/usr/include/libint2/util/vector_x86.h is in libint2-dev 2.3.0~beta3-2.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 | /*
* This file is a part of Libint.
* Copyright (C) 2004-2014 Edward F. Valeev
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Library General Public License, version 2,
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU Library General Public License
* along with this program. If not, see http://www.gnu.org/licenses/.
*
*/
#ifndef _libint2_src_lib_libint_vectorx86_h_
#define _libint2_src_lib_libint_vectorx86_h_
#include <cstring>
#include <libint2/util/cxxstd.h>
#include <libint2/util/type_traits.h>
#ifdef __SSE2__
#include <emmintrin.h>
#include <immintrin.h>
#include <cmath>
#include <iostream>
namespace libint2 { namespace simd {
/**
* SIMD vector of 2 double-precision floating-point real numbers, operations on which use SSE2 instructions
* available on all recent x86 hardware
*/
struct VectorSSEDouble {
typedef double T;
__m128d d;
/**
* creates a vector of default-initialized values.
*/
VectorSSEDouble() {}
/** Initializes all elements to the same value
* @param a the value to which all elements will be set
*/
VectorSSEDouble(T a) {
d = _mm_set_pd(a, a);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorSSEDouble(T (&a)[2]) {
d = _mm_loadu_pd(&a[0]);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorSSEDouble(T a0, T a1) {
d = _mm_set_pd(a1, a0);
}
/**
* converts a 128-bit SSE double vector type to VectorSSEDouble
*/
VectorSSEDouble(__m128d a) {
d = a;
}
VectorSSEDouble& operator=(T a) {
d = _mm_set_pd(a, a);
return *this;
}
VectorSSEDouble& operator+=(VectorSSEDouble a) {
d = _mm_add_pd(d, a.d);
return *this;
}
VectorSSEDouble& operator-=(VectorSSEDouble a) {
d = _mm_sub_pd(d, a.d);
return *this;
}
VectorSSEDouble operator-() const {
// TODO improve using bitflips
const static __m128d minus_one = _mm_set_pd(-1.0, -1.0);;
VectorSSEDouble result;
result.d = _mm_mul_pd(this->d, minus_one);
return result;
}
#if LIBINT2_CPLUSPLUS_STD >= 2011
explicit
#endif
operator double() const {
double d0[2];
::memcpy(&(d0[0]), &d, sizeof(__m128d));
// this segfaults on OS X
//_mm_storeu_pd(&(d0[0]), d);
// // aligned alternative requires C++11's alignas, but efficiency should not matter here
// alignas(__m128d) double d0[2];
// _mm_store_pd(&(d0[0]), d);
return d0[0];
}
/// implicit conversion to SSE 128-bit "register"
operator __m128d() const {
return d;
}
/// loads \c a to \c this
void load(T const* a) {
d = _mm_loadu_pd(a);
}
/// loads \c a to \c this \sa load()
/// @note \c a must be aligned to 16 bytes
void load_aligned(T const* a) {
d = _mm_load_pd(a);
}
/// writes \c this to \c a
void convert(T* a) const {
_mm_storeu_pd(&a[0], d);
}
/// writes \c this to \c a
/// @note \c a must be aligned to 16 bytes
void convert_aligned(T* a) const {
_mm_store_pd(&a[0], d);
}
};
//@{ arithmetic operators
inline VectorSSEDouble operator*(double a, VectorSSEDouble b) {
VectorSSEDouble c;
VectorSSEDouble _a(a);
c.d = _mm_mul_pd(_a.d, b.d);
return c;
}
inline VectorSSEDouble operator*(VectorSSEDouble a, double b) {
VectorSSEDouble c;
VectorSSEDouble _b(b);
c.d = _mm_mul_pd(a.d, _b.d);
return c;
}
inline VectorSSEDouble operator*(int a, VectorSSEDouble b) {
if (a == 1)
