File size: 112,489 Bytes
b7b614e | 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 | /*
* Copyright (C) 2010-2021 Arm Limited or its affiliates. All rights reserved.
*
* SPDX-License-Identifier: Apache-2.0
*
* 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
*
* 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.
*/
/* ----------------------------------------------------------------------
* Project: CMSIS NN Library
* Title: arm_nnfunctions.h
* Description: Public header file for CMSIS NN Library
*
* $Date: 19 January 2021
* $Revision: V.6.5.3
*
* Target Processor: Cortex-M CPUs
* -------------------------------------------------------------------- */
/**
\mainpage CMSIS NN Software Library
*
* Introduction
* ------------
*
* This user manual describes the CMSIS NN software library,
* a collection of efficient neural network kernels developed to maximize the
* performance and minimize the memory footprint of neural networks on Cortex-M processor cores.
*
* The library is divided into a number of functions each covering a specific category:
* - Convolution Functions
* - Activation Functions
* - Fully-connected Layer Functions
* - SVDF Layer Functions
* - Pooling Functions
* - Softmax Functions
* - Basic math Functions
*
* The library has separate functions for operating on different weight and activation data
* types including 8-bit integers (q7_t) and 16-bit integers (q15_t). The descrition of the
* kernels are included in the function description. The implementation details are also
* described in this paper [1].
*
* Function Classification
* --------
* The functions can be classified into two segments
* - Legacy functions supporting ARM's internal symmetric quantization(8 bits).
* - Functions that support TensorFlow Lite framework with symmetric quantization(8 bits).
*
* The legacy functions can be identified with their suffix of _q7 or _q15 and are no new development is done there.
* The article in [2] describes in detail how to run a network using the legacy functions.
*
* The functions supporting TensorFlow Lite framework is identified by the _s8 suffix and can be invoked from TFL
* micro. The functions are bit exact to TensorFlow Lite. Refer to the TensorFlow's documentation in [3] on how to run
* a TensorFlow Lite model using optimized CMSIS-NN kernels.
*
* Block Diagram
* --------
* \image html CMSIS-NN-OVERVIEW.PNG
*
* Examples
* --------
*
* The library ships with a number of examples which demonstrate how to use the library functions.
*
* Pre-processor Macros
* ------------
*
* Each library project have different pre-processor macros.
*
* - ARM_MATH_DSP:
*
* Define macro ARM_MATH_DSP, If the silicon supports DSP instructions(DSP extension).
*
* - ARM_MATH_MVEI:
*
* Define macro ARM_MATH_MVEI, If the silicon supports M-Profile Vector Extension.
* - ARM_MATH_AUTOVECTORIZE
* Used in conjucture with ARM_MATH_MVEI to let the compiler auto vectorize for the functions that uses inline
* assembly. It does not affect functions that use C or intrinsics.
* - ARM_MATH_BIG_ENDIAN:
*
* Define macro ARM_MATH_BIG_ENDIAN to build the library for big endian targets. This is supported only for the legacy
* functions i.e, functions targetted at TensorFlow Lite do not support big endianness. By default library builds for
* little endian targets.
*
* - ARM_NN_TRUNCATE:
*
* Define macro ARM_NN_TRUNCATE to use floor instead of round-to-the-nearest-int for the computation.
*
*
* Copyright Notice
* ------------
*
* Copyright (C) 2010-2019 Arm Limited. All rights reserved.
*
* [1] CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs https://arxiv.org/abs/1801.06601
*
* [2] Converting a Neural Network for Arm Cortex-M with CMSIS-NN
*
https://developer.arm.com/solutions/machine-learning-on-arm/developer-material/how-to-guides/converting-a-neural-network-for-arm-cortex-m-with-cmsis-nn/single-page
* [3] https://www.tensorflow.org/lite/microcontrollers/library
*
* [4] https://github.com/ARM-software/CMSIS_5/tree/develop/CMSIS/NN#legacy-vs-tfl-micro-compliant-apis
*/
/**
* @defgroup groupNN Neural Network Functions
* A collection of functions to perform basic operations for neural network layers. Functions with a _s8 suffix support
* TensorFlow Lite framework.
*/
#ifndef _ARM_NNFUNCTIONS_H
#define _ARM_NNFUNCTIONS_H
#include "edge-impulse-sdk/CMSIS/DSP/Include/dsp/none.h"
#include "edge-impulse-sdk/CMSIS/DSP/Include/arm_math_types.h"
#include "arm_nn_types.h"
#define USE_INTRINSIC
//#define ARM_NN_TRUNCATE /* This config the rounding model to floor or round to the nearest int */
#ifdef __cplusplus
extern "C"
{
#endif
/**
* @brief Struct for specifying activation function types
*
*/
typedef enum
{
ARM_SIGMOID = 0,
/**< Sigmoid activation function */
ARM_TANH = 1,
/**< Tanh activation function */
} arm_nn_activation_type;
/**
* @defgroup NNConv Convolution Functions
*
* Collection of convolution, depthwise convolution functions and their variants.
*
* The convolution is implemented in 2 steps: im2col and GEMM
*
* im2col is a process of converting each patch of image data into
* a column. After im2col, the convolution is computed as matrix-matrix
* multiplication.
*
* To reduce the memory footprint, the im2col is performed partially.
* Each iteration, only a few column (i.e., patches) are generated and
* computed with GEMM kernels similar to CMSIS-DSP arm_mat_mult functions.
*
*/
/**
* @brief s8 convolution layer wrapper function with the main purpose to call the optimal kernel available in
cmsis-nn
* to perform the convolution.
*
* @param[in, out] ctx Function context that contains the additional buffer if required by the function.
arm_convolve_wrapper_s8_get_buffer_size will return the buffer_size if required
* @param[in] conv_params Convolution parameters (e.g. strides, dilations, pads,...).
* Range of conv_params->input_offset : [-127, 128]
* Range of conv_params->output_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, HK, WK, C_IN] where HK and WK are the
* spatial filter dimensions
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [N, H, W, C_OUT]
* @param[out] output_data Output data pointer. Data type: int8
*
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> if argument constraints fail. or,
* <code>ARM_MATH_SUCCESS</code> on successful completion.
*
*/
arm_status arm_convolve_wrapper_s8(const cmsis_nn_context *ctx,
const cmsis_nn_conv_params *conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for arm_convolve_wrapper_s8
*
* @param[in] conv_params Convolution parameters (e.g. strides, dilations, pads,...).
* Range of conv_params->input_offset : [-127, 128]
* Range of conv_params->output_offset : [-128, 127]
* @param[in] input_dims Input (activation) dimensions. Format: [N, H, W, C_IN]
* @param[in] filter_dims Filter dimensions. Format: [C_OUT, HK, WK, C_IN] where HK and WK are the spatial
* filter dimensions
* @param[in] output_dims Output tensor dimensions. Format: [N, H, W, C_OUT]
*
* @return The function returns required buffer size(bytes)
*
*/
int32_t arm_convolve_wrapper_s8_get_buffer_size(const cmsis_nn_conv_params *conv_params,
const cmsis_nn_dims *input_dims,
const cmsis_nn_dims *filter_dims,
const cmsis_nn_dims *output_dims);
/**
* @brief Basic s8 convolution function
* @param[in, out] ctx Function context that contains the additional buffer if required by the function.
arm_convolve_s8_get_buffer_size will return the buffer_size if required
* @param[in] conv_params Convolution parameters (e.g. strides, dilations, pads,...).
