File size: 21,133 Bytes
dffe123 | 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 | // 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.
#ifndef OPENCV_DNN_SRC_CUDA_FUNCTORS_HPP
#define OPENCV_DNN_SRC_CUDA_FUNCTORS_HPP
#include <cuda_runtime.h>
#include "math.hpp"
#include "../cuda4dnn/csl/nvcc_defs.hpp"
namespace cv { namespace dnn { namespace cuda4dnn { namespace kernels {
template <class T>
struct IdentityFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE IdentityFunctor() { }
CUDA4DNN_DEVICE IdentityFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
return value;
};
};
template <class T>
struct ReLUFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : slope(0) { }
CUDA4DNN_HOST_DEVICE Params(T slope_) : slope(slope_) { }
T slope;
};
CUDA4DNN_DEVICE ReLUFunctor() : ReLUFunctor(Params{}) { }
CUDA4DNN_DEVICE ReLUFunctor(const Params& params) : slope(params.slope) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::log1pexp;
return value >= T(0) ? value : slope * value;
}
T slope;
};
template <class T>
struct ClippedReLUFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : floor(0), ceiling(6) { }
CUDA4DNN_HOST_DEVICE Params(T floor_, T ceiling_) : floor(floor_), ceiling(ceiling_) { }
T floor, ceiling;
};
CUDA4DNN_DEVICE ClippedReLUFunctor() : ClippedReLUFunctor(Params{}) { }
CUDA4DNN_DEVICE ClippedReLUFunctor(const Params& params) : floor{params.floor}, ceiling{params.ceiling} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::clamp;
return clamp(value, floor, ceiling);
}
T floor, ceiling;
};
template <class T>
struct TanHFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE TanHFunctor() { }
CUDA4DNN_DEVICE TanHFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::tanh;
return tanh(value);
}
};
template <class T>
struct SwishFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SwishFunctor() { }
CUDA4DNN_DEVICE SwishFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
// f(x) = x * sigmoid(x)
using csl::device::fast_divide;
using csl::device::fast_exp;
return fast_divide(value, static_cast<T>(1) + fast_exp(-value));
}
};
template <class T>
struct MishFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE MishFunctor() { }
CUDA4DNN_DEVICE MishFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::tanh;
using csl::device::log1pexp;
return value * tanh(log1pexp(value));
}
};
template <>
struct MishFunctor<float> {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE MishFunctor() { }
CUDA4DNN_DEVICE MishFunctor(const Params& params) { }
CUDA4DNN_DEVICE float operator()(float value) {
// f(x) = x * tanh(log1pexp(x));
using csl::device::fast_divide;
using csl::device::fast_exp;
auto e = fast_exp(value);
auto n = e * e + 2 * e;
if (value <= -0.6f)
return value * fast_divide(n, n + 2);
return value - 2 * fast_divide(value, n + 2);
}
};
#if !defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 530)
template <>
struct MishFunctor<__half> {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE MishFunctor() { }
CUDA4DNN_DEVICE MishFunctor(const Params& params) { }
CUDA4DNN_DEVICE __half operator()(__half value) {
return MishFunctor<float>()(value);
}
};
#endif
template <class T>
struct SigmoidFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SigmoidFunctor() { }
CUDA4DNN_DEVICE SigmoidFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::fast_sigmoid;
return fast_sigmoid(value);
}
};
template <class T>
struct ELUFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : alpha(1) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_) : alpha(alpha_) { }
T alpha;
};
CUDA4DNN_DEVICE ELUFunctor() : ELUFunctor(Params{}) { }
CUDA4DNN_DEVICE ELUFunctor(const Params& params) : alpha{params.alpha} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::expm1;
return value >= T(0) ? value : alpha * expm1(value);
}
T alpha;
};
template <class T>
struct AbsFunctor {
struct Params { };
CUDA4DNN_DEVICE AbsFunctor() { }
CUDA4DNN_DEVICE AbsFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::abs;
return abs(value);
}
};
template <class T>
struct BNLLFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE BNLLFunctor() { }
CUDA4DNN_DEVICE BNLLFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::log1pexp;
return value > T(0) ? value + log1pexp(-value) : log1pexp(value);
}
};
template <class T>
struct CeilFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE CeilFunctor() { }
CUDA4DNN_DEVICE CeilFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::ceil;
return ceil(value);
}
};
template <class T>
struct FloorFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE FloorFunctor() { }
CUDA4DNN_DEVICE FloorFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::floor;
return floor(value);
}
};
template <class T>
struct LogFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE LogFunctor() { }
CUDA4DNN_DEVICE LogFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::log;
return log(value);
}
};
template <class T>
struct RintFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE RintFunctor() { }
CUDA4DNN_DEVICE RintFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::rint;
return rint(value);
}
};
template <class T>
struct SqrtFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SqrtFunctor() { }
CUDA4DNN_DEVICE SqrtFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::sqrt;
return sqrt(value);
}
};
template <class T>