return b;
else {
VectorSSEDouble c;
VectorSSEDouble _a((double)a);
c.d = _mm_mul_pd(_a.d, b.d);
return c;
}
}
inline VectorSSEDouble operator*(VectorSSEDouble a, int b) {
if (b == 1)
return a;
else {
VectorSSEDouble c;
VectorSSEDouble _b((double)b);
c.d = _mm_mul_pd(a.d, _b.d);
return c;
}
}
inline VectorSSEDouble operator*(VectorSSEDouble a, VectorSSEDouble b) {
VectorSSEDouble c;
c.d = _mm_mul_pd(a.d, b.d);
return c;
}
inline VectorSSEDouble operator+(VectorSSEDouble a, VectorSSEDouble b) {
VectorSSEDouble c;
c.d = _mm_add_pd(a.d, b.d);
return c;
}
inline VectorSSEDouble operator-(VectorSSEDouble a, VectorSSEDouble b) {
VectorSSEDouble c;
c.d = _mm_sub_pd(a.d, b.d);
return c;
}
inline VectorSSEDouble operator/(VectorSSEDouble a, VectorSSEDouble b) {
VectorSSEDouble c;
c.d = _mm_div_pd(a.d, b.d);
return c;
}
#if defined(__FMA__)
inline VectorSSEDouble fma_plus(VectorSSEDouble a, VectorSSEDouble b, VectorSSEDouble c) {
VectorSSEDouble d;
d.d = _mm_fmadd_pd(a.d, b.d, c.d);
return d;
}
inline VectorSSEDouble fma_minus(VectorSSEDouble a, VectorSSEDouble b, VectorSSEDouble c) {
VectorSSEDouble d;
d.d = _mm_fmsub_pd(a.d, b.d, c.d);
return d;
}
#elif defined(__FMA4__)
inline VectorSSEDouble fma_plus(VectorSSEDouble a, VectorSSEDouble b, VectorSSEDouble c) {
VectorSSEDouble d;
d.d = _mm_macc_pd(a.d, b.d, c.d);
return d;
}
inline VectorSSEDouble fma_minus(VectorSSEDouble a, VectorSSEDouble b, VectorSSEDouble c) {
VectorSSEDouble d;
d.d = _mm_msub_pd(a.d, b.d, c.d);
return d;
}
#endif
/// Horizontal add
/// @param a input vector = {a[0], a[1]}
/// @return a[0] + a[1]
inline double horizontal_add (VectorSSEDouble const & a) {
// Agner Fog's version
#if defined(__SSE3__)
__m128d t1 = _mm_hadd_pd(a,a);
return _mm_cvtsd_f64(t1);
#else // SSE2 only
__m128 t0 = _mm_castpd_ps(a);
__m128d t1 = _mm_castps_pd(_mm_movehl_ps(t0,t0));
__m128d t2 = _mm_add_sd(a,t1);
return _mm_cvtsd_f64(t2);
#endif
}
/// Horizontal add of a pair of vectors
/// @param a input vector = {a[0], a[1]}
/// @param b input vector = {b[0], b[1]}
/// @return {a[0] + a[1], b[0] + b[1]}
inline VectorSSEDouble horizontal_add (VectorSSEDouble const & a, VectorSSEDouble const & b) {
#if defined(__SSE3__)
return _mm_hadd_pd(a, b);
#else // will be very inefficient without SSE3
return VectorSSEDouble(horizontal_add(a), horizontal_add(b));
#endif
}
//@}
//@{ standard functions
inline VectorSSEDouble exp(VectorSSEDouble a) {
#if HAVE_INTEL_SVML
VectorSSEDouble result;
result.d = _mm_exp_pd(a.d);
#else
double a_d[2]; a.convert(a_d);
for(int i=0; i<2; ++i) a_d[i] = std::exp(a_d[i]);
VectorSSEDouble result(a_d);
#endif
return result;
}
inline VectorSSEDouble sqrt(VectorSSEDouble a) {
#if HAVE_INTEL_SVML
VectorSSEDouble result;
result.d = _mm_sqrt_pd(a.d);
#else
double a_d[2]; a.convert(a_d);
for(int i=0; i<2; ++i) a_d[i] = std::sqrt(a_d[i]);
VectorSSEDouble result(a_d);
#endif
return result;
}
inline VectorSSEDouble erf(VectorSSEDouble a) {
#if HAVE_INTEL_SVML
VectorSSEDouble result;
result.d = _mm_erf_pd(a.d);
#else
double a_d[2]; a.convert(a_d);
for(int i=0; i<2; ++i) a_d[i] = ::erf(a_d[i]);
VectorSSEDouble result(a_d);
#endif
return result;
}
inline VectorSSEDouble erfc(VectorSSEDouble a) {
#if HAVE_INTEL_SVML
VectorSSEDouble result;
result.d = _mm_erfc_pd(a.d);
