1 /*===---- __clang_hip_cmath.h - HIP cmath decls -----------------------------=== 2 * 3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 * See https://llvm.org/LICENSE.txt for license information. 5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 * 7 *===-----------------------------------------------------------------------=== 8 */ 9 10 #ifndef __CLANG_HIP_CMATH_H__ 11 #define __CLANG_HIP_CMATH_H__ 12 13 #if !defined(__HIP__) && !defined(__OPENMP_AMDGCN__) 14 #error "This file is for HIP and OpenMP AMDGCN device compilation only." 15 #endif 16 17 #if !defined(__HIPCC_RTC__) 18 #if defined(__cplusplus) 19 #include <limits> 20 #include <type_traits> 21 #include <utility> 22 #endif 23 #include <limits.h> 24 #include <stdint.h> 25 #endif // !defined(__HIPCC_RTC__) 26 27 #pragma push_macro("__DEVICE__") 28 #pragma push_macro("__CONSTEXPR__") 29 #ifdef __OPENMP_AMDGCN__ 30 #define __DEVICE__ static __attribute__((always_inline, nothrow)) 31 #define __CONSTEXPR__ constexpr 32 #else 33 #define __DEVICE__ static __device__ inline __attribute__((always_inline)) 34 #define __CONSTEXPR__ 35 #endif // __OPENMP_AMDGCN__ 36 37 // Start with functions that cannot be defined by DEF macros below. 38 #if defined(__cplusplus) 39 #if defined __OPENMP_AMDGCN__ 40 __DEVICE__ __CONSTEXPR__ float fabs(float __x) { return ::fabsf(__x); } 41 __DEVICE__ __CONSTEXPR__ float sin(float __x) { return ::sinf(__x); } 42 __DEVICE__ __CONSTEXPR__ float cos(float __x) { return ::cosf(__x); } 43 #endif 44 __DEVICE__ __CONSTEXPR__ double abs(double __x) { return ::fabs(__x); } 45 __DEVICE__ __CONSTEXPR__ float abs(float __x) { return ::fabsf(__x); } 46 __DEVICE__ __CONSTEXPR__ long long abs(long long __n) { return ::llabs(__n); } 47 __DEVICE__ __CONSTEXPR__ long abs(long __n) { return ::labs(__n); } 48 __DEVICE__ __CONSTEXPR__ float fma(float __x, float __y, float __z) { 49 return ::fmaf(__x, __y, __z); 50 } 51 #if !defined(__HIPCC_RTC__) 52 // The value returned by fpclassify is platform dependent, therefore it is not 53 // supported by hipRTC. 54 __DEVICE__ __CONSTEXPR__ int fpclassify(float __x) { 55 return __builtin_fpclassify(FP_NAN, FP_INFINITE, FP_NORMAL, FP_SUBNORMAL, 56 FP_ZERO, __x); 57 } 58 __DEVICE__ __CONSTEXPR__ int fpclassify(double __x) { 59 return __builtin_fpclassify(FP_NAN, FP_INFINITE, FP_NORMAL, FP_SUBNORMAL, 60 FP_ZERO, __x); 61 } 62 #endif // !defined(__HIPCC_RTC__) 63 64 __DEVICE__ __CONSTEXPR__ float frexp(float __arg, int *__exp) { 65 return ::frexpf(__arg, __exp); 66 } 67 68 #if defined(__OPENMP_AMDGCN__) 69 // For OpenMP we work around some old system headers that have non-conforming 70 // `isinf(float)` and `isnan(float)` implementations that return an `int`. We do 71 // this by providing two versions of these functions, differing only in the 72 // return type. To avoid conflicting definitions we disable implicit base 73 // function generation. That means we will end up with two specializations, one 74 // per type, but only one has a base function defined by the system header. 75 #pragma omp begin declare variant match( \ 76 implementation = {extension(disable_implicit_base)}) 77 78 // FIXME: We lack an extension to customize the mangling of the variants, e.g., 79 // add a suffix. This means we would clash with the names of the variants 80 // (note that we do not create implicit base functions here). To avoid 81 // this clash we add a new trait to some of them that is always true 82 // (this is LLVM after all ;)). It will only influence the mangled name 83 // of the variants inside the inner region and avoid the clash. 