1 /*===---- __clang_cuda_math.h - Device-side CUDA math support --------------=== 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 #ifndef __CLANG_CUDA_MATH_H__ 10 #define __CLANG_CUDA_MATH_H__ 11 #ifndef __CUDA__ 12 #error "This file is for CUDA compilation only." 13 #endif 14 15 #ifndef __OPENMP_NVPTX__ 16 #if CUDA_VERSION < 9000 17 #error This file is intended to be used with CUDA-9+ only. 18 #endif 19 #endif 20 21 // __DEVICE__ is a helper macro with common set of attributes for the wrappers 22 // we implement in this file. We need static in order to avoid emitting unused 23 // functions and __forceinline__ helps inlining these wrappers at -O1. 24 #pragma push_macro("__DEVICE__") 25 #ifdef __OPENMP_NVPTX__ 26 #if defined(__cplusplus) 27 #define __DEVICE__ static constexpr __attribute__((always_inline, nothrow)) 28 #else 29 #define __DEVICE__ static __attribute__((always_inline, nothrow)) 30 #endif 31 #else 32 #define __DEVICE__ static __device__ __forceinline__ 33 #endif 34 35 // Specialized version of __DEVICE__ for functions with void return type. Needed 36 // because the OpenMP overlay requires constexpr functions here but prior to 37 // c++14 void return functions could not be constexpr. 38 #pragma push_macro("__DEVICE_VOID__") 39 #if defined(__OPENMP_NVPTX__) && defined(__cplusplus) && __cplusplus < 201402L 40 #define __DEVICE_VOID__ static __attribute__((always_inline, nothrow)) 41 #else 42 #define __DEVICE_VOID__ __DEVICE__ 43 #endif 44 45 // libdevice provides fast low precision and slow full-recision implementations 46 // for some functions. Which one gets selected depends on 47 // __CLANG_CUDA_APPROX_TRANSCENDENTALS__ which gets defined by clang if 48 // -ffast-math or -fcuda-approx-transcendentals are in effect. 49 #pragma push_macro("__FAST_OR_SLOW") 50 #if defined(__CLANG_CUDA_APPROX_TRANSCENDENTALS__) 51 #define __FAST_OR_SLOW(fast, slow) fast 52 #else 53 #define __FAST_OR_SLOW(fast, slow) slow 54 #endif 55 56 __DEVICE__ int abs(int __a) { return __nv_abs(__a); } 57 __DEVICE__ double fabs(double __a) { return __nv_fabs(__a); } 58 __DEVICE__ double acos(double __a) { return __nv_acos(__a); } 59 __DEVICE__ float acosf(float __a) { return __nv_acosf(__a); } 60 __DEVICE__ double acosh(double __a) { return __nv_acosh(__a); } 61 __DEVICE__ float acoshf(float __a) { return __nv_acoshf(__a); } 62 __DEVICE__ double asin(double __a) { return __nv_asin(__a); } 63 __DEVICE__ float asinf(float __a) { return __nv_asinf(__a); } 64 __DEVICE__ double asinh(double __a) { return __nv_asinh(__a); } 65 __DEVICE__ float asinhf(float __a) { return __nv_asinhf(__a); } 66 __DEVICE__ double atan(double __a) { return __nv_atan(__a); } 67 __DEVICE__ double atan2(double __a, double __b) { return __nv_atan2(__a, __b); } 68 __DEVICE__ float atan2f(float __a, float __b) { return __nv_atan2f(__a, __b); } 69 __DEVICE__ float atanf(float __a) { return __nv_atanf(__a); } 70 __DEVICE__ double atanh(double __a) { return __nv_atanh(__a); } 71 __DEVICE__ float atanhf(float __a) { return __nv_atanhf(__a); } 72 __DEVICE__ double cbrt(double __a) { return __nv_cbrt(__a); } 73 __DEVICE__ float cbrtf(float __a) { return __nv_cbrtf(__a); } 74 __DEVICE__ double