#else
double a_d[2]; a.convert(a_d);
for(int i=0; i<2; ++i) a_d[i] = ::erfc(a_d[i]);
VectorSSEDouble result(a_d);
#endif
return result;
}
//@}
};}; // namespace libint2::simd
//@{ standard stream operations
inline std::ostream& operator<<(std::ostream& os, libint2::simd::VectorSSEDouble a) {
double ad[2];
a.convert(ad);
os << "{" << ad[0] << "," << ad[1] << "}";
return os;
}
//@}
namespace libint2 {
//@{ vector traits of VectorSSEDouble
template <>
struct is_vector<simd::VectorSSEDouble> {
static const bool value = true;
};
template <>
struct vector_traits<simd::VectorSSEDouble> {
typedef double value_type;
static const size_t extent = 2;
};
//@}
} // namespace libint2
#endif // SSE2-only
#ifdef __SSE__
#include <xmmintrin.h>
namespace libint2 { namespace simd {
/**
* SIMD vector of 4 single-precision floating-point real numbers, operations on which use SSE instructions
* available on all recent x86 hardware.
*/
struct VectorSSEFloat {
typedef float T;
__m128 d;
/**
* creates a vector of default-initialized values.
*/
VectorSSEFloat() {}
/** Initializes all elements to the same value
* @param a the value to which all elements will be set
*/
VectorSSEFloat(T a) {
d = _mm_set_ps(a, a, a, a);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorSSEFloat(T (&a)[4]) {
d = _mm_loadu_ps(&a[0]);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorSSEFloat(T a0, T a1, T a2, T a3) {
d = _mm_set_ps(a3, a2, a1, a0);
}
VectorSSEFloat& operator=(T a) {
d = _mm_set_ps(a, a, a, a);
return *this;
}
VectorSSEFloat& operator+=(VectorSSEFloat a) {
d = _mm_add_ps(d, a.d);
return *this;
}
VectorSSEFloat& operator-=(VectorSSEFloat a) {
d = _mm_sub_ps(d, a.d);
return *this;
}
VectorSSEFloat operator-() const {
// TODO improve using bitflips
const static __m128 minus_one = _mm_set_ps(-1.0, -1.0, -1.0, -1.0);;
VectorSSEFloat result;
result.d = _mm_mul_ps(this->d, minus_one);
return result;
}
#if LIBINT2_CPLUSPLUS_STD >= 2011
explicit
#endif
operator float() const {
float d0[4];
::memcpy(&(d0[0]), &d, sizeof(__m128));
// this segfaults on OS X
//_mm_storeu_ps(&(d0[0]), d);
// // aligned alternative requires C++11's alignas, but efficiency should not matter here
// alignas(__m128) float d0[4];
// _mm_store_ps(&(d0[0]), d);
return d0[0];
}
#if LIBINT2_CPLUSPLUS_STD >= 2011
explicit
#endif
operator double() const {
const float result_flt = this->operator float();
return result_flt;
}
/// implicit conversion to SSE 128-bit "register"
operator __m128() const {
return d;
}
/// loads \c a to \c this
void load(T const* a) {
d = _mm_loadu_ps(a);
}
/// loads \c a to \c this \sa load()
/// @note \c a must be aligned to 16 bytes
void load_aligned(T const* a) {
d = _mm_load_ps(a);
}
/// writes \c this to \c a
void convert(T* a) const {
_mm_storeu_ps(&a[0], d);
}
/// writes \c this to \c a
/// @note \c a must be aligned to 32 bytes
void convert_aligned(T* a) const {
_mm_store_ps(&a[0], d);
}
};
//@{ arithmetic operators
inline VectorSSEFloat operator*(float a, VectorSSEFloat b) {
VectorSSEFloat c;
VectorSSEFloat _a(a);
c.d = _mm_mul_ps(_a.d, b.d);
return c;
}
inline VectorSSEFloat operator*(VectorSSEFloat a, float b) {
VectorSSEFloat c;
VectorSSEFloat _b(b);
c.d = _mm_mul_ps(a.d, _b.d);
return c;
}
// narrows a!