* Range of conv_params->input_offset : [-127, 128]
* Range of conv_params->output_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, HK, WK, C_IN] where HK and WK are the
* spatial filter dimensions
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Optional bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [N, H, W, C_OUT]
* @param[out] output_data Output data pointer. Data type: int8
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
* @details
* 1. Supported framework: TensorFlow Lite micro
* 2. q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
* 3. Additional memory is required for optimization. Refer to argument 'ctx' for details.
*
*/
arm_status arm_convolve_s8(const cmsis_nn_context *ctx,
const cmsis_nn_conv_params *conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for s8 convolution function
*
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, HK, WK, C_IN] where HK and WK
* are the spatial filter dimensions
* @return The function returns required buffer size(bytes)
*
*/
int32_t arm_convolve_s8_get_buffer_size(const cmsis_nn_dims *input_dims, const cmsis_nn_dims *filter_dims);
/**
* @brief Basic Q7 convolution function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_convolve_HWC_q7_basic(const q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Basic Q7 convolution function (non-square shape)
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in_x input tensor dimension x
* @param[in] dim_im_in_y input tensor dimension y
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel_x filter kernel size x
* @param[in] dim_kernel_y filter kernel size y
* @param[in] padding_x padding size x
* @param[in] padding_y padding size y
* @param[in] stride_x convolution stride x
* @param[in] stride_y convolution stride y
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out_x output tensor dimension x
* @param[in] dim_im_out_y output tensor dimension y
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*/
arm_status arm_convolve_HWC_q7_basic_nonsquare(const q7_t *Im_in,
const uint16_t dim_im_in_x,
const uint16_t dim_im_in_y,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel_x,
const uint16_t dim_kernel_y,
const uint16_t padding_x,
const uint16_t padding_y,
const uint16_t stride_x,
const uint16_t stride_y,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out_x,
const uint16_t dim_im_out_y,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Basic Q15 convolution function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_convolve_HWC_q15_basic(const q15_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q15_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q15_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q15_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast Q7 convolution function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 4
* ch_im_out is multiple of 2
*/
arm_status arm_convolve_HWC_q7_fast(const q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast Q7 convolution function (non-sqaure shape)
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in_x input tensor dimension x
* @param[in] dim_im_in_y input tensor dimension y
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel_x filter kernel size x
* @param[in] dim_kernel_y filter kernel size y
* @param[in] padding_x padding size x
* @param[in] padding_y padding size y
* @param[in] stride_x convolution stride x
* @param[in] stride_y convolution stride y
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out_x output tensor dimension x
* @param[in] dim_im_out_y output tensor dimension y
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 4
* ch_im_out is multiple of 2
*/
arm_status arm_convolve_HWC_q7_fast_nonsquare(const q7_t *Im_in,
const uint16_t dim_im_in_x,
const uint16_t dim_im_in_y,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel_x,
const uint16_t dim_kernel_y,
const uint16_t padding_x,
const uint16_t padding_y,
const uint16_t stride_x,
const uint16_t stride_y,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out_x,
const uint16_t dim_im_out_y,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast Q7 version of 1x1 convolution (non-sqaure shape)
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in_x input tensor dimension x
* @param[in] dim_im_in_y input tensor dimension y
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel_x filter kernel size x
* @param[in] dim_kernel_y filter kernel size y
* @param[in] padding_x padding size x
* @param[in] padding_y padding size y
* @param[in] stride_x convolution stride x
* @param[in] stride_y convolution stride y
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out_x output tensor dimension x
* @param[in] dim_im_out_y output tensor dimension y
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> if argument constraints fail. or,
* <code>ARM_MATH_SUCCESS</code> on successful completion.
*
* This function implement convolution with 1x1 kernel size (i.e., dim_kernel_x=1
* and dim_kernel_y=1). It can be used for
* second half of MobileNets after depthwise separable convolution.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 4
* ch_im_out is multiple of 2
*/
arm_status arm_convolve_1x1_HWC_q7_fast_nonsquare(const q7_t *Im_in,
const uint16_t dim_im_in_x,
const uint16_t dim_im_in_y,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel_x,
const uint16_t dim_kernel_y,
const uint16_t padding_x,
const uint16_t padding_y,
const uint16_t stride_x,
const uint16_t stride_y,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out_x,
const uint16_t dim_im_out_y,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast s8 version for 1x1 convolution (non-square shape)
*
* @param[in, out] ctx Function context that contains the additional buffer if required by the function.
arm_convolve_1x1_s8_fast_get_buffer_size will return the buffer_size if required
* @param[in] conv_params Convolution parameters (e.g. strides, dilations, pads,...).
* Range of conv_params->input_offset : [-127, 128]
* Range of conv_params->output_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, 1, 1, C_IN]
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Optional bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [N, H, W, C_OUT]
* @param[out] output_data Output data pointer. Data type: int8
*
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> if argument constraints fail. or,
* <code>ARM_MATH_SUCCESS</code> on successful completion.
*
* @details
* - Supported framework : TensorFlow Lite Micro
* - The following constrains on the arguments apply
* -# input_dims->c is a multiple of 4
* -# conv_params->padding.w = conv_params->padding.h = 0
* -# conv_params->stride.w = conv_params->stride.h = 1
*
*/
arm_status arm_convolve_1x1_s8_fast(const cmsis_nn_context *ctx,
const cmsis_nn_conv_params *conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for arm_convolve_1x1_s8_fast
*
* @param[in] input_dims Input (activation) dimensions
* @return The function returns the required buffer size in bytes
*
*/
int32_t arm_convolve_1x1_s8_fast_get_buffer_size(const cmsis_nn_dims *input_dims);
/**
* @brief 1xn convolution
*
* @param[in, out] ctx Function context that contains the additional buffer if required by the function.
arm_convolve_1_x_n_s8_get_buffer_size will return the buffer_size if required
* @param[in] conv_params Convolution parameters (e.g. strides, dilations, pads,...).
* Range of conv_params->input_offset : [-127, 128]
* Range of conv_params->output_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, 1, WK, C_IN] where WK is the horizontal
* spatial filter dimension
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Optional bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [N, H, W, C_OUT]
* @param[out] output_data Output data pointer. Data type: int8
*
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> if argument constraints fail. or,
* <code>ARM_MATH_SUCCESS</code> on successful completion.
*
* @details
* - Supported framework : TensorFlow Lite Micro
* - The following constrains on the arguments apply
* -# input_dims->n equals 1
* -# ouput_dims->w is a multiple of 4
* -# Explicit constraints(since it is for 1xN convolution)
* -## input_dims->h equals 1
* -## output_dims->h equals 1
* -## filter_dims->h equals 1
*@todo Remove constraint on output_dims->w to make the function generic.