ceil(double __a) { return __nv_ceil(__a); } 75 __DEVICE__ float ceilf(float __a) { return __nv_ceilf(__a); } 76 __DEVICE__ double copysign(double __a, double __b) { 77 return __nv_copysign(__a, __b); 78 } 79 __DEVICE__ float copysignf(float __a, float __b) { 80 return __nv_copysignf(__a, __b); 81 } 82 __DEVICE__ double cos(double __a) { return __nv_cos(__a); } 83 __DEVICE__ float cosf(float __a) { 84 return __FAST_OR_SLOW(__nv_fast_cosf, __nv_cosf)(__a); 85 } 86 __DEVICE__ double cosh(double __a) { return __nv_cosh(__a); } 87 __DEVICE__ float coshf(float __a) { return __nv_coshf(__a); } 88 __DEVICE__ double cospi(double __a) { return __nv_cospi(__a); } 89 __DEVICE__ float cospif(float __a) { return __nv_cospif(__a); } 90 __DEVICE__ double cyl_bessel_i0(double __a) { return __nv_cyl_bessel_i0(__a); } 91 __DEVICE__ float cyl_bessel_i0f(float __a) { return __nv_cyl_bessel_i0f(__a); } 92 __DEVICE__ double cyl_bessel_i1(double __a) { return __nv_cyl_bessel_i1(__a); } 93 __DEVICE__ float cyl_bessel_i1f(float __a) { return __nv_cyl_bessel_i1f(__a); } 94 __DEVICE__ double erf(double __a) { return __nv_erf(__a); } 95 __DEVICE__ double erfc(double __a) { return __nv_erfc(__a); } 96 __DEVICE__ float erfcf(float __a) { return __nv_erfcf(__a); } 97 __DEVICE__ double erfcinv(double __a) { return __nv_erfcinv(__a); } 98 __DEVICE__ float erfcinvf(float __a) { return __nv_erfcinvf(__a); } 99 __DEVICE__ double erfcx(double __a) { return __nv_erfcx(__a); } 100 __DEVICE__ float erfcxf(float __a) { return __nv_erfcxf(__a); } 101 __DEVICE__ float erff(float __a) { return __nv_erff(__a); } 102 __DEVICE__ double erfinv(double __a) { return __nv_erfinv(__a); } 103 __DEVICE__ float erfinvf(float __a) { return __nv_erfinvf(__a); } 104 __DEVICE__ double exp(double __a) { return __nv_exp(__a); } 105 __DEVICE__ double exp10(double __a) { return __nv_exp10(__a); } 106 __DEVICE__ float exp10f(float __a) { return __nv_exp10f(__a); } 107 __DEVICE__ double exp2(double __a) { return __nv_exp2(__a); } 108 __DEVICE__ float exp2f(float __a) { return __nv_exp2f(__a); } 109 __DEVICE__ float expf(float __a) { return __nv_expf(__a); } 110 __DEVICE__ double expm1(double __a) { return __nv_expm1(__a); } 111 __DEVICE__ float expm1f(float __a) { return __nv_expm1f(__a); } 112 __DEVICE__ float fabsf(float __a) { return __nv_fabsf(__a); } 113 __DEVICE__ double fdim(double __a, double __b) { return __nv_fdim(__a, __b); } 114 __DEVICE__ float fdimf(float __a, float __b) { return __nv_fdimf(__a, __b); } 115 __DEVICE__ double fdivide(double __a, double __b) { return __a / __b; } 116 __DEVICE__ float fdividef(float __a, float __b) { 117 #if __FAST_MATH__ && !__CUDA_PREC_DIV 118 return __nv_fast_fdividef(__a, __b); 119 #else 120 return __a / __b; 121 #endif 122 } 123 __DEVICE__ double floor(double __f) { return __nv_floor(__f); } 124 __DEVICE__ float floorf(float __f) { return __nv_floorf(__f); } 125 __DEVICE__ double fma(double __a, double __b, double __c) { 126 return __nv_fma(__a, __b, __c); 127 } 128 __DEVICE__ float fmaf(float __a, float __b, float __c) { 129 return __nv_fmaf(__a, __b, __c); 130 } 131 __DEVICE__ double fmax(double __a, double __b) { return __nv_fmax(__a, __b); } 