84 #pragma omp begin declare variant match(implementation = {vendor(llvm)}) 85 86 __DEVICE__ __CONSTEXPR__ int isinf(float __x) { return ::__isinff(__x); } 87 __DEVICE__ __CONSTEXPR__ int isinf(double __x) { return ::__isinf(__x); } 88 __DEVICE__ __CONSTEXPR__ int isfinite(float __x) { return ::__finitef(__x); } 89 __DEVICE__ __CONSTEXPR__ int isfinite(double __x) { return ::__finite(__x); } 90 __DEVICE__ __CONSTEXPR__ int isnan(float __x) { return ::__isnanf(__x); } 91 __DEVICE__ __CONSTEXPR__ int isnan(double __x) { return ::__isnan(__x); } 92 93 #pragma omp end declare variant 94 #endif // defined(__OPENMP_AMDGCN__) 95 96 __DEVICE__ __CONSTEXPR__ bool isinf(float __x) { return ::__isinff(__x); } 97 __DEVICE__ __CONSTEXPR__ bool isinf(double __x) { return ::__isinf(__x); } 98 __DEVICE__ __CONSTEXPR__ bool isfinite(float __x) { return ::__finitef(__x); } 99 __DEVICE__ __CONSTEXPR__ bool isfinite(double __x) { return ::__finite(__x); } 100 __DEVICE__ __CONSTEXPR__ bool isnan(float __x) { return ::__isnanf(__x); } 101 __DEVICE__ __CONSTEXPR__ bool isnan(double __x) { return ::__isnan(__x); } 102 103 #if defined(__OPENMP_AMDGCN__) 104 #pragma omp end declare variant 105 #endif // defined(__OPENMP_AMDGCN__) 106 107 __DEVICE__ __CONSTEXPR__ bool isgreater(float __x, float __y) { 108 return __builtin_isgreater(__x, __y); 109 } 110 __DEVICE__ __CONSTEXPR__ bool isgreater(double __x, double __y) { 111 return __builtin_isgreater(__x, __y); 112 } 113 __DEVICE__ __CONSTEXPR__ bool isgreaterequal(float __x, float __y) { 114 return __builtin_isgreaterequal(__x, __y); 115 } 116 __DEVICE__ __CONSTEXPR__ bool isgreaterequal(double __x, double __y) { 117 return __builtin_isgreaterequal(__x, __y); 118 } 119 __DEVICE__ __CONSTEXPR__ bool isless(float __x, float __y) { 120 return __builtin_isless(__x, __y); 121 } 122 __DEVICE__ __CONSTEXPR__ bool isless(double __x, double __y) { 123 return __builtin_isless(__x, __y); 124 } 125 __DEVICE__ __CONSTEXPR__ bool islessequal(float __x, float __y) { 126 return __builtin_islessequal(__x, __y); 127 } 128 __DEVICE__ __CONSTEXPR__ bool islessequal(double __x, double __y) { 129 return __builtin_islessequal(__x, __y); 130 } 131 __DEVICE__ __CONSTEXPR__ bool islessgreater(float __x, float __y) { 132 return __builtin_islessgreater(__x, __y); 133 } 134 __DEVICE__ __CONSTEXPR__ bool islessgreater(double __x, double __y) { 135 return __builtin_islessgreater(__x, __y); 136 } 137 __DEVICE__ __CONSTEXPR__ bool isnormal(float __x) { 138 return __builtin_isnormal(__x); 139 } 140 __DEVICE__ __CONSTEXPR__ bool isnormal(double __x) { 141 return __builtin_isnormal(__x); 142 } 143 __DEVICE__ __CONSTEXPR__ bool isunordered(float __x, float __y) { 144 return __builtin_isunordered(__x, __y); 145 } 146 __DEVICE__ __CONSTEXPR__ bool isunordered(double __x, double __y) { 147 return __builtin_isunordered(__x, __y); 148 } 149 __DEVICE__ __CONSTEXPR__ float modf(float __x, float *__iptr) { 150 return ::modff(__x, __iptr); 151 } 152 __DEVICE__ __CONSTEXPR__ float pow(float __base, int __iexp) { 153 return ::powif(__base, __iexp); 154 } 155 __DEVICE__ __CONSTEXPR__ double pow(double __base, int __iexp) { 156 return ::powi(__base, __iexp); 157 } 158 __DEVICE__ __CONSTEXPR__ float remquo(float __x, float __y, int *__quo) { 159 return ::remquof(__x, __y, __quo); 160 } 161 __DEVICE__ __CONSTEXPR__ float scalbln(float __x, long int __n) { 162 return ::scalblnf(__x, __n); 163 } 164 __DEVICE__ __CONSTEXPR__ bool signbit(float __x) { return ::__signbitf(__x); } 165 __DEVICE__ __CONSTEXPR__ bool signbit(double __x) { return ::__signbit(__x); } 166 167 // Notably missing above is nexttoward. We omit it because 168 // ocml doesn't provide an implementation, and we don't want to be in the 169 // business of implementing tricky libm functions in this header. 