inline VectorSSEFloat operator*(double a, VectorSSEFloat b) {
VectorSSEFloat c;
VectorSSEFloat _a((float)a);
c.d = _mm_mul_ps(_a.d, b.d);
return c;
}
// narrows b!
inline VectorSSEFloat operator*(VectorSSEFloat a, double b) {
VectorSSEFloat c;
VectorSSEFloat _b((float)b);
c.d = _mm_mul_ps(a.d, _b.d);
return c;
}
inline VectorSSEFloat operator*(int a, VectorSSEFloat b) {
if (a == 1)
return b;
else {
VectorSSEFloat c;
VectorSSEFloat _a((float)a);
c.d = _mm_mul_ps(_a.d, b.d);
return c;
}
}
inline VectorSSEFloat operator*(VectorSSEFloat a, int b) {
if (b == 1)
return a;
else {
VectorSSEFloat c;
VectorSSEFloat _b((float)b);
c.d = _mm_mul_ps(a.d, _b.d);
return c;
}
}
inline VectorSSEFloat operator*(VectorSSEFloat a, VectorSSEFloat b) {
VectorSSEFloat c;
c.d = _mm_mul_ps(a.d, b.d);
return c;
}
inline VectorSSEFloat operator+(VectorSSEFloat a, VectorSSEFloat b) {
VectorSSEFloat c;
c.d = _mm_add_ps(a.d, b.d);
return c;
}
inline VectorSSEFloat operator-(VectorSSEFloat a, VectorSSEFloat b) {
VectorSSEFloat c;
c.d = _mm_sub_ps(a.d, b.d);
return c;
}
inline VectorSSEFloat operator/(VectorSSEFloat a, VectorSSEFloat b) {
VectorSSEFloat c;
c.d = _mm_div_ps(a.d, b.d);
return c;
}
#if defined(__FMA__)
inline VectorSSEFloat fma_plus(VectorSSEFloat a, VectorSSEFloat b, VectorSSEFloat c) {
VectorSSEFloat d;
d.d = _mm_fmadd_ps(a.d, b.d, c.d);
return d;
}
inline VectorSSEFloat fma_minus(VectorSSEFloat a, VectorSSEFloat b, VectorSSEFloat c) {
VectorSSEFloat d;
d.d = _mm_fmsub_ps(a.d, b.d, c.d);
return d;
}
#elif defined(__FMA4__)
inline VectorSSEFloat fma_plus(VectorSSEFloat a, VectorSSEFloat b, VectorSSEFloat c) {
VectorSSEFloat d;
d.d = _mm_macc_ps(a.d, b.d, c.d);
return d;
}
inline VectorSSEFloat fma_minus(VectorSSEFloat a, VectorSSEFloat b, VectorSSEFloat c) {
VectorSSEFloat d;
d.d = _mm_msub_ps(a.d, b.d, c.d);
return d;
}
#endif
//@}
//@{ standard functions
inline VectorSSEFloat exp(VectorSSEFloat a) {
#if HAVE_INTEL_SVML
VectorSSEFloat result;
result.d = _mm_exp_ps(a.d);
#else
float a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = std::exp(a_d[i]);
VectorSSEFloat result(a_d);
#endif
return result;
}
inline VectorSSEFloat sqrt(VectorSSEFloat a) {
#if HAVE_INTEL_SVML
VectorSSEFloat result;
result.d = _mm_sqrt_ps(a.d);
#else
float a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = std::sqrt(a_d[i]);
VectorSSEFloat result(a_d);
#endif
return result;
}
inline VectorSSEFloat erf(VectorSSEFloat a) {
#if HAVE_INTEL_SVML
VectorSSEFloat result;
result.d = _mm_erf_ps(a.d);
#else
float a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::erf(a_d[i]);
VectorSSEFloat result(a_d);
#endif
return result;
}
inline VectorSSEFloat erfc(VectorSSEFloat a) {
#if HAVE_INTEL_SVML
VectorSSEFloat result;
result.d = _mm_erfc_ps(a.d);
#else
float a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::erfc(a_d[i]);
VectorSSEFloat result(a_d);
#endif
return result;
}
//@}
};}; // namespace libint2::simd
//@{ standard stream operations
inline std::ostream& operator<<(std::ostream& os, libint2::simd::VectorSSEFloat a) {
float ad[4];
a.convert(ad);
os << "{" << ad[0] << "," << ad[1] << "," << ad[2] << "," << ad[3] << "}";
return os;
}
//@}
namespace libint2 {
//@{ vector traits of VectorSSEFloat
template <>
struct is_vector<simd::VectorSSEFloat> {
static const bool value = true;
};
template <>
struct vector_traits<simd::VectorSSEFloat> {
typedef float value_type;
static const size_t extent = 4;
};
//@}
} // namespace libint2
#endif // SSE-only
#ifdef __AVX__
#include <immintrin.h>
namespace libint2 { namespace simd {
/**
* SIMD vector of 4 double-precision floating-point real numbers, operations on which use AVX instructions
* available on recent x86 hardware from Intel (starting with Sandy Bridge processors released in 2011)
* and AMD (starting with Bulldozer released in 2011).