132 __DEVICE__ float fmaxf(float __a, float __b) { return __nv_fmaxf(__a, __b); } 133 __DEVICE__ double fmin(double __a, double __b) { return __nv_fmin(__a, __b); } 134 __DEVICE__ float fminf(float __a, float __b) { return __nv_fminf(__a, __b); } 135 __DEVICE__ double fmod(double __a, double __b) { return __nv_fmod(__a, __b); } 136 __DEVICE__ float fmodf(float __a, float __b) { return __nv_fmodf(__a, __b); } 137 __DEVICE__ double frexp(double __a, int *__b) { return __nv_frexp(__a, __b); } 138 __DEVICE__ float frexpf(float __a, int *__b) { return __nv_frexpf(__a, __b); } 139 __DEVICE__ double hypot(double __a, double __b) { return __nv_hypot(__a, __b); } 140 __DEVICE__ float hypotf(float __a, float __b) { return __nv_hypotf(__a, __b); } 141 __DEVICE__ int ilogb(double __a) { return __nv_ilogb(__a); } 142 __DEVICE__ int ilogbf(float __a) { return __nv_ilogbf(__a); } 143 __DEVICE__ double j0(double __a) { return __nv_j0(__a); } 144 __DEVICE__ float j0f(float __a) { return __nv_j0f(__a); } 145 __DEVICE__ double j1(double __a) { return __nv_j1(__a); } 146 __DEVICE__ float j1f(float __a) { return __nv_j1f(__a); } 147 __DEVICE__ double jn(int __n, double __a) { return __nv_jn(__n, __a); } 148 __DEVICE__ float jnf(int __n, float __a) { return __nv_jnf(__n, __a); } 149 #if defined(__LP64__) || defined(_WIN64) 150 __DEVICE__ long labs(long __a) { return __nv_llabs(__a); }; 151 #else 152 __DEVICE__ long labs(long __a) { return __nv_abs(__a); }; 153 #endif 154 __DEVICE__ double ldexp(double __a, int __b) { return __nv_ldexp(__a, __b); } 155 __DEVICE__ float ldexpf(float __a, int __b) { return __nv_ldexpf(__a, __b); } 156 __DEVICE__ double lgamma(double __a) { return __nv_lgamma(__a); } 157 __DEVICE__ float lgammaf(float __a) { return __nv_lgammaf(__a); } 158 __DEVICE__ long long llabs(long long __a) { return __nv_llabs(__a); } 159 __DEVICE__ long long llmax(long long __a, long long __b) { 160 return __nv_llmax(__a, __b); 161 } 162 __DEVICE__ long long llmin(long long __a, long long __b) { 163 return __nv_llmin(__a, __b); 164 } 165 __DEVICE__ long long llrint(double __a) { return __nv_llrint(__a); } 166 __DEVICE__ long long llrintf(float __a) { return __nv_llrintf(__a); } 167 __DEVICE__ long long llround(double __a) { return __nv_llround(__a); } 168 __DEVICE__ long long llroundf(float __a) { return __nv_llroundf(__a); } 169 __DEVICE__ double round(double __a) { return __nv_round(__a); } 170 __DEVICE__ float roundf(float __a) { return __nv_roundf(__a); } 171 __DEVICE__ double log(double __a) { return __nv_log(__a); } 172 __DEVICE__ double log10(double __a) { return __nv_log10(__a); } 173 __DEVICE__ float log10f(float __a) { return __nv_log10f(__a); } 174 __DEVICE__ double log1p(double __a) { return __nv_log1p(__a); } 175 __DEVICE__ float log1pf(float __a) { return __nv_log1pf(__a); } 176 __DEVICE__ double log2(double __a) { return __nv_log2(__a); } 177 __DEVICE__ float log2f(float __a) { 178 return __FAST_OR_SLOW(__nv_fast_log2f, __nv_log2f)(__a); 179 } 180 __DEVICE__ double logb(double __a) { return __nv_logb(__a); } 181 __DEVICE__ float logbf(float __a) { return __nv_logbf(__a); } 182 __DEVICE__ float logf(float __a) { 183 return __FAST_OR_SLOW(__nv_fast_logf, __nv_logf)(__a); 184 } 185 #if defined(__LP64__) || defined(_WIN64) 186 __DEVICE__ long lrint(double __a) { return llrint(__a); } 187 __DEVICE__ long lrintf(float __a) { return __float2ll_rn(__a); } 188 __DEVICE__ long lround(double __a) { return llround(__a); } 189 __DEVICE__ long lroundf(float __a) { return llroundf(__a); } 190 #else 191 __DEVICE__ long lrint(double __a) { return (long)rint(__a); } 192 __DEVICE__ long lrintf(float __a) { return __float2int_rn(__a); } 193 __DEVICE__ long lround(double __a) { return round(__a); } 194 __DEVICE__ long lroundf(float __a) { return roundf(__a); } 195 #endif 196 __DEVICE__ int max(int __a, int __b) { return __nv_max(__a, __b); } 197 __DEVICE__ int min(int __a, int __b) { return __nv_min(__a, __b); } 198 __DEVICE__ double modf(double __a, double *__b) { return __nv_modf(__a, __b); } 199 __DEVICE__ float modff(float __a, float *__b) { return __nv_modff(__a, __b); } 200 __DEVICE__ double nearbyint(double __a) { return __builtin_nearbyint(__a); } 201 __DEVICE__ float nearbyintf(float __a) { return __builtin_nearbyintf(__a); } 202 __DEVICE__ double nextafter(double __a, double __b) { 203 return __nv_nextafter(__a, __b); 204 } 205 __DEVICE__ float nextafterf(float __a, float __b) { 206 return __nv_nextafterf(__a, __b); 207 } 208 __DEVICE__ double norm(int __dim, const double *__t) { 209 return __nv_norm(__dim, __t); 210 } 211 __DEVICE__ double norm3d(double __a, double __b, double __c) { 212 return __nv_norm3d(__a, __b, __c); 213 } 214 __DEVICE__ float norm3df(float __a, float __b, float __c) { 215 return __nv_norm3df(__a, __b, __c); 216 } 217 __DEVICE__ double norm4d(double __a, double __b, double __c, double __d) { 218 return __nv_norm4d(__a, __b, __c, __d); 219 } 220 __DEVICE__ float norm4df(float __a, float __b, float __c, float __d) { 221 return __nv_norm4df(__a, __b, __c, __d); 222 } 223 __DEVICE__ double normcdf(double __a) { return __nv_normcdf(__a); } 224 __DEVICE__ float normcdff(float __a) { return __nv_normcdff(__a); } 225 __DEVICE__ double normcdfinv(double __a) { return __nv_normcdfinv(__a); } 226 __DEVICE__ float normcdfinvf(float __a) { return __nv_normcdfinvf(__a); } 227 __DEVICE__ float normf(int __dim, const float *__t) { 228 return __nv_normf(__dim, __t); 229 } 230 __DEVICE__ double pow(double __a, double __b) { return __nv_pow(__a, __b); } 231 __DEVICE__ float powf(float __a, float __b) { return __nv_powf(__a, __b); } 232 __DEVICE__ double powi(double __a, int __b) { return __nv_powi(__a, __b); } 233 __DEVICE__ float powif(float __a, int __b) { return __nv_powif(__a, __b); } 234 __DEVICE__ double rcbrt(double __a) { return __nv_rcbrt(__a); } 235 __DEVICE__ float rcbrtf(float __a) { return __nv_rcbrtf(__a); } 236 __DEVICE__ double remainder(double __a, double __b) { 237 return __nv_remainder(__a, __b); 238 } 239 __DEVICE__ float remainderf(float __a, float __b) { 240 return __nv_remainderf(__a, __b); 241 } 242 __DEVICE__ double remquo(double __a, double __b, int *__c) { 243 return __nv_remquo(__a, __b, __c); 244 } 245 __DEVICE__ float remquof(float __a, float __b, int *__c) { 246 return __nv_remquof(__a, __b, __c); 247 } 248 __DEVICE__ double rhypot(double __a, double __b) { 249 return __nv_rhypot(__a, __b); 250 } 251 __DEVICE__ float rhypotf(float __a, float __b) { 252 return __nv_rhypotf(__a, __b); 253 } 254 // __nv_rint* in libdevice is buggy and produces incorrect results. 