170 171 // Other functions. 172 __DEVICE__ __CONSTEXPR__ _Float16 fma(_Float16 __x, _Float16 __y, 173 _Float16 __z) { 174 return __ocml_fma_f16(__x, __y, __z); 175 } 176 __DEVICE__ __CONSTEXPR__ _Float16 pow(_Float16 __base, int __iexp) { 177 return __ocml_pown_f16(__base, __iexp); 178 } 179 180 #ifndef __OPENMP_AMDGCN__ 181 // BEGIN DEF_FUN and HIP_OVERLOAD 182 183 // BEGIN DEF_FUN 184 185 #pragma push_macro("__DEF_FUN1") 186 #pragma push_macro("__DEF_FUN2") 187 #pragma push_macro("__DEF_FUN2_FI") 188 189 // Define cmath functions with float argument and returns __retty. 190 #define __DEF_FUN1(__retty, __func) \ 191 __DEVICE__ __CONSTEXPR__ __retty __func(float __x) { return __func##f(__x); } 192 193 // Define cmath functions with two float arguments and returns __retty. 194 #define __DEF_FUN2(__retty, __func) \ 195 __DEVICE__ __CONSTEXPR__ __retty __func(float __x, float __y) { \ 196 return __func##f(__x, __y); \ 197 } 198 199 // Define cmath functions with a float and an int argument and returns __retty. 200 #define __DEF_FUN2_FI(__retty, __func) \ 201 __DEVICE__ __CONSTEXPR__ __retty __func(float __x, int __y) { \ 202 return __func##f(__x, __y); \ 203 } 204 205 __DEF_FUN1(float, acos) 206 __DEF_FUN1(float, acosh) 207 __DEF_FUN1(float, asin) 208 __DEF_FUN1(float, asinh) 209 __DEF_FUN1(float, atan) 210 __DEF_FUN2(float, atan2) 211 __DEF_FUN1(float, atanh) 212 __DEF_FUN1(float, cbrt) 213 __DEF_FUN1(float, ceil) 214 __DEF_FUN2(float, copysign) 215 __DEF_FUN1(float, cos) 216 __DEF_FUN1(float, cosh) 217 __DEF_FUN1(float, erf) 218 __DEF_FUN1(float, erfc) 219 __DEF_FUN1(float, exp) 220 __DEF_FUN1(float, exp2) 221 __DEF_FUN1(float, expm1) 222 __DEF_FUN1(float, fabs) 223 __DEF_FUN2(float, fdim) 224 __DEF_FUN1(float, floor) 225 __DEF_FUN2(float, fmax) 226 __DEF_FUN2(float, fmin) 227 __DEF_FUN2(float, fmod) 228 __DEF_FUN2(float, hypot) 229 __DEF_FUN1(int, ilogb) 230 __DEF_FUN2_FI(float, ldexp) 231 __DEF_FUN1(float, lgamma) 232 __DEF_FUN1(float, log) 233 __DEF_FUN1(float, log10) 234 __DEF_FUN1(float, log1p) 235 __DEF_FUN1(float, log2) 236 __DEF_FUN1(float, logb) 237 __DEF_FUN1(long long, llrint) 238 __DEF_FUN1(long long, llround) 239 __DEF_FUN1(long, lrint) 240 __DEF_FUN1(long, lround) 241 __DEF_FUN1(float, nearbyint) 242 __DEF_FUN2(float, nextafter) 243 __DEF_FUN2(float, pow) 244 __DEF_FUN2(float, remainder) 245 __DEF_FUN1(float, rint) 246 __DEF_FUN1(float, round) 247 __DEF_FUN2_FI(float, scalbn) 248 __DEF_FUN1(float, sin) 249 __DEF_FUN1(float, sinh) 250 __DEF_FUN1(float, sqrt) 251 __DEF_FUN1(float, tan) 252 __DEF_FUN1(float, tanh) 253 __DEF_FUN1(float, tgamma) 254 __DEF_FUN1(float, trunc) 255 256 #pragma pop_macro("__DEF_FUN1") 257 #pragma pop_macro("__DEF_FUN2") 258 #pragma pop_macro("__DEF_FUN2_FI") 259 260 // END DEF_FUN 261 262 // BEGIN HIP_OVERLOAD 263 264 #pragma push_macro("__HIP_OVERLOAD1") 265 #pragma push_macro("__HIP_OVERLOAD2") 266 267 // __hip_enable_if::type is a type function which returns __T if __B is true. 268 template <bool __B, class __T = void> struct __hip_enable_if {}; 269 270 template <class __T> struct __hip_enable_if<true, __T> { typedef __T type; }; 271 272 namespace __hip { 273 template <class _Tp> struct is_integral { 274 enum { value = 0 }; 275 }; 276 template <> struct is_integral<bool> { 277 enum { value = 1 }; 278 }; 279 template <> struct is_integral<char> { 280 enum { value = 1 }; 281 }; 282 template <> struct is_integral<signed char> { 283 enum { value = 1 }; 284 }; 285 template <> struct is_integral<unsigned char> { 286 enum { value = 1 }; 287 }; 288 template <> struct is_integral<wchar_t> { 289 enum { value = 1 }; 290 }; 291 template <> struct is_integral<short> { 292 enum { value = 1 }; 293 }; 294 template <> struct is_integral<unsigned short> { 295 enum { value = 1 }; 296 }; 297 template <> struct is_integral<int> { 298 enum { value = 1 }; 299 }; 300 template <> struct is_integral<unsigned int> { 301 enum { value = 1 }; 302 }; 303 template <> struct is_integral<long> { 304 enum { value = 1 }; 305 }; 306 template <> struct is_integral<unsigned long> { 307 enum { value = 1 }; 308 }; 309 template <> struct is_integral<long long> { 310 enum { value = 1 }; 311 }; 312 template <> struct is_integral<unsigned long long> { 313 enum { value = 1 }; 314 }; 315 316 // ToDo: specializes is_arithmetic<_Float16> 317 template <class _Tp> struct is_arithmetic { 318 enum { value = 0 }; 319 }; 320 template <> struct is_arithmetic<bool> { 321 enum { value = 1 }; 322 }; 323 template <> struct is_arithmetic<char> { 324 enum { value = 1 }; 325 }; 326 template <> struct is_arithmetic<signed char> { 327 enum { value = 1 }; 328 }; 329 template <> struct is_arithmetic<unsigned char> { 330 enum { value = 1 }; 331 }; 332 template <> struct is_arithmetic<wchar_t> { 333 enum { value = 1 }; 334 }; 335 template <> struct is_arithmetic<short> { 336 enum { value = 1 }; 337 }; 338 template <> struct is_arithmetic<unsigned short> { 339 enum { value = 1 }; 340 }; 341 template <> struct is_arithmetic<int> { 342 enum { value = 1 }; 343 }; 344 template <> struct is_arithmetic<unsigned int> { 345 enum { value = 1 }; 346 }; 347 template <> struct is_arithmetic<long> { 348 enum { value = 1 }; 349 }; 350 template <> struct is_arithmetic<unsigned long> { 351 enum { value = 1 }; 352 }; 353 template <> struct is_arithmetic<long long> { 354 enum { value = 1 }; 355 }; 356 template <> struct is_arithmetic<unsigned long long> { 357 enum { value = 1 }; 358 }; 359 template <> struct is_arithmetic<float> { 360 enum { value = 1 }; 361 }; 362 template <> struct is_arithmetic<double> { 363 enum { value = 1 }; 364 }; 365 366 struct true_type { 367 static const __constant__ bool value = true; 368 }; 369 struct false_type { 370 static const __constant__ bool value = false; 371 }; 372 373 template <typename __T, typename __U> struct is_same : public false_type {}; 374 template <typename __T> struct is_same<__T, __T> : public true_type {}; 375 376 template <typename __T> struct add_rvalue_reference { typedef __T &&type; }; 377 378 template <typename __T> typename add_rvalue_reference<__T>::type declval(); 379 380 // decltype is only available in C++11 and above. 381 #if __cplusplus >= 201103L 382 // __hip_promote 383 template <class _Tp> struct __numeric_type { 384 static void __test(...); 385 static _Float16 __test(_Float16); 386 static float __test(float); 387 static double __test(char); 388 static double __test(int); 389 static double __test(unsigned); 390 static double __test(long); 391 static double __test(unsigned long); 392 static double __test(long long); 393 static double __test(unsigned long long); 394 static double __test(double); 395 // No support for long double, use double instead. 