwd;
int k = 0;
for(; k < dk; k++, wptr += wstep)
packed_wptr[k] = *wptr;
for(; k < CONV_MR_FP32; k++)
packed_wptr[k] = 0.f;
}
}
}});
}
}
else
CV_Error(cv::Error::StsUnsupportedFormat, "Unknown convolution type.");
// store bias; append some zero's to make sure that
// we can always read MR elements starting from any valid index
{
int k = 0, nbias = K + VEC_ALIGN;
conv->biasBuf.resize(nbias);
float* biasBufPtr = conv->biasBuf.data();
for(; k < K; k++)
biasBufPtr[k] = srcBias ? srcBias[k] : 0.f;
for(; k < nbias; k++)
biasBufPtr[k] = 0.f;
}
return conv;
}
static inline void packData8(char*& inpbuf, float*& inptrIn, int& in_w, int& x0, int& s0, const int* ofstab,
const int stride_w, const int ksize, const int esz)
{
char * inpbufC = inpbuf + s0 * esz;
float* inptrInC = (float* )inptrIn;
#ifdef CONV_ARM_FP16
__fp16* inpbufC_FP16 = (__fp16 *)inpbufC;
if (esz == sizeof(__fp16))
{
if (stride_w == 1)
{
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
float32x4_t v0 = vld1q_f32(inptrInC + k1);
float32x4_t v1 = vld1q_f32(inptrInC + k1 + 4);
vst1q_f16((__fp16*)inpbufC_FP16 + k * CONV_NR_FP16, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
}
}
else
{
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
float32x4_t v0 = {inptrInC[k1], inptrInC[k1 + stride_w], inptrInC[k1 + 2*stride_w], inptrInC[k1 + 3*stride_w]};
float32x4_t v1 = {inptrInC[k1 + 4*stride_w], inptrInC[k1 + 5*stride_w], inptrInC[k1 + 6*stride_w], inptrInC[k1 + 7*stride_w]};
vst1q_f16((__fp16*)inpbufC_FP16 + k * CONV_NR_FP16, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
}
}
}
else // float 32
#endif
{
CV_Assert(esz == sizeof(float ));
float* inpbufC_FP32 = (float* )inpbufC;
if (stride_w == 1)
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
#if CV_SIMD256
vx_store(inpbufC_FP32 + k*CONV_NR_FP32, vx_load(inptrInC + k1));
#elif CV_SIMD128
v_float32x4 vv0 = v_load(inptrInC + k1);
v_float32x4 vv1 = v_load(inptrInC + k1 + 4);
v_store(inpbufC_FP32 + k*CONV_NR_FP32, vv0);
v_store(inpbufC_FP32 + k*CONV_NR_FP32 + 4, vv1);
#else
float v0 = inptrInC[k1];
float v1 = inptrInC[k1 + 1];
float v2 = inptrInC[k1 + 2];
float v3 = inptrInC[k1 + 3];
float v4 = inptrInC[k1 + 4];
float v5 = inptrInC[k1 + 5];
float v6 = inptrInC[k1 + 6];
float v7 = inptrInC[k1 + 7];
inpbufC_FP32[k*CONV_NR_FP32] = v0;
inpbufC_FP32[k*CONV_NR_FP32+1] = v1;
inpbufC_FP32[k*CONV_NR_FP32+2] = v2;
inpbufC_FP32[k*CONV_NR_FP32+3] = v3;
inpbufC_FP32[k*CONV_NR_FP32+4] = v4;
inpbufC_FP32[k*CONV_NR_FP32+5] = v5;
inpbufC_FP32[k*CONV_NR_FP32+6] = v6;
inpbufC_FP32[k*CONV_NR_FP32+7] = v7;
#endif
}
else
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
float v0 = inptrInC[k1];
float v1 = inptrInC[k1 + stride_w];
float v2 = inptrInC[k1 + 2*stride_w];
float v3 = inptrInC[k1 + 3*stride_w];
float v4 = inptrInC[k1 + 4*stride_w];
float v5 = inptrInC[k1 + 5*stride_w];
float v6 = inptrInC[k1 + 6*stride_w];
float v7 = inptrInC[k1 + 7*stride_w];
inpbufC_FP32[k*CONV_NR_FP32] = v0;
inpbufC_FP32[k*CONV_NR_FP32+1] = v1;
inpbufC_FP32[k*CONV_NR_FP32+2] = v2;
inpbufC_FP32[k*CONV_NR_FP32+3] = v3;
inpbufC_FP32[k*CONV_NR_FP32+4] = v4;
inpbufC_FP32[k*CONV_NR_FP32+5] = v5;
inpbufC_FP32[k*CONV_NR_FP32+6] = v6;
inpbufC_FP32[k*CONV_NR_FP32+7] = v7;
}
}
x0+=7;
s0+=7;
inptrIn += 7*stride_w;
in_w += 7*stride_w;
}
static inline void packData2(char *& inpbuf, float*& inptrIn, int& in_w, int& x0, int& s0, const int* ofstab,
const int stride_w, const int ksize, const int esz)
{
char* inpbufC = inpbuf + s0 * esz;
float* inptrInC = inptrIn;
#ifdef CONV_ARM_FP16
__fp16* inpbufC_FP16 = (__fp16 *)inpbufC;
if (esz == sizeof(__fp16))
{
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
float v0 = inptrInC[k1];
float v1 = inptrInC[k1 + stride_w];
inpbufC_FP16[k*CONV_NR_FP16] = (__fp16)v0;
inpbufC_FP16[k*CONV_NR_FP16+1] = (__fp16)v1;
}
} else
#endif
{
float * inpbufC_FP32 = (float *)inpbufC;
for (int k = 0; k < ksize; k++)
{
int k1 = ofstab[k];
float v0 = inptrInC[k1];
float v1 = inptrInC[k1 + stride_w];
inpbufC_FP32[k*CONV_NR_FP32] = v0;
inpbufC_FP32[k*CONV_NR_FP32+1] = v1;
}
}
x0++;
s0++;
inptrIn += stride_w;
in_w += stride_w;
}
static inline void packInputData(char* inpbuf_task, float* inp, const int* ofstab, const int* dhwTab, int zyx0, int zyx_limit,
int ksize, int stride_d, int stride_h, int stride_w, int pad_front, int pad_top, int pad_left,
int Dk, int Hk, int Wk, int dilation_d, int dilation_h, int dilation_w, int Di, int Hi, int Wi,
int H0, int W0, int Cg, int stripesize, int inp_plane_ofs, int inp_planesize, int conv_dim, int conv_type,
const int CONV_NR, const int esz, bool fast_1x1, bool useFP16)
{
for (int stripe = 0; zyx0 < zyx_limit; stripe++, zyx0 += CONV_NR)
{
char *inpbuf = inpbuf_task + stripe * stripesize * esz;
float *inptr = inp + inp_plane_ofs;
/*
1. pack the data. Copy the HkxWk CONV_NR-wide slices from
each feature plane of the input tensor to the input buffer.
*/
if (fast_1x1)
{
int slice_len = zyx_limit - zyx0;
bool partial = slice_len < CONV_NR;
const int CONV_NR_esz = CONV_NR * esz;
// Superfast branch for 1x1 convolutions with sy=sx=1.
// in this case each feature plane can be safely treated
// as 1D array, and we just extract next portion
// of CONV_NR elements from each feature plane and
// put it together.
inptr += zyx0;
if (!partial)
{
// Make special branch where memcpy() is called with a constant buffer size.
// Compilers will likely unroll this loop properly.