*
*/
arm_status arm_convolve_1_x_n_s8(const cmsis_nn_context *ctx,
const cmsis_nn_conv_params *conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required additional buffer size for 1xn convolution
*
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* @param[in] filter_dims Filter tensor dimensions. Format: [C_OUT, 1, WK, C_IN] where WK is the
* horizontal spatial filter dimension
* @return The function returns required buffer size(bytes)
*
*/
int32_t arm_convolve_1_x_n_s8_get_buffer_size(const cmsis_nn_dims *input_dims, const cmsis_nn_dims *filter_dims);
/**
* @brief Q7 version of convolution for RGB image
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This kernel is written exclusively for convolution with ch_im_in
* equals 3. This applies on the first layer of CNNs which has input
* image with RGB format.
*/
arm_status arm_convolve_HWC_q7_RGB(const q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast Q15 convolution function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 2
* ch_im_out is multiple of 2
*/
arm_status arm_convolve_HWC_q15_fast(const q15_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q15_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q15_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q15_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Fast Q15 convolution function (non-sqaure shape)
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in_x input tensor dimension x
* @param[in] dim_im_in_y input tensor dimension y
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel_x filter kernel size x
* @param[in] dim_kernel_y filter kernel size y
* @param[in] padding_x padding size x
* @param[in] padding_y padding size y
* @param[in] stride_x convolution stride x
* @param[in] stride_y convolution stride y
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out_x output tensor dimension x
* @param[in] dim_im_out_y output tensor dimension y
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* @details
*
* <b>Buffer size:</b>
*
* bufferA size: 2*ch_im_in*dim_kernel*dim_kernel
*
* bufferB size: 0
*
* <b>Input dimension constraints:</b>
*
* ch_im_in is multiple of 2
*
* ch_im_out is multipe of 2
*
*/
arm_status arm_convolve_HWC_q15_fast_nonsquare(const q15_t *Im_in,
const uint16_t dim_im_in_x,
const uint16_t dim_im_in_y,
const uint16_t ch_im_in,
const q15_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel_x,
const uint16_t dim_kernel_y,
const uint16_t padding_x,
const uint16_t padding_y,
const uint16_t stride_x,
const uint16_t stride_y,
const q15_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q15_t *Im_out,
const uint16_t dim_im_out_x,
const uint16_t dim_im_out_y,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Q7 depthwise separable convolution function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 2
* ch_im_out is multiple of 2
*/
arm_status arm_depthwise_separable_conv_HWC_q7(const q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Q7 depthwise separable convolution function (non-square shape)
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in_x input tensor dimension x
* @param[in] dim_im_in_y input tensor dimension y
* @param[in] ch_im_in number of input tensor channels
* @param[in] wt pointer to kernel weights
* @param[in] ch_im_out number of filters, i.e., output tensor channels
* @param[in] dim_kernel_x filter kernel size x
* @param[in] dim_kernel_y filter kernel size y
* @param[in] padding_x padding sizes x
* @param[in] padding_y padding sizes y
* @param[in] stride_x convolution stride x
* @param[in] stride_y convolution stride y
* @param[in] bias pointer to bias
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in,out] Im_out pointer to output tensor
* @param[in] dim_im_out_x output tensor dimension x
* @param[in] dim_im_out_y output tensor dimension y
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] bufferB pointer to buffer space for output
* @return The function returns either
* <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
*
* This function is the version with full list of optimization tricks, but with
* some contraints:
* ch_im_in is multiple of 2
* ch_im_out is multiple of 2
*/
arm_status arm_depthwise_separable_conv_HWC_q7_nonsquare(const q7_t *Im_in,
const uint16_t dim_im_in_x,
const uint16_t dim_im_in_y,
const uint16_t ch_im_in,
const q7_t *wt,
const uint16_t ch_im_out,
const uint16_t dim_kernel_x,
const uint16_t dim_kernel_y,
const uint16_t padding_x,
const uint16_t padding_y,
const uint16_t stride_x,
const uint16_t stride_y,
const q7_t *bias,
const uint16_t bias_shift,
const uint16_t out_shift,
q7_t *Im_out,
const uint16_t dim_im_out_x,
const uint16_t dim_im_out_y,
q15_t *bufferA,
q7_t *bufferB);
/**
* @brief Wrapper function to pick the right optimized s8 depthwise convolution function
*
* @param[in, out] ctx Function context (e.g. temporary buffer). Check the function
* definition file to see if an additional buffer is required.
* Optional function {API}_get_buffer_size() provides the buffer
* size if required.
* @param[in] dw_conv_params Depthwise convolution parameters (e.g. strides, dilations, pads,...)
* dw_conv_params->dilation is not used.
* Range of dw_conv_params->input_offset : [-127, 128]
* Range of dw_conv_params->output_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each
* output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [H, W, C_IN]
* Batch argument N is not used and assumed to be 1.
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [1, H, W, C_OUT]
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [1, H, W, C_OUT]
* @param[in, out] output_data Output data pointer. Data type: int8
* @return The function returns
* <code>ARM_MATH_SUCCESS</code> - Successful completion.
*
* @details
* - Supported framework: TensorFlow Lite
* - Picks one of the the following functions
* -# arm_depthwise_conv_s8()
* -# arm_depthwise_conv_3x3_s8() - Cortex-M CPUs with DSP extension only
* -# arm_depthwise_conv_s8_opt()
* - q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
* - Check details of arm_depthwise_conv_s8_opt() for potential data that can be accessed outside of the
* boundary.
*/
arm_status arm_depthwise_conv_wrapper_s8(const cmsis_nn_context *ctx,
const cmsis_nn_dw_conv_params *dw_conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get size of additional buffer required by arm_depthwise_conv_wrapper_s8()
*
* @param[in] dw_conv_params Depthwise convolution parameters (e.g. strides, dilations, pads,...)
* dw_conv_params->dilation is not used.
* Range of dw_conv_params->input_offset : [-127, 128]
* Range of dw_conv_params->input_offset : [-128, 127]
* @param[in] input_dims Input (activation) tensor dimensions. Format: [H, W, C_IN]
* Batch argument N is not used and assumed to be 1.
* @param[in] filter_dims Filter tensor dimensions. Format: [1, H, W, C_OUT]
* @param[in] output_dims Output tensor dimensions. Format: [1, H, W, C_OUT]
* @return Size of additional memory required for optimizations in bytes.
*
*/
int32_t arm_depthwise_conv_wrapper_s8_get_buffer_size(const cmsis_nn_dw_conv_params *dw_conv_params,
const cmsis_nn_dims *input_dims,
const cmsis_nn_dims *filter_dims,
const cmsis_nn_dims *output_dims);
/**
* @brief Basic s8 depthwise convolution function that doesn't have any constraints on the input dimensions.
*
* @param[in, out] ctx Function context (e.g. temporary buffer). Check the function
* definition file to see if an additional buffer is required.
* Optional function {API}_get_buffer_size() provides the buffer
* size if an additional buffer is required.
* exists if additional memory is.
* @param[in] dw_conv_params Depthwise convolution parameters (e.g. strides, dilations, pads,...)
* dw_conv_params->dilation is not used.