struct NotFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE NotFunctor() { }
CUDA4DNN_DEVICE NotFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::floor;
return floor(static_cast<T>(1.) - value);
}
};
template <class T>
struct AcosFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AcosFunctor() { }
CUDA4DNN_DEVICE AcosFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::acos;
return acos(value);
}
};
template <class T>
struct AcoshFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AcoshFunctor() { }
CUDA4DNN_DEVICE AcoshFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::acosh;
return acosh(value);
}
};
template <class T>
struct AsinFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AsinFunctor() { }
CUDA4DNN_DEVICE AsinFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::asin;
return asin(value);
}
};
template <class T>
struct AsinhFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AsinhFunctor() { }
CUDA4DNN_DEVICE AsinhFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::asinh;
return asinh(value);
}
};
template <class T>
struct AtanFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AtanFunctor() { }
CUDA4DNN_DEVICE AtanFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::atan;
return atan(value);
}
};
template <class T>
struct AtanhFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE AtanhFunctor() { }
CUDA4DNN_DEVICE AtanhFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::atanh;
return atanh(value);
}
};
template <class T>
struct CosFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE CosFunctor() { }
CUDA4DNN_DEVICE CosFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::cos;
return cos(value);
}
};
template <class T>
struct CoshFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE CoshFunctor() { }
CUDA4DNN_DEVICE CoshFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::cosh;
return cosh(value);
}
};
template <class T>
struct ErfFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE ErfFunctor() { }
CUDA4DNN_DEVICE ErfFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::erf;
return erf(value);
}
};
template <class T>
struct HardSwishFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE HardSwishFunctor() { }
CUDA4DNN_DEVICE HardSwishFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::clamp; // saturate?
return value * clamp(value / static_cast<T>(6.f) + static_cast<T>(0.5f), static_cast<T>(0.f), static_cast<T>(1.f));
}
};
template <class T>
struct SinFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SinFunctor() { }
CUDA4DNN_DEVICE SinFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::sin;
return sin(value);
}
};
template <class T>
struct SinhFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SinhFunctor() { }
CUDA4DNN_DEVICE SinhFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::sinh;
return sinh(value);
}
};
template <class T>
struct SoftplusFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SoftplusFunctor() { }
CUDA4DNN_DEVICE SoftplusFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::log1pexp;
return log1pexp(value);
}
};
template <class T>
struct SoftsignFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SoftsignFunctor() { }
CUDA4DNN_DEVICE SoftsignFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::abs;
return value / (static_cast<T>(1.f) + abs(value));
}
};
template <class T>
struct TanFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE TanFunctor() { }
CUDA4DNN_DEVICE TanFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::tan;
return tan(value);
}
};
template <class T>
struct CeluFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : alpha(1) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_) : alpha(alpha_) { }
T alpha;
};
CUDA4DNN_DEVICE CeluFunctor() : CeluFunctor(Params{}) { }
CUDA4DNN_DEVICE CeluFunctor(const Params& params) : alpha{params.alpha} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::min;
using csl::device::max;
using csl::device::expm1;
return max(T(0), value) + min(T(0), alpha * expm1(value / alpha));
}
T alpha;
};
template <class T>
struct HardSigmoidFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : alpha(0.2), beta(0.5) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_, T beta_) : alpha(alpha_), beta(beta_) { }
T alpha, beta;
};
CUDA4DNN_DEVICE HardSigmoidFunctor() : HardSigmoidFunctor(Params{}) { }
CUDA4DNN_DEVICE HardSigmoidFunctor(const Params& params): alpha{params.alpha}, beta{params.beta} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::clamp;
return clamp(alpha * value + beta, T(0), T(1));
}
T alpha, beta;
};
template <class T>
struct SeluFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : alpha(1.6732632423543772848170429916717),
gamma(1.0507009873554804934193349852946) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_, T gamma_) : alpha(alpha_), gamma(gamma_) { }
T alpha, gamma;
};
CUDA4DNN_DEVICE SeluFunctor() : SeluFunctor(Params{}) { }
CUDA4DNN_DEVICE SeluFunctor(const Params& params): alpha{params.alpha}, gamma{params.gamma} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::expm1;
return gamma * (value > T(0) ? value : alpha * expm1(value));
}
T alpha, gamma;
};
template <class T>
struct GeluFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE GeluFunctor() { }
CUDA4DNN_DEVICE GeluFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::erf;
return static_cast<T>(0.5f) * value * (static_cast<T>(1.f) + erf(value * static_cast<T>(M_SQRT1_2)));