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
_cvt32f16f(inptr, (__fp16 *)inpbuf, CONV_NR);
}
else
#endif
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
memcpy(inpbuf, inptr, CONV_NR_esz);
}
else
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
{
_cvt32f16f(inptr, (__fp16 *)inpbuf, slice_len);
}
}
else
#endif
for (int c = 0; c < Cg; c++, inptr += inp_planesize, inpbuf += CONV_NR_esz)
{
memcpy(inpbuf, inptr, slice_len * esz);
}
}
}
else if (conv_type == CONV_TYPE_DEPTHWISE_REMAIN)
{
CV_Assert(Cg == 1);
const int HW0 = H0 * W0;
const int HWi = Hi * Wi;
int slice_len = std::min(zyx_limit - zyx0, CONV_NR);
// here some non-continuous sub-row of the row will not be
// filled from the tensor; we need to make sure that the uncovered
// elements are explicitly set to 0's. the easiest way is to
// set all the elements to 0's before the loop.
memset(inpbuf, 0, stripesize * esz);
int z0 = zyx0 / HW0, yx0 = zyx0 - z0 * HW0;
int y0 = yx0 / W0, x0 = yx0 - y0 * W0;
if (conv_dim == CONV_1D)
{
for (int slice_i = 0; slice_i < slice_len; y0++, x0=0)
{
int delta = std::min(slice_len - slice_i, W0 - x0);
int x1 = x0 + delta;
int in_w = x0 * stride_w - pad_left;
float* inptrIn = inptr + in_w;
int s0 = slice_i;
for (; x0 < x1; x0++, s0++, inptrIn += stride_w, in_w += stride_w)
{
// Pack 8
if (x0 + 8 <= x1 && 0 <= in_w &&
in_w + stride_w*8 <= Wi - (Wk-1)*dilation_w)
{
packData8(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else if (x0 + 2 <= x1 && 0 <= in_w &&
in_w + stride_w*2 <= Wi - (Wk-1)*dilation_w)
{
packData2(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else
{
int w0 = std::max(0, (-in_w + dilation_w-1)/dilation_w);
int w1 = std::min(Wk, (Wi - in_w + dilation_w-1)/dilation_w);
const float* inptrInC = inptrIn;
#ifdef CONV_ARM_FP16
if (useFP16)
{
__fp16* inpbufC = (__fp16 *)inpbuf + s0;
for (int w = w0; w < w1; w++)
{
int imgofs = w*dilation_w;
inpbufC[w*CONV_NR] = (__fp16)inptrInC[imgofs];
}
}
else
#endif
{
float* inpbufC = (float *)inpbuf + s0;
for (int w = w0; w < w1; w++)
{
int imgofs = w*dilation_w;
inpbufC[w*CONV_NR] = inptrInC[imgofs];
}
}
}
}
slice_i += delta;
}
}
else if (conv_dim == CONV_2D)
{
for (int slice_i = 0; slice_i < slice_len; y0++, x0=0)
{
int delta = std::min(slice_len - slice_i, W0 - x0);
int x1 = x0 + delta;
int in_h = y0 * stride_h - pad_top;
int in_w = x0 * stride_w - pad_left;
float* inptrIn = inptr + in_h*Wi + in_w;
bool ok_i = 0 <= in_h && in_h < Hi - (Hk-1)*dilation_h;
int h0 = std::max(0, (-in_h + dilation_h-1)/dilation_h);
int h1 = std::min(Hk, (Hi - in_h + dilation_h-1)/dilation_h);
int s0 = slice_i;
for (; x0 < x1; x0++, s0++, inptrIn += stride_w, in_w += stride_w)
{
// Pack 8
if (ok_i && x0 + 8 <= x1 && 0 <= in_w &&
in_w + stride_w*8 <= Wi - (Wk-1)*dilation_w)
{
packData8(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else if (ok_i && x0 + 2 <= x1 && 0 <= in_w &&
in_w + stride_w*2 <= Wi - (Wk-1)*dilation_w)
{
packData2(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else
{
int w0 = std::max(0, (-in_w + dilation_w-1)/dilation_w);
int w1 = std::min(Wk, (Wi - in_w + dilation_w-1)/dilation_w);
const float* inptrInC = inptrIn;
#ifdef CONV_ARM_FP16
if (useFP16)
{
__fp16* inpbufC = (__fp16 *)inpbuf + s0;
for (int h = h0; h < h1; h++)
{
for (int w = w0; w < w1; w++)
{
int imgofs = h*(dilation_h*Wi) + w*dilation_w;
inpbufC[(h*Wk + w)*CONV_NR] = (__fp16)inptrInC[imgofs];
}
}
}
else
#endif
{
float* inpbufC = (float *)inpbuf + s0;
for (int h = h0; h < h1; h++)
{
for (int w = w0; w < w1; w++)
{
int imgofs = h*(dilation_h*Wi) + w*dilation_w;
inpbufC[(h*Wk + w)*CONV_NR] = inptrInC[imgofs];
}
}
}
}
}
slice_i += delta;
}
}
else if (conv_dim == CONV_3D)
{
for (int slice_i = 0; slice_i < slice_len; z0 += (y0+1)/H0, y0 = (y0+1)%H0, x0=0)
{
int delta = std::min(slice_len - slice_i, W0 - x0);
int x1 = x0 + delta;
int in_d = z0 * stride_d - pad_front;
int in_h = y0 * stride_h - pad_top;
int in_w = x0 * stride_w - pad_left;
float* inptrIn = inptr + in_d*HWi + in_h*Wi + in_w;
int d0 = std::max(0, (-in_d + dilation_d - 1) / dilation_d);
int d1 = std::min(Dk, (Di - in_d + dilation_d - 1) / dilation_d);
bool ok_i = 0 <= in_d && in_d < Di - (Dk-1)*dilation_d &&
0 <= in_h && in_h < Hi - (Hk-1)*dilation_h;
int h0 = std::max(0, (-in_h + dilation_h-1)/dilation_h);
int h1 = std::min(Hk, (Hi - in_h + dilation_h-1)/dilation_h);
int s0 = slice_i;
for (; x0 < x1; x0++, s0++, inptrIn += stride_w, in_w += stride_w)
{
// Pack 8
if (ok_i && x0 + 8 <= x1 && 0 <= in_w &&
in_w + stride_w*8 <= Wi - (Wk-1)*dilation_w)
{
packData8(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else if (ok_i && x0 + 2 <= x1 && 0 <= in_w &&
in_w + stride_w*2 <= Wi - (Wk-1)*dilation_w)
{
packData2(inpbuf, inptrIn, in_w, x0, s0, ofstab, stride_w, ksize, esz);
}
else
{
int w0 = std::max(0, (-in_w + dilation_w-1)/dilation_w);
int w1 = std::min(Wk, (Wi - in_w + dilation_w-1)/dilation_w);
const float* inptrInC = inptrIn;
#ifdef CONV_ARM_FP16
if (useFP16)
{
__fp16* inpbufC = (__fp16* )inpbuf + s0;
for ( int d = d0; d < d1; d++)
{
for (int h = h0; h < h1; h++)
{
for (int w = w0; w < w1; w++)
{
int imgofs = d*dilation_d*HWi + h*(dilation_h*Wi) + w*dilation_w;
inpbufC[((d*Hk + h)*Wk + w)*CONV_NR] = (__fp16)inptrInC[imgofs];
}
}
}
}
else
#endif
{
float* inpbufC = (float* )inpbuf + s0;
for ( int d = d0; d < d1; d++)
{
for (int h = h0; h < h1; h++)
{
for (int w = w0; w < w1; w++)
{
int imgofs = d*dilation_d*HWi + h*(dilation_h*Wi) + w*dilation_w;
inpbufC[((d*Hk + h)*Wk + w)*CONV_NR] = inptrInC[imgofs];