* Range of dw_conv_params->input_offset : [-127, 128]
* Range of dw_conv_params->input_offset : [-128, 127]
* @param[in] quant_params Per-channel quantization info.
* It contains the multiplier and shift values to be applied to each
* output channel
* @param[in] input_dims Input (activation) tensor dimensions. Format: [1, H, W, C_IN]
* Batch argument N is not used.
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [1, H, W, C_OUT]
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* @param[in] bias_data Bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [1, H, W, C_OUT]
* @param[in, out] output_data Output data pointer. Data type: int8
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
* @details
* - Supported framework: TensorFlow Lite
* - q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
*/
arm_status arm_depthwise_conv_s8(const cmsis_nn_context *ctx,
const cmsis_nn_dw_conv_params *dw_conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Optimized s8 depthwise convolution function for 3x3 kernel size with some constraints on
* the input arguments(documented below). Refer arm_depthwise_conv_s8() for function
* argument details.
*
* @return The function returns one of the following
* <code>ARM_MATH_SIZE_MISMATCH</code> - Unsupported dimension of tensors
* <code>ARM_MATH_ARGUMENT_ERROR</code> - Unsupported pad size along the x axis
* <code>ARM_MATH_SUCCESS</code> - Successful operation
*
* @details
* - Supported framework : TensorFlow Lite Micro
* - The following constrains on the arguments apply
* -# Number of input channel equals number of output channels
* -# Filter height and width equals 3
* -# Padding along x is either 0 or 1.
*
*/
arm_status arm_depthwise_conv_3x3_s8(const cmsis_nn_context *ctx,
const cmsis_nn_dw_conv_params *dw_conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Optimized s8 depthwise convolution function with constraint that in_channel equals out_channel.
* Refer arm_depthwise_conv_s8() for function argument details.
*
* @return The function returns one of the following
* <code>ARM_MATH_SIZE_MISMATCH</code> - input channel != output channel or
* ch_mult != 1
* <code>ARM_MATH_SUCCESS</code> - Successful operation
*
* @note If number of channels is not a multiple of 4, upto 3 elements outside the boundary will be read out
* for the following if MVE optimizations(Arm Helium Technology) are used.
* - Output shift
* - Output multiplier
* - Output bias
* - kernel
* @details
* - Supported framework: TensorFlow Lite
* - The following constrains on the arguments apply
* -# Number of input channel equals number of output channels or ch_mult equals 1
* - q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
* - Reccomended when number of channels is 4 or greater.
*
*/
arm_status arm_depthwise_conv_s8_opt(const cmsis_nn_context *ctx,
const cmsis_nn_dw_conv_params *dw_conv_params,
const cmsis_nn_per_channel_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for optimized s8 depthwise convolution
* function with constraint that in_channel equals out_channel.
* @param[in] input_dims Input (activation) tensor dimensions. Format: [1, H, W, C_IN]
* Batch argument N is not used.
* @param[in] filter_dims Filter tensor dimensions. Format: [1, H, W, C_OUT]
* @return The function returns required buffer size in bytes
*
*/
int32_t arm_depthwise_conv_s8_opt_get_buffer_size(const cmsis_nn_dims *input_dims,
const cmsis_nn_dims *filter_dims);
/**
* @defgroup FC Fully-connected Layer Functions
*
* Collection of fully-connected and matrix multiplication functions.
*
* Fully-connected layer is basically a matrix-vector multiplication
* with bias. The matrix is the weights and the input/output vectors
* are the activation values. Supported {weight, activation} precisions
* include {8-bit, 8-bit}, {16-bit, 16-bit}, and {8-bit, 16-bit}.
*
* Here we have two types of kernel functions. The basic function
* implements the function using regular GEMV approach. The opt functions
* operates with weights in interleaved formats.
*
*/
/**
*@brief Q7 basic fully-connected layer function
*@param[in] pV pointer to input vector
*@param[in] pM pointer to matrix weights
*@param[in] dim_vec length of the vector
*@param[in] num_of_rows number of rows in weight matrix
*@param[in] bias_shift amount of left-shift for bias
*@param[in] out_shift amount of right-shift for output
*@param[in] bias pointer to bias
*@param[in,out] pOut pointer to output vector
*@param[in,out] vec_buffer pointer to buffer space for input
*@return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_q7(const q7_t *pV,
const q7_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q7_t *pOut,
q15_t *vec_buffer);
/**
* @brief Basic s8 Fully Connected function.
*
* @param[in, out] ctx Function context (e.g. temporary buffer). Check the function
* definition file to see if an additional buffer is required.
* Optional function {API}_get_buffer_size() provides the buffer
* size if an additional buffer is required.
* @param[in] fc_params Fully Connected layer parameters (e.g. strides, dilations, pads,...)
* Range of fc_params->input_offset : [-127, 128]
* Range of fc_params->filter_offset : [-127, 128]
* Range of fc_params->output_offset : [-128, 127]
* @param[in] quant_params Per-tensor quantization info.
* It contains the multiplier and shift values to be applied to the output tensor.
* @param[in] input_dims Input (activation) tensor dimensions. Format: [N, H, W, C_IN]
* Input dimension is taken as Nx(H * W * C_IN)
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Two dimensional filter dimensions. Format: [N, C]
* N : accumulation depth and equals (H * W * C_IN) from input_dims
* C : output depth and equals C_OUT in output_dims
* H & W : Not used
* @param[in] filter_data Filter data pointer. Data type: int8
* @param[in] bias_dims Bias tensor dimensions. Format: [C_OUT]
* N, H, W : Not used
* @param[in] bias_data Bias data pointer. Data type: int32
* @param[in] output_dims Output tensor dimensions. Format: [N, C_OUT]
* N : Batches
* C_OUT : Output depth
* H & W : Not used.