*/
struct VectorAVXDouble {
typedef double T;
__m256d d;
/**
* creates a vector of default-initialized values.
*/
VectorAVXDouble() {}
/** Initializes all elements to the same value
* @param a the value to which all elements will be set
*/
VectorAVXDouble(T a) {
d = _mm256_set_pd(a, a, a, a);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorAVXDouble(T (&a)[4]) {
d = _mm256_loadu_pd(&a[0]);
}
/**
* creates a vector of values initialized by an ordinary static-sized array
*/
VectorAVXDouble(T a0, T a1, T a2, T a3) {
d = _mm256_set_pd(a3, a2, a1, a0);
}
/**
* converts a 256-bit AVX double vector type to VectorAVXDouble
*/
VectorAVXDouble(__m256d a) {
d = a;
}
VectorAVXDouble& operator=(T a) {
d = _mm256_set_pd(a, a, a, a);
return *this;
}
VectorAVXDouble& operator+=(VectorAVXDouble a) {
d = _mm256_add_pd(d, a.d);
return *this;
}
VectorAVXDouble& operator-=(VectorAVXDouble a) {
d = _mm256_sub_pd(d, a.d);
return *this;
}
VectorAVXDouble operator-() const {
// TODO improve using bitflips
const static __m256d minus_one = _mm256_set_pd(-1.0, -1.0, -1.0, -1.0);;
VectorAVXDouble result;
result.d = _mm256_mul_pd(this->d, minus_one);
return result;
}
#if LIBINT2_CPLUSPLUS_STD >= 2011
explicit
#endif
operator double() const {
double d0[4];
::memcpy(&(d0[0]), &d, sizeof(__m256d));
// this segfaults on OS X
// _mm256_storeu_pd(&d0[0], d);
// // aligned alternative requires C++11's alignas, but efficiency should not matter here
// // alignas(__m256d) double d0[4];
// // _mm256_store_pd(&(d0[0]), d);
return d0[0];
}
/// implicit conversion to AVX 256-bit "register"
operator __m256d() const {
return d;
}
/// loads \c a to \c this
void load(T const* a) {
d = _mm256_loadu_pd(a);
}
/// loads \c a to \c this \sa load()
/// @note \c a must be aligned to 32 bytes
void load_aligned(T const* a) {
d = _mm256_load_pd(a);
}
/// writes \c this to \c a
void convert(T* a) const {
_mm256_storeu_pd(&a[0], d);
}
/// writes \c this to \c a
/// @note \c a must be aligned to 32 bytes
void convert_aligned(T* a) const {
_mm256_store_pd(&a[0], d);
}
};
//@{ arithmetic operators
inline VectorAVXDouble operator*(double a, VectorAVXDouble b) {
VectorAVXDouble c;
VectorAVXDouble _a(a);
c.d = _mm256_mul_pd(_a.d, b.d);
return c;
}
inline VectorAVXDouble operator*(VectorAVXDouble a, double b) {
VectorAVXDouble c;
VectorAVXDouble _b(b);
c.d = _mm256_mul_pd(a.d, _b.d);
return c;
}
inline VectorAVXDouble operator*(int a, VectorAVXDouble b) {
if (a == 1)
return b;
else {
VectorAVXDouble c;
VectorAVXDouble _a((double)a);
c.d = _mm256_mul_pd(_a.d, b.d);
return c;
}
}
inline VectorAVXDouble operator*(VectorAVXDouble a, int b) {
if (b == 1)
return a;
else {
VectorAVXDouble c;
VectorAVXDouble _b((double)b);
c.d = _mm256_mul_pd(a.d, _b.d);
return c;
}
}