}
}
}
}
}
}
slice_i += delta;
}
}
}
else
{
const int HW0 = H0 * W0;
const int HWi = Hi * Wi;
int z0_ = zyx0 / HW0, yx0 = zyx0 - z0_ * HW0;
int y0_ = yx0 / W0, x0_ = yx0 - y0_ * W0;
for (int k = 0; k < ksize; k++)
{
int dz = dhwTab[k * 3], dy = dhwTab[k * 3 + 1], dx = dhwTab[k * 3 + 2];
int i = 0, z0 = z0_, y0 = y0_, x0 = x0_;
for (; i < CONV_NR;)
{
float* inpbuf_ki = (float* )inpbuf + k * CONV_NR * Cg + i;
#ifdef CONV_ARM_FP16
__fp16 * inpbuf_ki_FP16 = (__fp16 *)inpbuf + k * CONV_NR * Cg + i;
#endif
int zi = z0 * stride_d + dz - pad_front;
int yi = y0 * stride_h + dy - pad_top;
int xi = x0 * stride_w + dx - pad_left;
if ((unsigned) zi < (unsigned) Di && (unsigned) yi < (unsigned) Hi &&
(unsigned) xi < (unsigned) Wi)
{
const float *inptr_ki = inptr + zi * HWi + yi * Wi + xi;
if (i + 8 <= CONV_NR && x0 + 8 <= W0 && xi + stride_w * 8 <= Wi)
{
if (stride_w == 1)
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
{
float32x4_t v0 = vld1q_f32(inptr_ki);
float32x4_t v1 = vld1q_f32(inptr_ki + 4);
vst1q_f16((__fp16* )inpbuf_ki_FP16, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
}
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
{
float t0 = inptr_ki[0], t1 = inptr_ki[1];
float t2 = inptr_ki[2], t3 = inptr_ki[3];
float t4 = inptr_ki[4], t5 = inptr_ki[5];
float t6 = inptr_ki[6], t7 = inptr_ki[7];
inpbuf_ki[0] = t0;
inpbuf_ki[1] = t1;
inpbuf_ki[2] = t2;
inpbuf_ki[3] = t3;
inpbuf_ki[4] = t4;
inpbuf_ki[5] = t5;
inpbuf_ki[6] = t6;
inpbuf_ki[7] = t7;
}
}
else if (stride_w == 2)
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
{
float32x4_t v0 = {inptr_ki[0], inptr_ki[2], inptr_ki[4], inptr_ki[6]};
float32x4_t v1 = {inptr_ki[8], inptr_ki[10], inptr_ki[12], inptr_ki[14]};
vst1q_f16((__fp16* )inpbuf_ki_FP16, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
}
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
{
float t0 = inptr_ki[0], t1 = inptr_ki[2];
float t2 = inptr_ki[4], t3 = inptr_ki[6];
float t4 = inptr_ki[8], t5 = inptr_ki[10];
float t6 = inptr_ki[12], t7 = inptr_ki[14];
inpbuf_ki[0] = t0;
inpbuf_ki[1] = t1;
inpbuf_ki[2] = t2;
inpbuf_ki[3] = t3;
inpbuf_ki[4] = t4;
inpbuf_ki[5] = t5;
inpbuf_ki[6] = t6;
inpbuf_ki[7] = t7;
}
}
else
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
{
float32x4_t v0 = {inptr_ki[0], inptr_ki[stride_w], inptr_ki[stride_w * 2], inptr_ki[stride_w * 3]};
float32x4_t v1 = {inptr_ki[stride_w * 4], inptr_ki[stride_w * 5], inptr_ki[stride_w * 6], inptr_ki[stride_w * 7]};
vst1q_f16((__fp16* )inpbuf_ki_FP16, vcombine_f16(vcvt_f16_f32(v0), vcvt_f16_f32(v1)));
}
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
{
float t0 = inptr_ki[0], t1 = inptr_ki[stride_w];
float t2 = inptr_ki[stride_w * 2], t3 = inptr_ki[stride_w * 3];
float t4 = inptr_ki[stride_w * 4], t5 = inptr_ki[stride_w * 5];
float t6 = inptr_ki[stride_w * 6], t7 = inptr_ki[stride_w * 7];
inpbuf_ki[0] = t0;
inpbuf_ki[1] = t1;
inpbuf_ki[2] = t2;
inpbuf_ki[3] = t3;
inpbuf_ki[4] = t4;
inpbuf_ki[5] = t5;
inpbuf_ki[6] = t6;
inpbuf_ki[7] = t7;
}
}
i += 8;
x0 += 8;
}
else if (i + 4 <= CONV_NR && x0 + 4 <= W0 && xi + stride_w * 4 <= Wi)
{
if (stride_w == 1)
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
{
float32x4_t v0 = vld1q_f32(inptr_ki);
vst1_f16((__fp16* )inpbuf_ki_FP16, vcvt_f16_f32(v0));
}
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
{
float t0 = inptr_ki[0], t1 = inptr_ki[1];
float t2 = inptr_ki[2], t3 = inptr_ki[3];
inpbuf_ki[0] = t0;
inpbuf_ki[1] = t1;
inpbuf_ki[2] = t2;
inpbuf_ki[3] = t3;
}
}
else
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
{
float32x4_t v0 = {inptr_ki[0], inptr_ki[stride_w], inptr_ki[stride_w * 2], inptr_ki[stride_w * 3]};
vst1_f16((__fp16* )inpbuf_ki_FP16, vcvt_f16_f32(v0));
}
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
{
float t0 = inptr_ki[0], t1 = inptr_ki[stride_w];
float t2 = inptr_ki[stride_w * 2], t3 = inptr_ki[stride_w * 3];
inpbuf_ki[0] = t0;
inpbuf_ki[1] = t1;
inpbuf_ki[2] = t2;
inpbuf_ki[3] = t3;
}
}
i += 4;
x0 += 4;
}
else
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR, inptr_ki += inp_planesize)
inpbuf_ki_FP16[0] = (__fp16)(*inptr_ki);
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR, inptr_ki += inp_planesize)
*inpbuf_ki = *inptr_ki;
i++;
x0++;
}
}
else
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
for (int c = 0; c < Cg; c++, inpbuf_ki_FP16 += CONV_NR)
inpbuf_ki_FP16[0] = (__fp16)0.f;
}
else
#endif
for (int c = 0; c < Cg; c++, inpbuf_ki += CONV_NR)
inpbuf_ki[0] = 0.f;
i++;
x0++;
}
int mask = x0 >= W0;
y0 += mask;
x0 &= mask - 1;
mask = y0 >= H0; // Only Conv 3D need jump at z0 dimension
if (mask && conv_dim != CONV_3D)
break;
z0 += mask;
y0 &= mask - 1;
}
}
}
}
}
void runFastConv(InputArray _input, OutputArray _output, const Ptr<FastConv>& conv, int ntasks,
const Ptr<ActivationLayer>& actLayer, const std::vector<float>& reluslope, bool fusedAdd)
{
Mat input = _input.getMat();
Mat output = _output.getMat();
int conv_dim = conv->conv_dim;
CV_Assert_N(input.dims == output.dims,
input.size[0] == output.size[0],
conv->C == input.size[1],
conv->K == output.size[1],
input.type() == output.type(),
input.isContinuous(),
output.isContinuous());
const bool useFP16 = conv->useFP16;
Mat fusedAddMat;
if (fusedAdd)
{
CV_Assert(conv->conv_dim != CONV_3D && "Conv3D does not support Conv+Add fusion optimization!");
fusedAddMat = _output.getMat();
}
if (conv->conv_type == CONV_TYPE_DEPTHWISE)
{
// Depthwise-Convolution layer should not be followed by Add layer.