* @param[in, out] output_data Output data pointer. Data type: int8
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
* @details
* - Supported framework: TensorFlow Lite
* - q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
*/
arm_status arm_fully_connected_s8(const cmsis_nn_context *ctx,
const cmsis_nn_fc_params *fc_params,
const cmsis_nn_per_tensor_quant_params *quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const q7_t *filter_data,
const cmsis_nn_dims *bias_dims,
const int32_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for S8 basic fully-connected and
* matrix multiplication layer function for TF Lite
* @param[in] filter_dims dimension of filter
* @return The function returns required buffer size in bytes
*
*/
int32_t arm_fully_connected_s8_get_buffer_size(const cmsis_nn_dims *filter_dims);
/**
* @brief Q7 opt fully-connected layer function
* @param[in] pV pointer to input vector
* @param[in] pM pointer to matrix weights
* @param[in] dim_vec length of the vector
* @param[in] num_of_rows number of rows in weight matrix
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias pointer to bias
* @param[in,out] pOut pointer to output vector
* @param[in,out] vec_buffer pointer to buffer space for input
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_q7_opt(const q7_t *pV,
const q7_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q7_t *pOut,
q15_t *vec_buffer);
/**
* @brief Q15 basic fully-connected layer function
* @param[in] pV pointer to input vector
* @param[in] pM pointer to matrix weights
* @param[in] dim_vec length of the vector
* @param[in] num_of_rows number of rows in weight matrix
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias pointer to bias
* @param[in,out] pOut pointer to output vector
* @param[in,out] vec_buffer pointer to buffer space for input
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_q15(const q15_t *pV,
const q15_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q15_t *bias,
q15_t *pOut,
q15_t *vec_buffer);
/**
* @brief Q15 opt fully-connected layer function
* @param[in] pV pointer to input vector
* @param[in] pM pointer to matrix weights
* @param[in] dim_vec length of the vector
* @param[in] num_of_rows number of rows in weight matrix
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias pointer to bias
* @param[in,out] pOut pointer to output vector
* @param[in,out] vec_buffer pointer to buffer space for input
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_q15_opt(const q15_t *pV,
const q15_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q15_t *bias,
q15_t *pOut,
q15_t *vec_buffer);
/**
* @brief Mixed Q15-Q7 fully-connected layer function
* @param[in] pV pointer to input vector
* @param[in] pM pointer to matrix weights
* @param[in] dim_vec length of the vector
* @param[in] num_of_rows number of rows in weight matrix
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias pointer to bias
* @param[in,out] pOut pointer to output vector
* @param[in,out] vec_buffer pointer to buffer space for input
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_mat_q7_vec_q15(const q15_t *pV,
const q7_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q15_t *pOut,
q15_t *vec_buffer);
/**
* @brief Mixed Q15-Q7 opt fully-connected layer function
* @param[in] pV pointer to input vector
* @param[in] pM pointer to matrix weights
* @param[in] dim_vec length of the vector
* @param[in] num_of_rows number of rows in weight matrix
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias pointer to bias
* @param[in,out] pOut pointer to output vector
* @param[in,out] vec_buffer pointer to buffer space for input
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
*/
arm_status arm_fully_connected_mat_q7_vec_q15_opt(const q15_t *pV,
const q7_t *pM,
const uint16_t dim_vec,
const uint16_t num_of_rows,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q15_t *pOut,
q15_t *vec_buffer);
/**
* @brief Matrix-Multiplication Kernels for Convolution
*
* These functions are used within convolution layer functions for
* matrix multiplication.
*
* The implementation is similar to CMSIS-DSP arm_mat_mult functions
* with one Q7 and one Q15 operands. The Q15 operand is the im2col
* output which is always with 2 columns.
*
*/
/**
* @brief Matrix-multiplication function for convolution
* @param[in] pA pointer to operand A
* @param[in] pInBuffer pointer to operand B, always conssists of 2 vectors
* @param[in] ch_im_out numRow of A
* @param[in] numCol_A numCol of A
* @param[in] bias_shift amount of left-shift for bias
* @param[in] out_shift amount of right-shift for output
* @param[in] bias the bias
* @param[in,out] pOut pointer to output
* @return The function returns the incremented output pointer
*/
q7_t *arm_nn_mat_mult_kernel_q7_q15(const q7_t *pA,
const q15_t *pInBuffer,
const uint16_t ch_im_out,
const uint16_t numCol_A,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q7_t *pOut);
/**
* @brief Matrix-multiplication function for convolution with per-channel requantization.
* @param[in] input_a pointer to operand A
* @param[in] input_b pointer to operand B, always consists of 2 vectors.
* @param[in] output_ch number of rows of A
* @param[in] out_shift pointer to per output channel requantization shift parameter.
* @param[in] out_mult pointer to per output channel requantization multiplier parameter.
* @param[in] out_offset output tensor offset.
* @param[in] activation_min minimum value to clamp the output to. Range : int8
* @param[in] activation_max maximum value to clamp the output to. Range : int8
* @param[in] num_col_a number of columns of A
* @param[in] output_bias per output channel bias. Range : int32
* @param[in,out] out_0 pointer to output
* @return The function returns one of the two
* 1. The incremented output pointer for a successful operation or
* 2. NULL if implementation is not available.
*
* @details This function does the matrix multiplication of weight matrix for all output channels
* with 2 columns from im2col and produces two elements/output_channel. The outputs are
* clamped in the range provided by activation min and max.
* Supported framework: TensorFlow Lite micro.
*/
q7_t *arm_nn_mat_mult_kernel_s8_s16(const q7_t *input_a,
const q15_t *input_b,
const uint16_t output_ch,
const int32_t *out_shift,
const int32_t *out_mult,
const int32_t out_offset,
const int16_t activation_min,
const int16_t activation_max,
const uint16_t num_col_a,
const int32_t *const output_bias,
q7_t *out_0);
/**
* @brief Matrix-multiplication of re-ordered input B with A.
*
* @details For arguments, refer arm_nn_mat_mult_kernel_s8_s16. The re-ordering is a consequence
* of sign extension done by the SXTB16 command on input_b. The outputs are clamped in the range
* provided by activation min and max.
* * @details
* - Supported framework : TensorFlow Lite Micro
* - The following constrains on the arguments apply
* -# num_col_a is a multiple of 4
* -# output_ch is a multiple of 2
*
*/
q7_t *arm_nn_mat_mult_kernel_s8_s16_reordered(const q7_t *input_a,
const q15_t *input_b,
const uint16_t output_ch,
const int32_t *out_shift,
const int32_t *out_mult,
const int32_t out_offset,
const int16_t activation_min,
const int16_t activation_max,
const uint16_t num_col_a,
const int32_t *const output_bias,
q7_t *out_0);
/**
*@brief Matrix-multiplication function for convolution with reordered columns
*@param[in] pA pointer to operand A
*@param[in] pInBuffer pointer to operand B, always conssists of 2 vectors
*@param[in] ch_im_out numRow of A
*@param[in] numCol_A numCol of A
*@param[in] bias_shift amount of left-shift for bias
*@param[in] out_shift amount of right-shift for output
*@param[in] bias the bias
*@param[in,out] pOut pointer to output
*@return The function returns the incremented output pointer
*
*@details This function assumes that data in pInBuffer are reordered
*/
q7_t *arm_nn_mat_mult_kernel_q7_q15_reordered(const q7_t *pA,
const q15_t *pInBuffer,
const uint16_t ch_im_out,
const uint16_t numCol_A,
const uint16_t bias_shift,
const uint16_t out_shift,
const q7_t *bias,
q7_t *pOut);
#ifdef __cplusplus
}
#endif
/*
* Other functions
* These layers are typically not timing critical
* Basic implementation is supported here
*/
#ifdef __cplusplus
extern "C"
{
#endif
/**
* @defgroup BasicMath Basic math functions
*
* Element wise add and multiplication functions.