396 static double __test(long double); 397 398 typedef decltype(__test(declval<_Tp>())) type; 399 static const bool value = !is_same<type, void>::value; 400 }; 401 402 template <> struct __numeric_type<void> { static const bool value = true; }; 403 404 template <class _A1, class _A2 = void, class _A3 = void, 405 bool = __numeric_type<_A1>::value &&__numeric_type<_A2>::value 406 &&__numeric_type<_A3>::value> 407 class __promote_imp { 408 public: 409 static const bool value = false; 410 }; 411 412 template <class _A1, class _A2, class _A3> 413 class __promote_imp<_A1, _A2, _A3, true> { 414 private: 415 typedef typename __promote_imp<_A1>::type __type1; 416 typedef typename __promote_imp<_A2>::type __type2; 417 typedef typename __promote_imp<_A3>::type __type3; 418 419 public: 420 typedef decltype(__type1() + __type2() + __type3()) type; 421 static const bool value = true; 422 }; 423 424 template <class _A1, class _A2> class __promote_imp<_A1, _A2, void, true> { 425 private: 426 typedef typename __promote_imp<_A1>::type __type1; 427 typedef typename __promote_imp<_A2>::type __type2; 428 429 public: 430 typedef decltype(__type1() + __type2()) type; 431 static const bool value = true; 432 }; 433 434 template <class _A1> class __promote_imp<_A1, void, void, true> { 435 public: 436 typedef typename __numeric_type<_A1>::type type; 437 static const bool value = true; 438 }; 439 440 template <class _A1, class _A2 = void, class _A3 = void> 441 class __promote : public __promote_imp<_A1, _A2, _A3> {}; 442 #endif //__cplusplus >= 201103L 443 } // namespace __hip 444 445 // __HIP_OVERLOAD1 is used to resolve function calls with integer argument to 446 // avoid compilation error due to ambibuity. e.g. floor(5) is resolved with 447 // floor(double). 448 #define __HIP_OVERLOAD1(__retty, __fn) \ 449 template <typename __T> \ 450 __DEVICE__ __CONSTEXPR__ \ 451 typename __hip_enable_if<__hip::is_integral<__T>::value, __retty>::type \ 452 __fn(__T __x) { \ 453 return ::__fn((double)__x); \ 454 } 455 456 // __HIP_OVERLOAD2 is used to resolve function calls with mixed float/double 457 // or integer argument to avoid compilation error due to ambibuity. e.g. 458 // max(5.0f, 6.0) is resolved with max(double, double). 459 #if __cplusplus >= 201103L 460 #define __HIP_OVERLOAD2(__retty, __fn) \ 461 template <typename __T1, typename __T2> \ 462 __DEVICE__ __CONSTEXPR__ typename __hip_enable_if< \ 463 __hip::is_arithmetic<__T1>::value && __hip::is_arithmetic<__T2>::value, \ 464 typename __hip::__promote<__T1, __T2>::type>::type \ 465 __fn(__T1 __x, __T2 __y) { \ 466 typedef typename __hip::__promote<__T1, __T2>::type __result_type; \ 467 return __fn((__result_type)__x, (__result_type)__y); \ 468 } 469 #else 470 #define __HIP_OVERLOAD2(__retty, __fn) \ 471 template <typename __T1, typename __T2> \ 472 __DEVICE__ __CONSTEXPR__ \ 473 typename __hip_enable_if<__hip::is_arithmetic<__T1>::value && \ 474 __hip::is_arithmetic<__T2>::value, \ 475 __retty>::type \ 476 __fn(__T1 __x, __T2 __y) { \ 477 return __fn((double)__x, (double)__y); \ 478 } 479 #endif 480 481 __HIP_OVERLOAD1(double, acos) 482 __HIP_OVERLOAD1(double, acosh) 483 __HIP_OVERLOAD1(double, asin) 484 __HIP_OVERLOAD1(double, asinh) 485 __HIP_OVERLOAD1(double, atan) 486 __HIP_OVERLOAD2(double, atan2) 487 __HIP_OVERLOAD1(double, atanh) 488 __HIP_OVERLOAD1(double, cbrt) 489 __HIP_OVERLOAD1(double, ceil) 490 __HIP_OVERLOAD2(double, copysign) 