CV_Assert((conv_dim == CONV_1D || conv_dim == CONV_2D) && !useFP16);
return runDepthwise(input, output, conv, actLayer.get(), reluslope, fusedAdd);
}
MatShape inputShape = shape(input);
MatShape outputShape = shape(output);
CV_Assert(inputShape.size() == outputShape.size());
ActivationLayer* activ = nullptr;
float minval = -FLT_MAX, maxval = FLT_MAX;
bool ifMinMaxAct = false;
if (actLayer)
{
Ptr<ReLULayer> activ_relu = actLayer.dynamicCast<ReLULayer>();
Ptr<ReLU6Layer> activ_relu6 = actLayer.dynamicCast<ReLU6Layer>();
if (!activ_relu.empty())
{
if (activ_relu->negativeSlope == 0.0f)
{
minval = 0.0f;
ifMinMaxAct = true;
activ = nullptr;
}
else // Leaky ReLU
{
activ = actLayer.get();
}
}
else if (!activ_relu6.empty())
{
minval = activ_relu6->minValue;
maxval = activ_relu6->maxValue;
ifMinMaxAct = true;
activ = nullptr;
}
else
activ = actLayer.get();
}
else
activ = nullptr;
if (conv->conv_type == CONV_TYPE_WINOGRAD3X3) // winograd
{
CV_Assert((!conv->weightsWinoBuf.empty() || !conv->weightsWinoBuf_FP16.empty()) && input.dims == 4 && conv_dim == CONV_2D);
if (runWinograd63(input, fusedAddMat, output, conv, ntasks, minval, maxval, activ, ifMinMaxAct))
return;
}
int N = inputShape[0], C = inputShape[1];
// input shape: [N, C, D, H, W] for Conv3D, [N, C, H, W] for Conv2D, [N, C, W] for Conv1D.
int Di = conv_dim == CONV_3D ? inputShape[2] : 1;
int Hi = conv_dim == CONV_1D ? 1 : inputShape[inputShape.size() - 2];
int Wi = inputShape[inputShape.size() - 1];
int ngroups = conv->ngroups;
int K = conv->K, Dk = conv->Dk, Hk = conv->Hk, Wk = conv->Wk;
int D0 = conv_dim == CONV_3D ? outputShape[2] : 1;
int H0 = conv_dim == CONV_1D ? 1 : outputShape[outputShape.size() - 2];
int W0 = outputShape[outputShape.size() - 1];
int Cg = C/ngroups, Kg = K/ngroups;
const size_t inp_planesize = (size_t)Di*Hi*Wi;
const size_t out_planesize = (size_t)D0*H0*W0;
int pad_front = conv->pad_front;
int pad_top = conv->pad_top;
int pad_left = conv->pad_left;
int stride_d = conv->stride_d, stride_h = conv->stride_h, stride_w = conv->stride_w;
int dilation_d = conv->dilation_d, dilation_h = conv->dilation_h, dilation_w = conv->dilation_w;
int ksize = Dk*Hk*Wk;
bool fast_1x1 = ksize == 1 && stride_d == 1 && stride_w == 1 && stride_h == 1
&& pad_front == 0 && pad_left == 0 && pad_top == 0;
int DkHkWkCg = Dk*Hk*Wk*Cg;
std::vector<int> ofstab_(Hk*Wk*Dk*4, 0);
int* ofstab = ofstab_.data();
int* dhwTab = ofstab + Hk*Wk*Dk;
int padded_ksize = ((ksize + VEC_ALIGN-1) / VEC_ALIGN) * VEC_ALIGN;
if (conv_dim == CONV_1D)
{
for( int w = 0; w < Wk; w++)
{
int dw = w*dilation_w;
dhwTab[w*3+2] = dw;
ofstab[w] = dw;
}
}
else if (conv_dim == CONV_2D)
{
for (int h = 0; h < Hk; h++)
for( int w = 0; w < Wk; w++)
{
int k = h*Wk + w;
int dh = h*dilation_h, dw = w*dilation_w;
dhwTab[k*3+1] = dh;
dhwTab[k*3+2] = dw;
ofstab[k] = dh*Wi + dw;
}
}
else
{
for (int d = 0; d < Dk; d++)
for (int h = 0; h < Hk; h++)
{
for (int w = 0; w < Wk; w++)
{
int k = d*Hk*Wk + h*Wk + w;
int dd = d*dilation_d, dh = h*dilation_h, dw = w*dilation_w;
dhwTab[k*3] = dd;
dhwTab[k*3+1] = dh;
dhwTab[k*3+2] = dw;
ofstab[k] = dd*Hi*Wi + dh*Wi + dw;
}
}
}
int CONV_NR = CONV_NR_FP32;
int CONV_MR = CONV_MR_FP32;
int esz = sizeof(float );
#ifdef CONV_ARM_FP16
if (useFP16)
{
// works at FP 16.
CONV_NR = CONV_NR_FP16;
CONV_MR = CONV_MR_FP16;
esz = sizeof(__fp16);
}
#endif
int MAX_STRIPES = conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN ? 1 : (56 + CONV_NR - 1)/CONV_NR;
// Friendly to L1 cache
const int K_BLOCK_SIZE = conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN ? 1 : 32;
const int C_BLOCK_SIZE = 256;
int Kg_nblocks = (Kg + CONV_MR-1)/CONV_MR;
int Kg_aligned = Kg_nblocks * CONV_MR;
int stripes_per_plane0 = ((int)out_planesize + CONV_NR - 1) / CONV_NR;
int stripes_per_plane = stripes_per_plane0;
if (stripes_per_plane < ntasks * 4 || conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN)
{
MAX_STRIPES = 1;
stripes_per_plane = 1;
}
else
Kg_nblocks = 1;
bool separateIm2col = (fast_1x1 || stripes_per_plane == 1) && conv->conv_type != CONV_TYPE_DEPTHWISE_REMAIN;
int Kstripes = Kg_nblocks * stripes_per_plane;
int nsubtasks = N * ngroups * Kstripes;
size_t stripesize = alignSize(CONV_NR * ksize * Cg, VEC_ALIGN);
size_t cbufsize = alignSize(CONV_NR * K_BLOCK_SIZE * MAX_STRIPES, VEC_ALIGN);
size_t taskbufsize = cbufsize * sizeof(float );
if (!separateIm2col)
taskbufsize += MAX_STRIPES * stripesize * esz;
size_t totalbufsize_base = taskbufsize * ntasks;
size_t totalbufsize = totalbufsize_base;
if (separateIm2col)
totalbufsize += N * ngroups * stripes_per_plane0 * stripesize * esz;
AutoBuffer<char> inpbuf_all_;
char* inpbuf_all = nullptr;
inpbuf_all_.allocate(totalbufsize + VEC_ALIGN * sizeof(float ));
inpbuf_all = alignPtr(inpbuf_all_.data(), (int)(VEC_ALIGN * sizeof(float )));
char* inpbuf_all_0 = inpbuf_all + totalbufsize_base;
float* inp = input.ptr<float>();
float* out = output.ptr<float>();
float* fusedAddPtr0 = fusedAddMat.empty() ? 0 : fusedAddMat.ptr<float>();
// In the case of 1x1 convolution we first reorder the whole input tensor.