*
*/
/**
* @brief s8 element wise add of two vectors
* @param[in] input_1_vect pointer to input vector 1
* @param[in] input_2_vect pointer to input vector 2
* @param[in] input_1_offset offset for input 1. Range: Range: -127 to 128
* @param[in] input_1_mult multiplier for input 1
* @param[in] input_1_shift shift for input 1
* @param[in] input_2_offset offset for input 2. Range: Range: -127 to 128
* @param[in] input_2_mult multiplier for input 2
* @param[in] input_2_shift shift for input 2
* @param[in] left_shift input left shift
* @param[in,out] output pointer to output vector
* @param[in] out_offset output offset
* @param[in] out_mult output multiplier
* @param[in] out_shift output shift
* @param[in] out_activation_min minimum value to clamp output to
* @param[in] out_activation_max maximum value to clamp output to
* @param[in] block_size number of samples
* @return The function returns ARM_MATH_SUCCESS
*/
arm_status arm_elementwise_add_s8(const int8_t *input_1_vect,
const int8_t *input_2_vect,
const int32_t input_1_offset,
const int32_t input_1_mult,
const int32_t input_1_shift,
const int32_t input_2_offset,
const int32_t input_2_mult,
const int32_t input_2_shift,
const int32_t left_shift,
int8_t *output,
const int32_t out_offset,
const int32_t out_mult,
const int32_t out_shift,
const int32_t out_activation_min,
const int32_t out_activation_max,
const uint32_t block_size);
/**
* @brief s8 element wise multiplication
* @param[in] input_1_vect pointer to input vector 1
* @param[in] input_2_vect pointer to input vector 2
* @param[in] input_1_offset offset for input 1. Range: Range: -127 to 128
* @param[in] input_2_offset offset for input 2. Range: Range: -127 to 128
* @param[in,out] output pointer to output vector
* @param[in] out_offset output offset
* @param[in] out_mult output multiplier
* @param[in] out_shift output shift
* @param[in] out_activation_min minimum value to clamp output to
* @param[in] out_activation_max maximum value to clamp output to
* @param[in] block_size number of samples
* @return The function returns ARM_MATH_SUCCESS
*
* @details Supported framework: TensorFlow Lite micro
*/
arm_status arm_elementwise_mul_s8(const int8_t *input_1_vect,
const int8_t *input_2_vect,
const int32_t input_1_offset,
const int32_t input_2_offset,
int8_t *output,
const int32_t out_offset,
const int32_t out_mult,
const int32_t out_shift,
const int32_t out_activation_min,
const int32_t out_activation_max,
const uint32_t block_size);
/**
* @defgroup Acti Activation Functions
*
* Perform activation layers, including ReLU (Rectified Linear Unit),
* sigmoid and tanh
*
*/
/**
* @brief Q7 RELU function
* @param[in,out] data pointer to input
* @param[in] size number of elements
* @return none.
*/
void arm_relu_q7(q7_t *data, uint16_t size);
/**
* @brief s8 ReLU6 function
* @param[in,out] data pointer to input
* @param[in] size number of elements
*/
void arm_relu6_s8(q7_t *data, uint16_t size);
/**
* @brief Q15 RELU function
* @param[in,out] data pointer to input
* @param[in] size number of elements
* @return none.
*/
void arm_relu_q15(q15_t *data, uint16_t size);
/**
* @brief Q7 neural network activation function using direct table look-up
* @param[in,out] data pointer to input
* @param[in] size number of elements
* @param[in] int_width bit-width of the integer part, assume to be smaller than 3
* @param[in] type type of activation functions
* @return none.
*/
void arm_nn_activations_direct_q7(q7_t *data, uint16_t size, uint16_t int_width, arm_nn_activation_type type);
/**
* @brief Q15 neural network activation function using direct table look-up
* @param[in,out] data pointer to input
* @param[in] size number of elements
* @param[in] int_width bit-width of the integer part, assume to be smaller than 3
* @param[in] type type of activation functions
* @return none.
*
* @details
*
* This is the direct table look-up approach.
*
* Assume here the integer part of the fixed-point is <= 3.
* More than 3 just not making much sense, makes no difference with
* saturation followed by any of these activation functions.
*/
void arm_nn_activations_direct_q15(q15_t *data, uint16_t size, uint16_t int_width, arm_nn_activation_type type);
/**
* @defgroup Pooling Pooling Functions
*
* Perform pooling functions, including max pooling and average pooling
*
*/
/**
* @brief Q7 max pooling function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] Im_out pointer to output tensor
* @return none.
*
*/
void arm_maxpool_q7_HWC(q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const uint16_t dim_im_out,
q7_t *bufferA,
q7_t *Im_out);
/**
* @brief Q7 average pooling function
* @param[in] Im_in pointer to input tensor
* @param[in] dim_im_in input tensor dimension
* @param[in] ch_im_in number of input tensor channels
* @param[in] dim_kernel filter kernel size
* @param[in] padding padding sizes
* @param[in] stride convolution stride
* @param[in] dim_im_out output tensor dimension
* @param[in,out] bufferA pointer to buffer space for input
* @param[in,out] Im_out pointer to output tensor
* @return none.
*
*/
void arm_avepool_q7_HWC(q7_t *Im_in,
const uint16_t dim_im_in,
const uint16_t ch_im_in,
const uint16_t dim_kernel,
const uint16_t padding,
const uint16_t stride,
const uint16_t dim_im_out,
q7_t *bufferA,
q7_t *Im_out);
/**
* @brief s8 average pooling function.
*
* @param[in, out] ctx Function context (e.g. temporary buffer). Check the function
* definition file to see if an additional buffer is required.
* Optional function {API}_get_buffer_size() provides the buffer
* size if an additional buffer is required.
* @param[in] pool_params Pooling parameters
* @param[in] input_dims Input (activation) tensor dimensions. Format: [H, W, C_IN]
* Argument 'N' is not used.
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [H, W]
* Argument N and C are not used.
* @param[in] output_dims Output tensor dimensions. Format: [H, W, C_OUT]
* Argument N is not used.
* C_OUT equals C_IN.
* @param[in, out] output_data Output data pointer. Data type: int8
* @return The function returns
* <code>ARM_MATH_SUCCESS</code> - Successful operation
*
* @details
* - Supported Framework: TensorFlow Lite
*
*/
arm_status arm_avgpool_s8(const cmsis_nn_context *ctx,
const cmsis_nn_pool_params *pool_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @brief Get the required buffer size for S8 average pooling function
* @param[in] dim_dst_width output tensor dimension
* @param[in] ch_src number of input tensor channels
* @return The function returns required buffer size in bytes
*
*/
int32_t arm_avgpool_s8_get_buffer_size(const int dim_dst_width, const int ch_src);
/**
* @brief s8 max pooling function.
*
* @param[in, out] ctx Function context (e.g. temporary buffer). Check the function
* definition file to see if an additional buffer is required.
* Optional function {API}_get_buffer_size() provides the buffer
* size if an additional buffer is required.
* @param[in] pool_params Pooling parameters
* @param[in] input_dims Input (activation) tensor dimensions. Format: [H, W, C_IN]
* Argument 'N' is not used.
* @param[in] input_data Input (activation) data pointer. Data type: int8
* @param[in] filter_dims Filter tensor dimensions. Format: [H, W]
* Argument N and C are not used.
* @param[in] output_dims Output tensor dimensions. Format: [H, W, C_OUT]
* Argument N is not used.
* C_OUT equals C_IN.