491 __HIP_OVERLOAD1(double, cos) 492 __HIP_OVERLOAD1(double, cosh) 493 __HIP_OVERLOAD1(double, erf) 494 __HIP_OVERLOAD1(double, erfc) 495 __HIP_OVERLOAD1(double, exp) 496 __HIP_OVERLOAD1(double, exp2) 497 __HIP_OVERLOAD1(double, expm1) 498 __HIP_OVERLOAD1(double, fabs) 499 __HIP_OVERLOAD2(double, fdim) 500 __HIP_OVERLOAD1(double, floor) 501 __HIP_OVERLOAD2(double, fmax) 502 __HIP_OVERLOAD2(double, fmin) 503 __HIP_OVERLOAD2(double, fmod) 504 #if !defined(__HIPCC_RTC__) 505 __HIP_OVERLOAD1(int, fpclassify) 506 #endif // !defined(__HIPCC_RTC__) 507 __HIP_OVERLOAD2(double, hypot) 508 __HIP_OVERLOAD1(int, ilogb) 509 __HIP_OVERLOAD1(bool, isfinite) 510 __HIP_OVERLOAD2(bool, isgreater) 511 __HIP_OVERLOAD2(bool, isgreaterequal) 512 __HIP_OVERLOAD1(bool, isinf) 513 __HIP_OVERLOAD2(bool, isless) 514 __HIP_OVERLOAD2(bool, islessequal) 515 __HIP_OVERLOAD2(bool, islessgreater) 516 __HIP_OVERLOAD1(bool, isnan) 517 __HIP_OVERLOAD1(bool, isnormal) 518 __HIP_OVERLOAD2(bool, isunordered) 519 __HIP_OVERLOAD1(double, lgamma) 520 __HIP_OVERLOAD1(double, log) 521 __HIP_OVERLOAD1(double, log10) 522 __HIP_OVERLOAD1(double, log1p) 523 __HIP_OVERLOAD1(double, log2) 524 __HIP_OVERLOAD1(double, logb) 525 __HIP_OVERLOAD1(long long, llrint) 526 __HIP_OVERLOAD1(long long, llround) 527 __HIP_OVERLOAD1(long, lrint) 528 __HIP_OVERLOAD1(long, lround) 529 __HIP_OVERLOAD1(double, nearbyint) 530 __HIP_OVERLOAD2(double, nextafter) 531 __HIP_OVERLOAD2(double, pow) 532 __HIP_OVERLOAD2(double, remainder) 533 __HIP_OVERLOAD1(double, rint) 534 __HIP_OVERLOAD1(double, round) 535 __HIP_OVERLOAD1(bool, signbit) 536 __HIP_OVERLOAD1(double, sin) 537 __HIP_OVERLOAD1(double, sinh) 538 __HIP_OVERLOAD1(double, sqrt) 539 __HIP_OVERLOAD1(double, tan) 540 __HIP_OVERLOAD1(double, tanh) 541 __HIP_OVERLOAD1(double, tgamma) 542 __HIP_OVERLOAD1(double, trunc) 543 544 // Overload these but don't add them to std, they are not part of cmath. 545 __HIP_OVERLOAD2(double, max) 546 __HIP_OVERLOAD2(double, min) 547 548 // Additional Overloads that don't quite match HIP_OVERLOAD. 549 #if __cplusplus >= 201103L 550 template <typename __T1, typename __T2, typename __T3> 551 __DEVICE__ __CONSTEXPR__ typename __hip_enable_if< 552 __hip::is_arithmetic<__T1>::value && __hip::is_arithmetic<__T2>::value && 553 __hip::is_arithmetic<__T3>::value, 554 typename __hip::__promote<__T1, __T2, __T3>::type>::type 555 fma(__T1 __x, __T2 __y, __T3 __z) { 556 typedef typename __hip::__promote<__T1, __T2, __T3>::type __result_type; 557 return ::fma((__result_type)__x, (__result_type)__y, (__result_type)__z); 558 } 559 #else 560 template <typename __T1, typename __T2, typename __T3> 561 __DEVICE__ __CONSTEXPR__ 562 typename __hip_enable_if<__hip::is_arithmetic<__T1>::value && 563 __hip::is_arithmetic<__T2>::value && 564 __hip::is_arithmetic<__T3>::value, 565 double>::type 566 fma(__T1 __x, __T2 __y, __T3 __z) { 567 return ::fma((double)__x, (double)__y, (double)__z); 568 } 569 #endif 570 571 template <typename __T> 572 __DEVICE__ __CONSTEXPR__ 573 typename __hip_enable_if<__hip::is_integral<__T>::value, double>::type 574 frexp(__T __x, int *__exp) { 575 return ::frexp((double)__x, __exp); 576 } 577 578 template <typename __T> 579 __DEVICE__ __CONSTEXPR__ 580 typename __hip_enable_if<__hip::is_integral<__T>::value, double>::type 581 ldexp(__T __x, int __exp) { 582 return ::ldexp((double)__x, __exp); 583 } 584 585 template <typename __T> 586 __DEVICE__ __CONSTEXPR__ 587 typename __hip_enable_if<__hip::is_integral<__T>::value, double>::type 