// In general, im2row results in Hk*Wk-x unrolling factor
// (e.g. 3*3=9x unrolling for 3x3 convolution), thus for 1x1 convolution
// the reordered tensor will take as much space as the original tensor.
if (separateIm2col)
{
// the optional phase 1. im2row
parallel_for_(Range(0, ntasks), [&](const Range& r0) {
for (int task_id = r0.start; task_id < r0.end; task_id++)
{
if (fast_1x1)
{
int nc0 = task_id*N*C/ntasks, nc1 = (task_id+1)*N*C/ntasks, dc = 0;
for (; nc0 < nc1; nc0 += dc)
{
int n = nc0/C, c0 = nc0 - n*C;
int g = c0 / Cg;
c0 -= g*Cg;
dc = Cg - c0 <= nc1 - nc0 ? Cg - c0 : nc1 - nc0;
float * inptr_ = inp + (size_t)nc0*inp_planesize;
char* inpbuf_ = inpbuf_all_0 + ((n*ngroups + g)*stripes_per_plane0*stripesize + c0*CONV_NR)*esz;
packInputData(inpbuf_, inptr_, ofstab, dhwTab, 0, out_planesize, ksize, stride_d, stride_h,
stride_w, pad_front, pad_top, pad_left, Dk, Hk, Wk, dilation_d, dilation_h, dilation_w,
Di, Hi, Wi, H0, W0, dc, stripesize, 0, inp_planesize, conv->conv_dim,
conv->conv_type, CONV_NR, esz, fast_1x1, useFP16);
}
}
else
{
const int allTasks = N * ngroups * stripes_per_plane0;
int ngs0 = task_id*allTasks/ntasks, ngs1 = (task_id+1)*allTasks/ntasks, ds = 0;
for (; ngs0 < ngs1; ngs0 += ds)
{
int n = ngs0 / (ngroups * stripes_per_plane0), gs0 = ngs0 - n*ngroups*stripes_per_plane0;
int g = gs0 / stripes_per_plane0, s0 = gs0 - g*stripes_per_plane0;
ds = stripes_per_plane0 - s0 <= ngs1 - ngs0 ? stripes_per_plane0 - s0 : ngs1 - ngs0;
int zyx = s0 * CONV_NR;
int zyx_limit = (s0 + ds) * CONV_NR < out_planesize ? (s0 + ds) * CONV_NR : out_planesize;
float * inptr_ = inp + (size_t)(n * ngroups + g) * Cg * inp_planesize;
char* inpbuf_ = inpbuf_all_0 + ((n * ngroups + g) * stripes_per_plane0 * stripesize + s0 * stripesize) * esz;
packInputData(inpbuf_, inptr_, ofstab, dhwTab, zyx, zyx_limit, ksize, stride_d, stride_h,
stride_w, pad_front, pad_top, pad_left, Dk, Hk, Wk, dilation_d, dilation_h, dilation_w,
Di, Hi, Wi, H0, W0, Cg, stripesize, 0, inp_planesize, conv->conv_dim,
conv->conv_type, CONV_NR, esz, fast_1x1, useFP16);
}
}
}
});
}
// Compute
parallel_for_(Range(0, ntasks), [&](const Range& r0) {
for (int task_id = r0.start; task_id < r0.end; task_id++)
{
float * cbuf_task = (float *)(inpbuf_all + taskbufsize * task_id);
char * inpbuf_task = (char*)(cbuf_task + cbufsize);
int ngs0 = (int)((size_t)nsubtasks * task_id / ntasks);
int ngs1 = (int)((size_t)nsubtasks * (task_id+1) / ntasks);
for (int subtask = ngs0; subtask < ngs1; )
{
int ng = subtask / Kstripes;
int kzyx0 = subtask - ng * Kstripes;
int kzyx1 = kzyx0 + (ngs1 - subtask);
int n = ng / ngroups, g = ng % ngroups; // ng - n * ngroups;
size_t inp_plane_ofs = (size_t)(n * ngroups + g) * Cg * inp_planesize;
kzyx1 = kzyx1 <= Kstripes ? kzyx1 : Kstripes;
subtask += kzyx1 - kzyx0;
int k0, k1;
int zyx0, zyx_limit, zyx_block_limit = 0;
if (stripes_per_plane == 1 || conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN)
{
k0 = kzyx0 * CONV_MR;
k1 = kzyx1 * CONV_MR;
k1 = k1 <= Kg ? k1 : Kg;
zyx0 = 0;
zyx_limit = (int)out_planesize;
}
else
{
k0 = 0;
k1 = Kg;
zyx0 = kzyx0 * CONV_NR;
zyx_limit = kzyx1 * CONV_NR;
zyx_limit = zyx_limit < out_planesize ? zyx_limit : (int)out_planesize;
}
for (; zyx0 < zyx_limit; zyx0 = zyx_block_limit)
{
// step 1. extract part of input tensor and represent it in zigzag form
zyx_block_limit = zyx0 + CONV_NR * MAX_STRIPES;
zyx_block_limit = zyx_block_limit < zyx_limit ? zyx_block_limit : zyx_limit;
int nstripes = (zyx_block_limit - zyx0 + CONV_NR - 1) / CONV_NR;
CV_Assert(nstripes <= MAX_STRIPES);
if (!separateIm2col)
{
packInputData(inpbuf_task, inp, ofstab, dhwTab, zyx0, zyx_block_limit, ksize, stride_d, stride_h,
stride_w, pad_front, pad_top, pad_left, Dk, Hk, Wk, dilation_d, dilation_h, dilation_w,
Di, Hi, Wi, H0, W0, Cg, stripesize, inp_plane_ofs, inp_planesize, conv->conv_dim,
conv->conv_type, CONV_NR, esz, fast_1x1, useFP16);
}
char *weights = nullptr;
#ifdef CONV_ARM_FP16
if (useFP16)
{
CV_Assert(!conv->weightsBuf_FP16.empty());
weights = (char *)conv->getWeightsFP16();
}
else
#endif
{
CV_Assert(!conv->weightsBuf.empty());
weights = (char *)conv->getWeights();
}
// optional branch, only for depth-wise convolution which was implemented by generic convolution.
// In this case, CONV_MR is 1, and CONV_NR remains the same.