* @param[in, out] output_data Output data pointer. Data type: int8
* @return The function returns
* <code>ARM_MATH_SUCCESS</code> - Successful operation
*
* @details
* - Supported Framework: TensorFlow Lite
*
*/
arm_status arm_max_pool_s8(const cmsis_nn_context *ctx,
const cmsis_nn_pool_params *pool_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *filter_dims,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
/**
* @defgroup Softmax Softmax Functions
*
* EXP(2) based softmax functions.
*
*/
/**
* @brief Q7 softmax function
* @param[in] vec_in pointer to input vector
* @param[in] dim_vec input vector dimension
* @param[out] p_out pointer to output vector
*
* @note This function is an optimized version which is not bit-accurate with
* TensorFlow Lite's kernel
*
*/
void arm_softmax_q7(const q7_t *vec_in, const uint16_t dim_vec, q7_t *p_out);
/**
* @brief Q7 softmax function with batch parameter
* @param[in] vec_in pointer to input vector
* @param[in] nb_batches number of batches
* @param[in] dim_vec input vector dimension
* @param[out] p_out pointer to output vector
* @return none.
*
* @note This function is an optimized version which is not bit-accurate with
* TensorFlow Lite's kernel
*
*/
void arm_softmax_with_batch_q7(const q7_t *vec_in, const uint16_t nb_batches, const uint16_t dim_vec, q7_t *p_out);
/**
* @brief Q15 softmax function
* @param[in] vec_in pointer to input vector
* @param[in] dim_vec input vector dimension
* @param[out] p_out pointer to output vector
* @return none.
*
* @note This function is an optimized version which is not bit-accurate with
* TensorFlow Lite's kernel
*
*/
void arm_softmax_q15(const q15_t *vec_in, const uint16_t dim_vec, q15_t *p_out);
/**
* @brief S8 softmax function
* @param[in] input Pointer to the input tensor
* @param[in] num_rows Number of rows in the input tensor
* @param[in] row_size Number of elements in each input row
* @param[in] mult Input quantization multiplier
* @param[in] shift Input quantization shift within the range [0, 31]
* @param[in] diff_min Minimum difference with max in row. Used to check if
* the quantized exponential operation can be performed
* @param[out] output Pointer to the output tensor
*
* @note Supported framework: TensorFlow Lite micro (bit-accurate)
*
*/
void arm_softmax_s8(const int8_t *input,
const int32_t num_rows,
const int32_t row_size,
const int32_t mult,
const int32_t shift,
const int32_t diff_min,
int8_t *output);
/**
* @brief U8 softmax function
* @param[in] input Pointer to the input tensor
* @param[in] num_rows Number of rows in the input tensor
* @param[in] row_size Number of elements in each input row
* @param[in] mult Input quantization multiplier
* @param[in] shift Input quantization shift within the range [0, 31]
* @param[in] diff_min Minimum difference with max in row. Used to check if
* the quantized exponential operation can be performed
* @param[out] output Pointer to the output tensor
*
* @note Supported framework: TensorFlow Lite micro (bit-accurate)
*
*/
void arm_softmax_u8(const uint8_t *input,
const int32_t num_rows,
const int32_t row_size,
const int32_t mult,
const int32_t shift,
const int32_t diff_min,
uint8_t *output);
/**
* @brief uint8 depthwise convolution function with asymmetric quantization
* Unless specified otherwise, arguments are mandatory.
*
* @param[in] input Pointer to input tensor
* @param[in] input_x Width of input tensor
* @param[in] input_y Height of input tensor
* @param[in] input_ch Channels in input tensor
* @param[in] kernel Pointer to kernel weights
* @param[in] kernel_x Width of kernel
* @param[in] kernel_y Height of kernel
* @param[in] ch_mult Number of channel multiplier
* @param[in] pad_x Padding sizes x
* @param[in] pad_y Padding sizes y
* @param[in] stride_x stride along the width
* @param[in] stride_y stride along the height
* @param[in] dilation_x Dilation along width. Not used and intended for future enhancement.
* @param[in] dilation_y Dilation along height. Not used and intended for future enhancement.
* @param[in] bias Pointer to optional bias values. If no bias is
* availble, NULL is expected
* @param[in] input_offset Input tensor zero offset
* @param[in] filter_offset Kernel tensor zero offset
* @param[in] output_offset Output tensor zero offset
* @param[in,out] output Pointer to output tensor
* @param[in] output_x Width of output tensor
* @param[in] output_y Height of output tensor
* @param[in] output_activation_min Minimum value to clamp the output to. Range : {0, 255}
* @param[in] output_activation_max Minimum value to clamp the output to. Range : {0, 255}
* @param[in] out_shift Amount of right-shift for output
* @param[in] out_mult Output multiplier for requantization
* @return The function returns the following
* <code>ARM_MATH_SUCCESS</code> - Successful operation
*
*/
arm_status arm_depthwise_conv_u8_basic_ver1(const uint8_t *input,
const uint16_t input_x,
const uint16_t input_y,
const uint16_t input_ch,
const uint8_t *kernel,
const uint16_t kernel_x,
const uint16_t kernel_y,
const int16_t ch_mult,
const int16_t pad_x,
const int16_t pad_y,
const int16_t stride_x,
const int16_t stride_y,
const int16_t dilation_x,
const int16_t dilation_y,
const int32_t *bias,
const int32_t input_offset,
const int32_t filter_offset,
const int32_t output_offset,
uint8_t *output,
const uint16_t output_x,
const uint16_t output_y,
const int32_t output_activation_min,
const int32_t output_activation_max,
const int32_t out_shift,
const int32_t out_mult);
/**
* @defgroup Reshape Reshape Functions
*
*/
/**
* @brief Reshape a s8 vector into another with different shape
* @param[in] input points to the s8 input vector
* @param[out] output points to the s8 output vector
* @param[in] total_size total size of the input and output vectors in bytes
*
* @note The output is expected to be in a memory area that does not overlap with the input's
*
*/
void arm_reshape_s8(const int8_t *input, int8_t *output, const uint32_t total_size);
/**
* @defgroup Concatenation Concatenation Functions
*
*/
/**
* @brief int8/uint8 concatenation function to be used for concatenating N-tensors along the X axis
* This function should be called for each input tensor to concatenate. The argument offset_x
* will be used to store the input tensor in the correct position in the output tensor
*
* i.e. offset_x = 0
* for(i = 0 i < num_input_tensors; ++i)
* {
* arm_concatenation_s8_x(&input[i], ..., &output, ..., ..., offset_x)
* offset_x += input_x[i]
* }
*
* This function assumes that the output tensor has:
* -# The same height of the input tensor
* -# The same number of channels of the input tensor
* -# The same batch size of the input tensor
*
* Unless specified otherwise, arguments are mandatory.