588 modf(__T __x, double *__exp) { 589 return ::modf((double)__x, __exp); 590 } 591 592 #if __cplusplus >= 201103L 593 template <typename __T1, typename __T2> 594 __DEVICE__ __CONSTEXPR__ 595 typename __hip_enable_if<__hip::is_arithmetic<__T1>::value && 596 __hip::is_arithmetic<__T2>::value, 597 typename __hip::__promote<__T1, __T2>::type>::type 598 remquo(__T1 __x, __T2 __y, int *__quo) { 599 typedef typename __hip::__promote<__T1, __T2>::type __result_type; 600 return ::remquo((__result_type)__x, (__result_type)__y, __quo); 601 } 602 #else 603 template <typename __T1, typename __T2> 604 __DEVICE__ __CONSTEXPR__ 605 typename __hip_enable_if<__hip::is_arithmetic<__T1>::value && 606 __hip::is_arithmetic<__T2>::value, 607 double>::type 608 remquo(__T1 __x, __T2 __y, int *__quo) { 609 return ::remquo((double)__x, (double)__y, __quo); 610 } 611 #endif 612 613 template <typename __T> 614 __DEVICE__ __CONSTEXPR__ 615 typename __hip_enable_if<__hip::is_integral<__T>::value, double>::type 616 scalbln(__T __x, long int __exp) { 617 return ::scalbln((double)__x, __exp); 618 } 619 620 template <typename __T> 621 __DEVICE__ __CONSTEXPR__ 622 typename __hip_enable_if<__hip::is_integral<__T>::value, double>::type 623 scalbn(__T __x, int __exp) { 624 return ::scalbn((double)__x, __exp); 625 } 626 627 #pragma pop_macro("__HIP_OVERLOAD1") 628 #pragma pop_macro("__HIP_OVERLOAD2") 629 630 // END HIP_OVERLOAD 631 632 // END DEF_FUN and HIP_OVERLOAD 633 634 #endif // ifndef __OPENMP_AMDGCN__ 635 #endif // defined(__cplusplus) 636 637 #ifndef __OPENMP_AMDGCN__ 638 // Define these overloads inside the namespace our standard library uses. 639 #if !defined(__HIPCC_RTC__) 640 #ifdef _LIBCPP_BEGIN_NAMESPACE_STD 641 _LIBCPP_BEGIN_NAMESPACE_STD 642 #else 643 namespace std { 644 #ifdef _GLIBCXX_BEGIN_NAMESPACE_VERSION 645 _GLIBCXX_BEGIN_NAMESPACE_VERSION 646 #endif // _GLIBCXX_BEGIN_NAMESPACE_VERSION 647 #endif // _LIBCPP_BEGIN_NAMESPACE_STD 648 649 // Pull the new overloads we defined above into namespace std. 650 // using ::abs; - This may be considered for C++. 651 using ::acos; 652 using ::acosh; 653 using ::asin; 654 using ::asinh; 655 using ::atan; 656 using ::atan2; 657 using ::atanh; 658 using ::cbrt; 659 using ::ceil; 660 using ::copysign; 661 using ::cos; 662 using ::cosh; 663 using ::erf; 664 using ::erfc; 665 using ::exp; 666 using ::exp2; 667 using ::expm1; 668 using ::fabs; 669 using ::fdim; 670 using ::floor; 671 using ::fma; 672 using ::fmax; 673 using ::fmin; 674 using ::fmod; 675 using ::fpclassify; 676 using ::frexp; 677 using ::hypot; 678 using ::ilogb; 679 using ::isfinite; 680 using ::isgreater; 681 using ::isgreaterequal; 682 using ::isless; 683 using ::islessequal; 684 using ::islessgreater; 685 using ::isnormal; 686 using ::isunordered; 687 using ::ldexp; 688 using ::lgamma; 689 using ::llrint; 690 using ::llround; 691 using ::log; 692 using ::log10; 693 using ::log1p; 694 using ::log2; 695 using ::logb; 696 using ::lrint; 697 using ::lround; 698 using ::modf; 699 // using ::nan; - This may be considered for C++. 700 // using ::nanf; - This may be considered for C++. 701 // using ::nanl; - This is not yet defined. 702 using ::nearbyint; 703 using ::nextafter; 704 // using ::nexttoward; - Omit this since we do not have a definition. 