if (conv->conv_type == CONV_TYPE_DEPTHWISE_REMAIN)
{
CV_Assert(weights);
size_t outofs = (n * ngroups + g) * out_planesize + zyx0;
float *cptr0 = cbuf_task;
weights += g * padded_ksize * esz;
int out_width = zyx_block_limit - zyx0;
float *outptr = out + outofs;
const float biasVal = *(conv->biasBuf.data() + g);
const char *inptr_ = separateIm2col ? inpbuf_all_0 + (ng * stripes_per_plane0 + zyx0 / CONV_NR) * stripesize * esz :
inpbuf_task;
for (int stripe = 0; stripe < nstripes; stripe++)
{
const char *inptr = inptr_ + stripe * stripesize * esz;
const int outLen = std::min(out_width - stripe * CONV_NR, CONV_NR);
bool ifBuffer = outLen < CONV_NR;
float *cptr = outptr + stripe * CONV_NR;
if (ifBuffer)
{
memcpy(cptr0, cptr, outLen * sizeof(float ));
cptr = cptr0;
}
#if CV_NEON && CV_NEON_AARCH64
if (conv->useNEON)
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
opt_NEON_FP16::convBlockMR1_F16(DkHkWkCg, weights, inptr, cptr, biasVal, fusedAdd, minval, maxval, ifMinMaxAct, outLen, CONV_NR);
}
else
#endif
opt_NEON::convBlockMR1_F32(DkHkWkCg, (const float *)weights, (const float *)inptr, cptr, biasVal, fusedAdd, minval, maxval, ifMinMaxAct, outLen, CONV_NR);
}
else
#endif
convBlockMR1_F32(DkHkWkCg, (const float *)weights, (const float *)inptr, cptr, biasVal, fusedAdd, minval, maxval, ifMinMaxAct, outLen, CONV_NR);
if (ifBuffer)
{
memcpy(outptr + stripe * CONV_NR, cptr, outLen * sizeof(float ));
}
}
if (activ)
activ->forwardSlice(outptr, outptr, out_width, out_planesize, g, g + 1);
continue;
}
CV_Assert(weights);
weights += g * Kg_aligned * DkHkWkCg * esz;
const float *biasptr = conv->biasBuf.data() + Kg * g;
int ldc = nstripes * CONV_NR;
// 2. do convolution, compute Kg x (zyx_block_limit - zyx0) part of the output tensor
int out_width = zyx_block_limit - zyx0;
for (int k0_block = k0; k0_block < k1; k0_block += K_BLOCK_SIZE)
{
int k1_block = k0_block + K_BLOCK_SIZE < k1 ? k0_block + K_BLOCK_SIZE : k1;
for (int c0 = 0; c0 < DkHkWkCg; c0 += C_BLOCK_SIZE)
{
int c1 = c0 + C_BLOCK_SIZE < DkHkWkCg ? c0 + C_BLOCK_SIZE : DkHkWkCg;
const char *inptr = separateIm2col ? inpbuf_all_0 + (ng * stripes_per_plane0 + zyx0 / CONV_NR) * stripesize * esz :
inpbuf_task;
inptr += (c0 * CONV_NR) * esz;
for (int stripe = 0; stripe < nstripes; stripe++, inptr += stripesize * esz)
{
const int outLen = std::min(out_width - stripe * CONV_NR, CONV_NR);
char *wptr = weights + (k0_block * DkHkWkCg + c0 * CONV_MR) * esz;
float *cptr = cbuf_task + stripe * CONV_NR;
hfloat* cptr_f16 = (hfloat*)cbuf_task + stripe*CONV_NR;
for (int k = k0_block; k < k1_block; k += CONV_MR,
wptr += DkHkWkCg * CONV_MR * esz, cptr += CONV_MR * ldc, cptr_f16 += CONV_MR * ldc)
{
#if CV_TRY_AVX2
if (conv->useAVX2)
opt_AVX2::convBlock_F32(c1 - c0, (const float *)wptr, (const float *)inptr, cptr, ldc, c0 == 0, outLen, CONV_MR, CONV_NR);
else
#endif
#if CV_TRY_AVX
if (conv->useAVX)
opt_AVX::convBlock_F32(c1 - c0, (const float *)wptr, (const float *)inptr, cptr, ldc, c0 == 0, outLen, CONV_MR, CONV_NR);
else
#endif
#if CV_NEON
if (conv->useNEON)
{
#ifdef CONV_ARM_FP16
if (useFP16)
{
opt_NEON_FP16::convBlock_F16(c1 - c0, wptr, inptr, (char *)cptr_f16, ldc, c0 == 0, outLen, CONV_MR, CONV_NR);
}
else
#endif
opt_NEON::convBlock_F32(c1 - c0, (const float *)wptr, (const float *)inptr, cptr, ldc, c0 == 0, outLen, CONV_MR, CONV_NR);
}
else
#endif
// The possible outLen range is 24 or 8~1.
convBlock_F32(c1 - c0, (const float *)wptr, (const float *)inptr, cptr, ldc, c0 == 0, outLen, CONV_MR, CONV_NR);
}
}
}
size_t outofs = ((n * ngroups + g) * Kg + k0_block) * out_planesize + zyx0;
const float *cptr = cbuf_task;
const hfloat *cptr_fp16 = (const hfloat *)cbuf_task;
float *outptr = out + outofs;
const float *pbptr = fusedAddPtr0 ? fusedAddPtr0 + outofs : 0;
for (int k = k0_block; k < k1_block; k++,
cptr += ldc, cptr_fp16 += ldc, outptr += out_planesize,
pbptr += (pbptr ? out_planesize : 0))
{
float biasval = biasptr[k];
int j = 0;
#ifdef CONV_ARM_FP16
if (useFP16)
{
float32x4_t vbias = vdupq_n_f32(biasval);
float32x4_t vmax = vdupq_n_f32(maxval);
float32x4_t vmin = vdupq_n_f32(minval);
if (pbptr)
{
for (; j + 7 < out_width; j += 8)
{
float32x4_t v0 = vcvt_f32_f16(vld1_f16((const __fp16 *)cptr_fp16 + j)) + vbias;
float32x4_t v1 = vcvt_f32_f16(vld1_f16((const __fp16 *)cptr_fp16 + + j + 4)) + vbias;
v0 += vld1q_f32(pbptr + j);
v1 += vld1q_f32(pbptr + j + 4);
if (ifMinMaxAct)
{
v0 = vminq_f32(vmaxq_f32(v0, vmin), vmax);
v1 = vminq_f32(vmaxq_f32(v1, vmin), vmax);
}
vst1q_f32(outptr + j, v0);
vst1q_f32(outptr + j + 4, v1);
}
}
else
{
for (; j + 7 < out_width; j += 8)
{
float32x4_t v0 = vcvt_f32_f16(vld1_f16((const __fp16 *)cptr_fp16 + j)) + vbias;
float32x4_t v1 = vcvt_f32_f16(vld1_f16((const __fp16 *)cptr_fp16 + j + 4)) + vbias;
if (ifMinMaxAct)
{
v0 = vminq_f32(vmaxq_f32(v0, vmin), vmax);
v1 = vminq_f32(vmaxq_f32(v1, vmin), vmax);
}
vst1q_f32(outptr + j, v0);
vst1q_f32(outptr + j + 4, v1);
}
}
if (pbptr)
{
for (; j < out_width; j++)
{
float v = (float )cptr_fp16[j] + biasval;
v += pbptr[j];
if (ifMinMaxAct)
v = std::min(std::max(v, minval), maxval);
outptr[j] = v;
}
}
else
{
for (; j < out_width; j++)
{
float v = (float )cptr_fp16[j] + biasval;
if (ifMinMaxAct)
v = std::min(std::max(v, minval), maxval);
outptr[j] = v;
}
}
}
else
#endif
{
#if CV_SIMD128
v_float32x4 vbias = v_setall_f32(biasval);
v_float32x4 vmax = v_setall_f32(maxval);
v_float32x4 vmin = v_setall_f32(minval);
if (pbptr)
{
for (; j + 7 < out_width; j += 8)
{
v_float32x4 v0 = v_add(v_load(cptr + j), vbias);
v_float32x4 v1 = v_add(v_load(cptr + j + 4), vbias);
v0 = v_add(v0, v_load(pbptr + j));