inline VectorAVXDouble operator*(VectorAVXDouble a, VectorAVXDouble b) {
VectorAVXDouble c;
c.d = _mm256_mul_pd(a.d, b.d);
return c;
}
inline VectorAVXDouble operator+(VectorAVXDouble a, VectorAVXDouble b) {
VectorAVXDouble c;
c.d = _mm256_add_pd(a.d, b.d);
return c;
}
inline VectorAVXDouble operator+(int a, VectorAVXDouble b) {
if(a == 0){
return b;
}
else {
VectorAVXDouble c;
VectorAVXDouble _a = (static_cast<double>(a));
c.d = _mm256_add_pd(_a.d,b.d);
return c;
}
}
inline VectorAVXDouble operator+(VectorAVXDouble a, int b) {
if(b == 0){
return a;
}
else {
VectorAVXDouble c;
VectorAVXDouble _b = (static_cast<double>(b));
c.d = _mm256_add_pd(a.d,_b.d);
return c;
}
}
inline VectorAVXDouble operator-(VectorAVXDouble a, VectorAVXDouble b) {
VectorAVXDouble c;
c.d = _mm256_sub_pd(a.d, b.d);
return c;
}
inline VectorAVXDouble operator/(VectorAVXDouble a, VectorAVXDouble b) {
VectorAVXDouble c;
c.d = _mm256_div_pd(a.d, b.d);
return c;
}
#if defined(__FMA__)
inline VectorAVXDouble fma_plus(VectorAVXDouble a, VectorAVXDouble b, VectorAVXDouble c) {
VectorAVXDouble d;
d.d = _mm256_fmadd_pd(a.d, b.d, c.d);
return d;
}
inline VectorAVXDouble fma_minus(VectorAVXDouble a, VectorAVXDouble b, VectorAVXDouble c) {
VectorAVXDouble d;
d.d = _mm256_fmsub_pd(a.d, b.d, c.d);
return d;
}
#elif defined(__FMA4__)
inline VectorAVXDouble fma_plus(VectorAVXDouble a, VectorAVXDouble b, VectorAVXDouble c) {
VectorAVXDouble d;
d.d = _mm256_facc_pd(a.d, b.d, c.d);
return d;
}
inline VectorAVXDouble fma_minus(VectorAVXDouble a, VectorAVXDouble b, VectorAVXDouble c) {
VectorAVXDouble d;
d.d = _mm256_fsub_pd(a.d, b.d, c.d);
return d;
}
#endif
/// Horizontal add
/// @param a input vector = {a[0], a[1], a[2], a[3]}
/// @return a[0] + a[1] + a[2] + a[3]
inline double horizontal_add (VectorAVXDouble const & a) {
__m256d s = _mm256_hadd_pd(a,a);
return ((double*)&s)[0] + ((double*)&s)[2];
// Agner Fog's version
// __m256d t1 = _mm256_hadd_pd(a,a);
// __m128d t2 = _mm256_extractf128_pd(t1,1);
// __m128d t3 = _mm_add_sd(_mm256_castpd256_pd128(t1),t2);
// return _mm_cvtsd_f64(t3);
}
/// Horizontal add of a pair of vectors
/// @param a input vector = {a[0], a[1], a[2], a[3]}
/// @param b input vector = {b[0], b[1], b[2], b[3]}
/// @return {a[0] + a[1] + a[2] + a[3], b[0] + b[1] + b[2] + b[3]}
inline VectorSSEDouble horizontal_add (VectorAVXDouble const & a, VectorAVXDouble const & b) {
// solution from http://stackoverflow.com/questions/9775538/fastest-way-to-do-horizontal-vector-sum-with-avx-instructions
__m256d sum = _mm256_hadd_pd(a, b);
// extract upper 128 bits of result
__m128d sum_high = _mm256_extractf128_pd(sum, 1);
// add upper 128 bits of sum to its lower 128 bits
return _mm_add_pd(sum_high, _mm256_castpd256_pd128(sum));
}
/// Horizontal add of a set of 4 vectors