705 using ::pow; 706 using ::remainder; 707 using ::remquo; 708 using ::rint; 709 using ::round; 710 using ::scalbln; 711 using ::scalbn; 712 using ::signbit; 713 using ::sin; 714 using ::sinh; 715 using ::sqrt; 716 using ::tan; 717 using ::tanh; 718 using ::tgamma; 719 using ::trunc; 720 721 // Well this is fun: We need to pull these symbols in for libc++, but we can't 722 // pull them in with libstdc++, because its ::isinf and ::isnan are different 723 // than its std::isinf and std::isnan. 724 #ifndef __GLIBCXX__ 725 using ::isinf; 726 using ::isnan; 727 #endif 728 729 // Finally, pull the "foobarf" functions that HIP defines into std. 730 using ::acosf; 731 using ::acoshf; 732 using ::asinf; 733 using ::asinhf; 734 using ::atan2f; 735 using ::atanf; 736 using ::atanhf; 737 using ::cbrtf; 738 using ::ceilf; 739 using ::copysignf; 740 using ::cosf; 741 using ::coshf; 742 using ::erfcf; 743 using ::erff; 744 using ::exp2f; 745 using ::expf; 746 using ::expm1f; 747 using ::fabsf; 748 using ::fdimf; 749 using ::floorf; 750 using ::fmaf; 751 using ::fmaxf; 752 using ::fminf; 753 using ::fmodf; 754 using ::frexpf; 755 using ::hypotf; 756 using ::ilogbf; 757 using ::ldexpf; 758 using ::lgammaf; 759 using ::llrintf; 760 using ::llroundf; 761 using ::log10f; 762 using ::log1pf; 763 using ::log2f; 764 using ::logbf; 765 using ::logf; 766 using ::lrintf; 767 using ::lroundf; 768 using ::modff; 769 using ::nearbyintf; 770 using ::nextafterf; 771 // using ::nexttowardf; - Omit this since we do not have a definition. 772 using ::powf; 773 using ::remainderf; 774 using ::remquof; 775 using ::rintf; 776 using ::roundf; 777 using ::scalblnf; 778 using ::scalbnf; 779 using ::sinf; 780 using ::sinhf; 781 using ::sqrtf; 782 using ::tanf; 783 using ::tanhf; 784 using ::tgammaf; 785 using ::truncf; 786 787 #ifdef _LIBCPP_END_NAMESPACE_STD 788 _LIBCPP_END_NAMESPACE_STD 789 #else 790 #ifdef _GLIBCXX_BEGIN_NAMESPACE_VERSION 791 _GLIBCXX_END_NAMESPACE_VERSION 792 #endif // _GLIBCXX_BEGIN_NAMESPACE_VERSION 793 } // namespace std 794 #endif // _LIBCPP_END_NAMESPACE_STD 795 #endif // !defined(__HIPCC_RTC__) 796 797 // Define device-side math functions from <ymath.h> on MSVC. 798 #if !defined(__HIPCC_RTC__) 799 #if defined(_MSC_VER) 800 801 // Before VS2019, `<ymath.h>` is also included in `<limits>` and other headers. 802 // But, from VS2019, it's only included in `<complex>`. Need to include 803 // `<ymath.h>` here to ensure C functions declared there won't be markded as 804 // `__host__` and `__device__` through `<complex>` wrapper. 805 #include <ymath.h> 806 807 #if defined(__cplusplus) 808 extern "C" { 809 #endif // defined(__cplusplus) 810 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) double _Cosh(double x, 811 double y) { 812 return cosh(x) * y; 813 } 814 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) float _FCosh(float x, 815 float y) { 816 return coshf(x) * y; 817 } 818 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) short _Dtest(double *p) { 819 return fpclassify(*p); 820 } 821 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) short _FDtest(float *p) { 822 return fpclassify(*p); 823 } 824 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) double _Sinh(double x, 825 double y) { 826 return sinh(x) * y; 827 } 828 __DEVICE__ __CONSTEXPR__ __attribute__((overloadable)) float _FSinh(float x, 829 float y) { 830 return sinhf(x) * y; 831 } 832 #if defined(__cplusplus) 833 } 834 #endif // defined(__cplusplus) 835 #endif // defined(_MSC_VER) 836 #endif // !defined(__HIPCC_RTC__) 837 #endif // ifndef __OPENMP_AMDGCN__ 838 839 #pragma pop_macro("__DEVICE__") 840 #pragma pop_macro("__CONSTEXPR__") 841 842 #endif // __CLANG_HIP_CMATH_H__ 843