v1 = v_add(v1, v_load(pbptr + j + 4));
if (ifMinMaxAct)
{
v0 = v_min(v_max(v0, vmin), vmax);
v1 = v_min(v_max(v1, vmin), vmax);
}
v_store(outptr + j, v0);
v_store(outptr + j + 4, v1);
}
}
else
{
for (; j + 7 < out_width; j += 8)
{
v_float32x4 v0 = v_add(v_load(cptr + j), vbias);
v_float32x4 v1 = v_add(v_load(cptr + j + 4), vbias);
if (ifMinMaxAct)
{
v0 = v_min(v_max(v0, vmin), vmax);
v1 = v_min(v_max(v1, vmin), vmax);
}
v_store(outptr + j, v0);
v_store(outptr + j + 4, v1);
}
}
#endif
if (pbptr)
{
for (; j < out_width; j++)
{
float v = cptr[j] + biasval;
v += pbptr[j];
if (ifMinMaxAct)
v = std::min(std::max(v, minval), maxval);
outptr[j] = v;
}
}
else
{
for (; j < out_width; j++)
{
float v = cptr[j] + biasval;
if (ifMinMaxAct)
v = std::min(std::max(v, minval), maxval);
outptr[j] = v;
}
}
}
if (activ)
activ->forwardSlice(outptr, outptr, out_width, out_planesize, Kg * g + k, Kg * g + k + 1);
}
}
}
}
}
});
}
/****************************************************************************************\
SIMD and no-SIMD code for convBlock
\****************************************************************************************/
static inline void convBlockMR1NoSIMD(int np, const float* a, const float* b, float *c, const float bias, bool init_c,
const float minval, const float maxval, bool ifMinMaxAct, const int outLen, const int convNR)
{
std::vector<float> cbuffer(outLen, 0);
float* cbuf = cbuffer.data();
for( int p = 0; p < np; p++ )
{
float ai = a[p];
for( int j = 0; j < outLen; j++ )
cbuf[j] += b[convNR*p + j] * ai;
}
if (init_c)
{
for(int j = 0; j < outLen; j++)
{
c[j] += cbuf[j] + bias;
if (ifMinMaxAct)
c[j] = std::min(std::max(c[j], minval), maxval);
}
}
else
{
for(int j = 0; j < outLen; j++)
{
c[j] = cbuf[j] + bias;
if (ifMinMaxAct)
c[j] = std::min(std::max(c[j], minval), maxval);
}
}
}
#if CV_SIMD128
static inline void convBlockMR1x24(int np, const float* a, const float* b, float *c, const float bias, bool init_c,
const float minval, const float maxval, bool ifMinMaxAct, const int convNR)
{
CV_Assert(convNR == 24);
v_float32x4 c0 = v_setall_f32(bias), c1 = c0, c2 = c0;
v_float32x4 c3 = c0, c4 = c0, c5 = c0;
for (int p = 0; p < np; p++, a++, b += convNR)
{
v_float32x4 a0 = v_setall_f32(a[0]);
v_float32x4 b0 = v_load(b), b1 = v_load(b + 4), b2 = v_load(b + 8);
v_float32x4 b3 = v_load(b + 12), b4 = v_load(b + 16), b5 = v_load(b + 20);
c0 = v_fma(b0, a0, c0);
c1 = v_fma(b1, a0, c1);
c2 = v_fma(b2, a0, c2);
c3 = v_fma(b3, a0, c3);
c4 = v_fma(b4, a0, c4);
c5 = v_fma(b5, a0, c5);
}
if (init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + 4));
c2 = v_add(c2, v_load(c + 8));
c3 = v_add(c3, v_load(c + 12));
c4 = v_add(c4, v_load(c + 16));
c5 = v_add(c5, v_load(c + 20));
}
if (ifMinMaxAct)
{
v_float32x4 vmax = v_setall_f32(maxval), vmin = v_setall_f32(minval);
c0 = v_min(v_max(c0, vmin), vmax);
c1 = v_min(v_max(c1, vmin), vmax);
c2 = v_min(v_max(c2, vmin), vmax);
c3 = v_min(v_max(c3, vmin), vmax);
c4 = v_min(v_max(c4, vmin), vmax);
c5 = v_min(v_max(c5, vmin), vmax);
}
v_store(c, c0);
v_store(c + 4, c1);
v_store(c + 8, c2);
v_store(c + 12, c3);
v_store(c + 16, c4);
v_store(c + 20, c5);
}
static inline void convBlockMR1x12(int np, const float* a, const float* b, float *c, const float bias, bool init_c,
const float minval, const float maxval, bool ifMinMaxAct, const int convNR)
{
CV_Assert(convNR == 12);
v_float32x4 c0 = v_setall_f32(bias), c1 = c0, c2 = c0;
for (int p = 0; p < np; p++, a++, b += convNR)
{
v_float32x4 a0 = v_setall_f32(a[0]);
v_float32x4 b0 = v_load(b), b1 = v_load(b + 4), b2 = v_load(b + 8);
c0 = v_fma(b0, a0, c0);
c1 = v_fma(b1, a0, c1);
c2 = v_fma(b2, a0, c2);
}
if (init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + 4));
c2 = v_add(c2, v_load(c + 8));
}
if (ifMinMaxAct)
{
v_float32x4 vmax = v_setall_f32(maxval), vmin = v_setall_f32(minval);
c0 = v_min(v_max(c0, vmin), vmax);
c1 = v_min(v_max(c1, vmin), vmax);
c2 = v_min(v_max(c2, vmin), vmax);
}
v_store(c, c0);
v_store(c + 4, c1);
v_store(c + 8, c2);
}
#endif
void convBlockMR1_F32(int np, const float* a, const float* b, float *c, const float bias, bool init_c,
const float minval, const float maxval, bool ifMinMaxAct, const int outLen, const int convNR)
{
#if CV_SIMD128
// The outLen represents the valid output value in CONV_NR length.
// When outLen is very small, we use the no-SIMD branch.
const int convNRby3 = convNR/3;
if (outLen > convNRby3)
{
if (convNR == 24)
convBlockMR1x24(np, a, b, c, bias, init_c, minval, maxval, ifMinMaxAct, convNR);
else if (convNR == 12)
convBlockMR1x12(np, a, b, c, bias, init_c, minval, maxval, ifMinMaxAct, convNR);
else
convBlockMR1NoSIMD(np, a, b, c, bias, init_c, minval, maxval, ifMinMaxAct, outLen, convNR);
}
else
convBlockMR1NoSIMD(np, a, b, c, bias, init_c, minval, maxval, ifMinMaxAct, outLen, convNR);
#else
convBlockMR1NoSIMD(np, a, b, c, bias, init_c, minval, maxval, ifMinMaxAct, outLen, convNR);
#endif
}
#if CV_SIMD128
static inline void convBlock4x24(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int convMR, const int convNR)
{
v_float32x4 c0 = v_setzero_f32(), c1 = c0, c2 = c0, c3 = c0, c4 = c0, c5 = c0;
v_float32x4 c6 = v_setzero_f32(), c7 = c6, c8 = c6, c9 = c6, c10 = c6, c11 = c6;
v_float32x4 c12 = v_setzero_f32(), c13 = c12, c14 = c12, c15 = c12, c16 = c12, c17 = c12;
v_float32x4 c18 = v_setzero_f32(), c19 = c18, c20 = c18, c21 = c18, c22 = c18, c23 = c18;
for (int p = 0; p < np; p++, a += convMR, b += convNR)
{
v_float32x4 a0 = v_setall_f32(a[0]);
v_float32x4 b0 = v_load(b), b1 = v_load(b + 4), b2 = v_load(b + 8);