*
* @note This function, data layout independent, can be used to concatenate either int8 or uint8 tensors because it
* does not involve any arithmetic operation
*
* @param[in] input Pointer to input tensor
* @param[in] input_x Width of input tensor
* @param[in] input_y Height of input tensor
* @param[in] input_z Channels in input tensor
* @param[in] input_w Batch size in input tensor
* @param[out] output Pointer to output tensor
* @param[in] output_x Width of output tensor
* @param[in] offset_x The offset (in number of elements) on the X axis to start concatenating the input tensor
* It is user responsibility to provide the correct value
*
* <b> Input constraints</b>
* offset_x is less than output_x
*
*/
void arm_concatenation_s8_x(const int8_t *input,
const uint16_t input_x,
const uint16_t input_y,
const uint16_t input_z,
const uint16_t input_w,
int8_t *output,
const uint16_t output_x,
const uint32_t offset_x);
/**
* @brief int8/uint8 concatenation function to be used for concatenating N-tensors along the Y axis
* This function should be called for each input tensor to concatenate. The argument offset_y
* will be used to store the input tensor in the correct position in the output tensor
*
* i.e. offset_y = 0
* for(i = 0 i < num_input_tensors; ++i)
* {
* arm_concatenation_s8_y(&input[i], ..., &output, ..., ..., offset_y)
* offset_y += input_y[i]
* }
*
* This function assumes that the output tensor has:
* -# The same width of the input tensor
* -# The same number of channels of the input tensor
* -# The same batch size of the input tensor
*
* Unless specified otherwise, arguments are mandatory.
*
* @note This function, data layout independent, can be used to concatenate either int8 or uint8 tensors because it
* does not involve any arithmetic operation
*
* @param[in] input Pointer to input tensor
* @param[in] input_x Width of input tensor
* @param[in] input_y Height of input tensor
* @param[in] input_z Channels in input tensor
* @param[in] input_w Batch size in input tensor
* @param[out] output Pointer to output tensor
* @param[in] output_y Height of output tensor
* @param[in] offset_y The offset on the Y axis to start concatenating the input tensor
* It is user responsibility to provide the correct value
*
* <b> Input constraints</b>
* offset_y is less than output_y
*
*/
void arm_concatenation_s8_y(const int8_t *input,
const uint16_t input_x,
const uint16_t input_y,
const uint16_t input_z,
const uint16_t input_w,
int8_t *output,
const uint16_t output_y,
const uint32_t offset_y);
/**
* @brief int8/uint8 concatenation function to be used for concatenating N-tensors along the Z axis
* This function should be called for each input tensor to concatenate. The argument offset_z
* will be used to store the input tensor in the correct position in the output tensor
*
* i.e. offset_z = 0
* for(i = 0 i < num_input_tensors; ++i)
* {
* arm_concatenation_s8_z(&input[i], ..., &output, ..., ..., offset_z)
* offset_z += input_z[i]
* }
*
* This function assumes that the output tensor has:
* -# The same width of the input tensor
* -# The same height of the input tensor
* -# The same batch size of the input tensor
*
* Unless specified otherwise, arguments are mandatory.
*
* @note This function, data layout independent, can be used to concatenate either int8 or uint8 tensors because it
* does not involve any arithmetic operation
*
* @param[in] input Pointer to input tensor
* @param[in] input_x Width of input tensor
* @param[in] input_y Height of input tensor
* @param[in] input_z Channels in input tensor
* @param[in] input_w Batch size in input tensor
* @param[out] output Pointer to output tensor
* @param[in] output_z Channels in output tensor
* @param[in] offset_z The offset on the Z axis to start concatenating the input tensor
* It is user responsibility to provide the correct value
*
* <b> Input constraints</b>
* offset_z is less than output_z
*
*/
void arm_concatenation_s8_z(const int8_t *input,
const uint16_t input_x,
const uint16_t input_y,
const uint16_t input_z,
const uint16_t input_w,
int8_t *output,
const uint16_t output_z,
const uint32_t offset_z);
/**
* @brief int8/uint8 concatenation function to be used for concatenating N-tensors along the W axis (Batch size)
* This function should be called for each input tensor to concatenate. The argument offset_w
* will be used to store the input tensor in the correct position in the output tensor
*
* i.e. offset_w = 0
* for(i = 0 i < num_input_tensors; ++i)
* {
* arm_concatenation_s8_w(&input[i], ..., &output, ..., ..., offset_w)
* offset_w += input_w[i]
* }
*
* This function assumes that the output tensor has:
* -# The same width of the input tensor
* -# The same height of the input tensor
* -# The same number o channels of the input tensor
*
* Unless specified otherwise, arguments are mandatory.
*
* @note This function, data layout independent, can be used to concatenate either int8 or uint8 tensors because it
* does not involve any arithmetic operation
*
* @param[in] input Pointer to input tensor
* @param[in] input_x Width of input tensor
* @param[in] input_y Height of input tensor
* @param[in] input_z Channels in input tensor
* @param[in] input_w Batch size in input tensor
* @param[out] output Pointer to output tensor
* @param[in] offset_w The offset on the W axis to start concatenating the input tensor
* It is user responsibility to provide the correct value
*
*/
void arm_concatenation_s8_w(const int8_t *input,
const uint16_t input_x,
const uint16_t input_y,
const uint16_t input_z,
const uint16_t input_w,
int8_t *output,
const uint32_t offset_w);
/**
* @defgroup SVDF SVDF Layer Functions
*
*/
/**
* @brief s8 SVDF function
*
* @param[in] input_ctx Temporary scratch buffer
* @param[in] output_ctx Temporary output scratch buffer
* @param[in] svdf_params SVDF Parameters
* Range of svdf_params->input_offset : [-128, 127]
* Range of svdf_params->output_offset : [-128, 127]
* @param[in] input_quant_params Input quantization parameters
* @param[in] output_quant_params Output quantization parameters
* @param[in] input_dims Input tensor dimensions
* @param[in] input_data Pointer to input tensor
* @param[in] state_dims State tensor dimensions
* @param[in] state_data Pointer to state tensor
* @param[in] weights_feature_dims Weights (feature) tensor dimensions
* @param[in] weights_feature_data Pointer to the weights (feature) tensor
* @param[in] weights_time_dims Weights (time) tensor dimensions
* @param[in] weights_time_data Pointer to the weights (time) tensor
* @param[in] bias_dims Bias tensor dimensions
* @param[in] bias_data Pointer to bias tensor
* @param[in] output_dims Output tensor dimensions
* @param[out] output_data Pointer to the output tensor
*
* @return The function returns <code>ARM_MATH_SUCCESS</code>
*
* @details
* 1. Supported framework: TensorFlow Lite micro
* 2. q7 is used as data type eventhough it is s8 data. It is done so to be consistent with existing APIs.
*
*/
arm_status arm_svdf_s8(const cmsis_nn_context *input_ctx,
const cmsis_nn_context *output_ctx,
const cmsis_nn_svdf_params *svdf_params,
const cmsis_nn_per_tensor_quant_params *input_quant_params,
const cmsis_nn_per_tensor_quant_params *output_quant_params,
const cmsis_nn_dims *input_dims,
const q7_t *input_data,
const cmsis_nn_dims *state_dims,
q15_t *state_data,
const cmsis_nn_dims *weights_feature_dims,
const q7_t *weights_feature_data,
const cmsis_nn_dims *weights_time_dims,
const q15_t *weights_time_data,
const cmsis_nn_dims *bias_dims,
const q31_t *bias_data,
const cmsis_nn_dims *output_dims,
q7_t *output_data);
#ifdef __cplusplus
}
#endif
#endif
|