}
};
template <class T>
struct ThresholdedReluFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : alpha(1) { }
CUDA4DNN_HOST_DEVICE Params(T alpha_) : alpha(alpha_) { }
T alpha;
};
CUDA4DNN_DEVICE ThresholdedReluFunctor() : ThresholdedReluFunctor(Params{}) { }
CUDA4DNN_DEVICE ThresholdedReluFunctor(const Params& params) : alpha{params.alpha} { }
CUDA4DNN_DEVICE T operator()(T value) {
return (value > alpha) ? value : T(0);
}
T alpha;
};
template <class T>
struct PowerFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : exp(1), scale(1), shift(0) { }
CUDA4DNN_HOST_DEVICE Params(T exp_, T scale_, T shift_) : exp(exp_), scale(scale_), shift(shift_) { }
T exp, scale, shift;
};
CUDA4DNN_DEVICE PowerFunctor() : PowerFunctor(Params{}) { }
CUDA4DNN_DEVICE PowerFunctor(const Params& params) : exp{params.exp}, scale{params.scale}, shift{params.shift} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::pow;
return pow(shift + scale * value, exp);
}
T exp, scale, shift;
};
template <class T>
struct ExpFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : normScale(1), normShift(0) { }
CUDA4DNN_HOST_DEVICE Params(T nScale_, T nShift_) : normScale(nScale_), normShift(nShift_) { }
T normScale, normShift;
};
CUDA4DNN_DEVICE ExpFunctor() : ExpFunctor(Params{}) { }
CUDA4DNN_DEVICE ExpFunctor(const Params& params) : normScale{params.normScale}, normShift{params.normShift} { }
CUDA4DNN_DEVICE T operator()(T value) {
using csl::device::fast_exp;
return fast_exp(normShift + normScale * value);
}
T normScale, normShift;
};
template <class T>
struct MaxFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE MaxFunctor() { }
CUDA4DNN_DEVICE MaxFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) {
using csl::device::max;
return max(x, y);
}
};
template <class T>
struct MinFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE MinFunctor() { }
CUDA4DNN_DEVICE MinFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) {
using csl::device::min;
return min(x, y);
}
};
template <class T>
struct SumFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SumFunctor() { }
CUDA4DNN_DEVICE SumFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) { return x + y; }
};
template <class T>
struct ScaledSumFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : scale_x(1), scale_y(1) { }
CUDA4DNN_HOST_DEVICE Params(T scale_x_, T scale_y_) : scale_x(scale_x_), scale_y(scale_y_) { }
T scale_x, scale_y;
};
CUDA4DNN_DEVICE ScaledSumFunctor() : scale_x(1), scale_y(1) { }
CUDA4DNN_DEVICE ScaledSumFunctor(const Params& params) : scale_x{params.scale_x}, scale_y{params.scale_y} { }
CUDA4DNN_DEVICE T operator()(T x, T y) { return scale_x * x + scale_y * y; }
T scale_x, scale_y;
};
template <class T>
struct ProductFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE ProductFunctor() { }
CUDA4DNN_DEVICE ProductFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) { return x * y; }
};
template <class T>
struct DivFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE DivFunctor() { }
CUDA4DNN_DEVICE DivFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) { return x / y; }
};
template <class T>
struct SubFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() { }
};
CUDA4DNN_DEVICE SubFunctor() { }
CUDA4DNN_DEVICE SubFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) { return x - y; }
};
template <class T>
struct SignFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() {}
};
CUDA4DNN_DEVICE SignFunctor() { }
CUDA4DNN_DEVICE SignFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
return value > T(0) ? T(1) : (value < T(0) ? T(-1) : T(0));
}
};
template <class T>
struct ShrinkFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() : bias(0), lambd(0.5) { }
CUDA4DNN_HOST_DEVICE Params(T bias_, T lambd_) : bias(bias_), lambd(lambd_) { }
T bias, lambd;
};
CUDA4DNN_DEVICE ShrinkFunctor() : ShrinkFunctor(Params{}) { }
CUDA4DNN_DEVICE ShrinkFunctor(const Params& params) : bias{params.bias}, lambd{params.lambd} { }
CUDA4DNN_DEVICE T operator()(T value) {
return value > lambd ? value - bias : (value < -lambd ? value + bias : T(0));
}
T bias, lambd;
};
template <class T>
struct ReciprocalFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() {}
};
CUDA4DNN_DEVICE ReciprocalFunctor() { }
CUDA4DNN_DEVICE ReciprocalFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T value) {
return T(1.f)/value;
}
};
template <class T>
struct ModFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() {}
};
CUDA4DNN_DEVICE ModFunctor() { }
CUDA4DNN_DEVICE ModFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) {
int res = (int)x % (int)y;
T zero = T(0);
if ((res > (int)zero && y < zero) || (res < (int)zero && y > zero)) {
res += (int)y;
}
return res;
}
};
template <class T>
struct FModFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() {}
};
CUDA4DNN_DEVICE FModFunctor() { }
CUDA4DNN_DEVICE FModFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) {
using csl::device::fmod;
return fmod(x, y);
}
};
template <class T>
struct PowFunctor {
struct Params {
CUDA4DNN_HOST_DEVICE Params() {}
};
CUDA4DNN_DEVICE PowFunctor() { }
CUDA4DNN_DEVICE PowFunctor(const Params& params) { }
CUDA4DNN_DEVICE T operator()(T x, T y) {
using csl::device::pow;
return pow(x, y);
}
};
}}}} /* namespace cv::dnn::cuda4dnn::kernels */
#endif /* OPENCV_DNN_SRC_CUDA_FUNCTORS_HPP */
|