v_float32x4 b3 = v_load(b + 12), b4 = v_load(b + 16), b5 = v_load(b + 20);
c0 = v_fma(b0, a0, c0);
c1 = v_fma(b1, a0, c1);
c2 = v_fma(b2, a0, c2);
c3 = v_fma(b3, a0, c3);
c4 = v_fma(b4, a0, c4);
c5 = v_fma(b5, a0, c5);
a0 = v_setall_f32(a[1]);
c6 = v_fma(b0, a0, c6);
c7 = v_fma(b1, a0, c7);
c8 = v_fma(b2, a0, c8);
c9 = v_fma(b3, a0, c9);
c10 = v_fma(b4, a0, c10);
c11 = v_fma(b5, a0, c11);
a0 = v_setall_f32(a[2]);
c12 = v_fma(b0, a0, c12);
c13 = v_fma(b1, a0, c13);
c14 = v_fma(b2, a0, c14);
c15 = v_fma(b3, a0, c15);
c16 = v_fma(b4, a0, c16);
c17 = v_fma(b5, a0, c17);
a0 = v_setall_f32(a[3]);
c18 = v_fma(b0, a0, c18);
c19 = v_fma(b1, a0, c19);
c20 = v_fma(b2, a0, c20);
c21 = v_fma(b3, a0, c21);
c22 = v_fma(b4, a0, c22);
c23 = v_fma(b5, a0, c23);
}
if (!init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + 4));
c2 = v_add(c2, v_load(c + 8));
c3 = v_add(c3, v_load(c + 12));
c4 = v_add(c4, v_load(c + 16));
c5 = v_add(c5, v_load(c + 20));
c6 = v_add(c6 , v_load(c + ldc));
c7 = v_add(c7 , v_load(c + ldc + 4));
c8 = v_add(c8 , v_load(c + ldc + 8));
c9 = v_add(c9 , v_load(c + ldc + 12));
c10 = v_add(c10, v_load(c + ldc + 16));
c11 = v_add(c11, v_load(c + ldc + 20));
c12 = v_add(c12, v_load(c + ldc*2));
c13 = v_add(c13, v_load(c + ldc*2 + 4));
c14 = v_add(c14, v_load(c + ldc*2 + 8));
c15 = v_add(c15, v_load(c + ldc*2 + 12));
c16 = v_add(c16, v_load(c + ldc*2 + 16));
c17 = v_add(c17, v_load(c + ldc*2 + 20));
c18 = v_add(c18, v_load(c + ldc*3));
c19 = v_add(c19, v_load(c + ldc*3 + 4));
c20 = v_add(c20, v_load(c + ldc*3 + 8));
c21 = v_add(c21, v_load(c + ldc*3 + 12));
c22 = v_add(c22, v_load(c + ldc*3 + 16));
c23 = v_add(c23, v_load(c + ldc*3 + 20));
}
v_store(c, c0);
v_store(c + 4, c1);
v_store(c + 8, c2);
v_store(c + 12, c3);
v_store(c + 16, c4);
v_store(c + 20, c5);
v_store(c + ldc, c6);
v_store(c + ldc + 4, c7);
v_store(c + ldc + 8, c8);
v_store(c + ldc + 12, c9);
v_store(c + ldc + 16, c10);
v_store(c + ldc + 20, c11);
v_store(c + ldc * 2, c12);
v_store(c + ldc * 2 + 4, c13);
v_store(c + ldc * 2 + 8, c14);
v_store(c + ldc * 2 + 12, c15);
v_store(c + ldc * 2 + 16, c16);
v_store(c + ldc * 2 + 20, c17);
v_store(c + ldc * 3, c18);
v_store(c + ldc * 3 + 4, c19);
v_store(c + ldc * 3 + 8, c20);
v_store(c + ldc * 3 + 12, c21);
v_store(c + ldc * 3 + 16, c22);
v_store(c + ldc * 3 + 20, c23);
}
static inline void convBlock4x8(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int convMR, const int convNR)
{
CV_Assert(convNR >= 4);
v_float32x4 c0 = v_setzero_f32(), c1 = c0, c2 = c0, c3 = c0;
v_float32x4 c4 = c0, c5 = c0, c6 = c0, c7 = c0;
for (int p = 0; p < np; p++, a += convMR, b += convNR)
{
v_float32x4 a0 = v_setall_f32(a[0]);
v_float32x4 a1 = v_setall_f32(a[1]);
v_float32x4 a2 = v_setall_f32(a[2]);
v_float32x4 a3 = v_setall_f32(a[3]);
v_float32x4 b0 = v_load(b), b1 = v_load(b + 4);
c0 = v_fma(b0, a0, c0);
c1 = v_fma(b1, a0, c1);
c2 = v_fma(b0, a1, c2);
c3 = v_fma(b1, a1, c3);
c4 = v_fma(b0, a2, c4);
c5 = v_fma(b1, a2, c5);
c6 = v_fma(b0, a3, c6);
c7 = v_fma(b1, a3, c7);
}
if (!init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + 4));
c2 = v_add(c2, v_load(c + ldc));
c3 = v_add(c3, v_load(c + ldc + 4));
c4 = v_add(c4, v_load(c + ldc*2));
c5 = v_add(c5, v_load(c + ldc*2 + 4));
c6 = v_add(c6, v_load(c + ldc*3));
c7 = v_add(c7, v_load(c + ldc*3 + 4));
}
v_store(c, c0);
v_store(c + 4, c1);
v_store(c + ldc, c2);
v_store(c + ldc + 4, c3);
v_store(c + ldc * 2, c4);
v_store(c + ldc * 2 + 4, c5);
v_store(c + ldc * 3, c6);
v_store(c + ldc * 3 + 4, c7);
}
static inline void convBlock4x4(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int convMR, const int convNR)
{
CV_Assert(convNR >= 4);
v_float32x4 c0 = v_setzero_f32(), c1 = c0, c2 = c0, c3 = c0;
for (int p = 0; p < np; p++, a += convMR, b += convNR)
{
v_float32x4 a0 = v_setall_f32(a[0]);
v_float32x4 a1 = v_setall_f32(a[1]);
v_float32x4 a2 = v_setall_f32(a[2]);
v_float32x4 a3 = v_setall_f32(a[3]);
v_float32x4 b0 = v_load(b);
c0 = v_fma(b0, a0, c0);
c1 = v_fma(b0, a1, c1);
c2 = v_fma(b0, a2, c2);
c3 = v_fma(b0, a3, c3);
}
if (!init_c)
{
c0 = v_add(c0, v_load(c));
c1 = v_add(c1, v_load(c + ldc));
c2 = v_add(c2, v_load(c + ldc*2));
c3 = v_add(c3, v_load(c + ldc*3));
}
v_store(c, c0);
v_store(c + ldc, c1);
v_store(c + ldc * 2, c2);
v_store(c + ldc * 3, c3);
}
#endif
static inline void convBlockNoSIMD(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int outLen,
const int convMR, const int convNR)
{
std::vector<float> cbuffer(convMR * outLen, 0);
float* cbuf = cbuffer.data();
for( int p = 0; p < np; p++ )
{
for( int i = 0; i < convMR; i++ )
{
float ai = a[convMR*p + i];
for( int j = 0; j < outLen; j++ )
cbuf[i * outLen+j] += b[convNR*p + j] * ai;
}
}
if (!init_c)
{
for(int i = 0; i < convMR; i++)
{
for(int j = 0; j < outLen; j++)
c[i*ldc + j] += cbuf[i*outLen + j];
}
}
else
{
for(int i = 0; i < convMR; i++)
{
for(int j = 0; j < outLen; j++)
c[i*ldc + j] = cbuf[i*outLen + j];
}
}
}
void convBlock_F32(int np, const float* a, const float* b, float* c, int ldc, bool init_c, const int outLen,
const int convMR, const int convNR)
{
// The possible outLen range is [24, 8~1].
#if CV_SIMD128
CV_Assert(convMR == 4);
if (outLen > 8 && convNR == 24)
{
convBlock4x24(np, a, b, c, ldc, init_c, convMR, convNR);
return;
}
if (outLen <= 8 && outLen > 4)
{
convBlock4x8(np, a, b, c, ldc, init_c, convMR, convNR);
return;
}
if (outLen <= 4 && outLen > 1)
{
convBlock4x4(np, a, b, c, ldc, init_c, convMR, convNR);
return;
}
convBlockNoSIMD(np, a, b, c, ldc, init_c, outLen, convMR, convNR);
#else
convBlockNoSIMD(np, a, b, c, ldc, init_c, outLen, convMR, convNR);
#endif
}
}} // namespace cv::dnn
|