/// @param a input vector = {a[0], a[1], a[2], a[3]}
/// @param b input vector = {b[0], b[1], b[2], b[3]}
/// @param c input vector = {c[0], c[1], c[2], c[3]}
/// @param d input vector = {d[0], d[1], d[2], d[3]}
/// @return {a[0] + a[1] + a[2] + a[3], b[0] + b[1] + b[2] + b[3], c[0] + c[1] + c[2] + c[3], d[0] + d[1] + d[2] + d[3]}
inline VectorAVXDouble horizontal_add (VectorAVXDouble const & a,
VectorAVXDouble const & b,
VectorAVXDouble const & c,
VectorAVXDouble const & d) {
// solution from http://stackoverflow.com/questions/10833234/4-horizontal-double-precision-sums-in-one-go-with-avx?lq=1
// {a[0]+a[1], b[0]+b[1], a[2]+a[3], b[2]+b[3]}
__m256d sumab = _mm256_hadd_pd(a, b);
// {c[0]+c[1], d[0]+d[1], c[2]+c[3], d[2]+d[3]}
__m256d sumcd = _mm256_hadd_pd(c, d);
// {a[0]+a[1], b[0]+b[1], c[2]+c[3], d[2]+d[3]}
__m256d blend = _mm256_blend_pd(sumab, sumcd, 0b1100);
// {a[2]+a[3], b[2]+b[3], c[0]+c[1], d[0]+d[1]}
__m256d perm = _mm256_permute2f128_pd(sumab, sumcd, 0x21);
return _mm256_add_pd(perm, blend);
}
//@}
//@{ standard functions
inline VectorAVXDouble exp(VectorAVXDouble a) {
#if HAVE_INTEL_SVML
VectorAVXDouble result;
result.d = _mm256_exp_pd(a.d);
#else
double a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::exp(a_d[i]);
VectorAVXDouble result(a_d);
#endif
return result;
}
inline VectorAVXDouble sqrt(VectorAVXDouble a) {
#if HAVE_INTEL_SVML
VectorAVXDouble result;
result.d = _mm256_sqrt_pd(a.d);
#else
double a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::sqrt(a_d[i]);
VectorAVXDouble result(a_d);
#endif
return result;
}
inline VectorAVXDouble erf(VectorAVXDouble a) {
#if HAVE_INTEL_SVML
VectorAVXDouble result;
result.d = _mm256_erf_pd(a.d);
#else
double a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::erf(a_d[i]);
VectorAVXDouble result(a_d);
#endif
return result;
}
inline VectorAVXDouble erfc(VectorAVXDouble a) {
#if HAVE_INTEL_SVML
VectorAVXDouble result;
result.d = _mm256_erfc_pd(a.d);
#else
double a_d[4]; a.convert(a_d);
for(int i=0; i<4; ++i) a_d[i] = ::erfc(a_d[i]);
VectorAVXDouble result(a_d);
#endif
return result;
}
//@}
};}; // namespace libint2::simd
//@{ standard stream operations
inline std::ostream& operator<<(std::ostream& os, libint2::simd::VectorAVXDouble a) {
double ad[4];
a.convert(ad);
os << "{" << ad[0] << "," << ad[1] << "," << ad[2] << "," << ad[3] << "}";
return os;
}
//@}
namespace libint2 {
//@{ vector traits of VectorAVXDouble
template <>
struct is_vector<simd::VectorAVXDouble> {
static const bool value = true;
};
template <>
struct vector_traits<simd::VectorAVXDouble> {
typedef double value_type;
static const size_t extent = 4;
};
//@}
} // namespace libint2
#endif // AVX-only
#ifdef LIBINT2_HAVE_AGNER_VECTORCLASS
#include <vectorclass.h>
#endif
#endif // header guard
|