255 __DEVICE__ double rint(double __a) { return __builtin_rint(__a); } 256 __DEVICE__ float rintf(float __a) { return __builtin_rintf(__a); } 257 __DEVICE__ double rnorm(int __a, const double *__b) { 258 return __nv_rnorm(__a, __b); 259 } 260 __DEVICE__ double rnorm3d(double __a, double __b, double __c) { 261 return __nv_rnorm3d(__a, __b, __c); 262 } 263 __DEVICE__ float rnorm3df(float __a, float __b, float __c) { 264 return __nv_rnorm3df(__a, __b, __c); 265 } 266 __DEVICE__ double rnorm4d(double __a, double __b, double __c, double __d) { 267 return __nv_rnorm4d(__a, __b, __c, __d); 268 } 269 __DEVICE__ float rnorm4df(float __a, float __b, float __c, float __d) { 270 return __nv_rnorm4df(__a, __b, __c, __d); 271 } 272 __DEVICE__ float rnormf(int __dim, const float *__t) { 273 return __nv_rnormf(__dim, __t); 274 } 275 __DEVICE__ double rsqrt(double __a) { return __nv_rsqrt(__a); } 276 __DEVICE__ float rsqrtf(float __a) { return __nv_rsqrtf(__a); } 277 __DEVICE__ double scalbn(double __a, int __b) { return __nv_scalbn(__a, __b); } 278 __DEVICE__ float scalbnf(float __a, int __b) { return __nv_scalbnf(__a, __b); } 279 __DEVICE__ double scalbln(double __a, long __b) { 280 if (__b > INT_MAX) 281 return __a > 0 ? HUGE_VAL : -HUGE_VAL; 282 if (__b < INT_MIN) 283 return __a > 0 ? 0.0 : -0.0; 284 return scalbn(__a, (int)__b); 285 } 286 __DEVICE__ float scalblnf(float __a, long __b) { 287 if (__b > INT_MAX) 288 return __a > 0 ? HUGE_VALF : -HUGE_VALF; 289 if (__b < INT_MIN) 290 return __a > 0 ? 0.f : -0.f; 291 return scalbnf(__a, (int)__b); 292 } 293 __DEVICE__ double sin(double __a) { return __nv_sin(__a); } 294 __DEVICE_VOID__ void sincos(double __a, double *__s, double *__c) { 295 return __nv_sincos(__a, __s, __c); 296 } 297 __DEVICE_VOID__ void sincosf(float __a, float *__s, float *__c) { 298 return __FAST_OR_SLOW(__nv_fast_sincosf, __nv_sincosf)(__a, __s, __c); 299 } 300 __DEVICE_VOID__ void sincospi(double __a, double *__s, double *__c) { 301 return __nv_sincospi(__a, __s, __c); 302 } 303 __DEVICE_VOID__ void sincospif(float __a, float *__s, float *__c) { 304 return __nv_sincospif(__a, __s, __c); 305 } 306 __DEVICE__ float sinf(float __a) { 307 return __FAST_OR_SLOW(__nv_fast_sinf, __nv_sinf)(__a); 308 } 309 __DEVICE__ double sinh(double __a) { return __nv_sinh(__a); } 310 __DEVICE__ float sinhf(float __a) { return __nv_sinhf(__a); } 311 __DEVICE__ double sinpi(double __a) { return __nv_sinpi(__a); } 312 __DEVICE__ float sinpif(float __a) { return __nv_sinpif(__a); } 313 __DEVICE__ double sqrt(double __a) { return __nv_sqrt(__a); } 314 __DEVICE__ float sqrtf(float __a) { return __nv_sqrtf(__a); } 315 __DEVICE__ double tan(double __a) { return __nv_tan(__a); } 316 __DEVICE__ float tanf(float __a) { return __nv_tanf(__a); } 317 __DEVICE__ double tanh(double __a) { return __nv_tanh(__a); } 318 __DEVICE__ float tanhf(float __a) { return __nv_tanhf(__a); } 319 __DEVICE__ double tgamma(double __a) { return __nv_tgamma(__a); } 320 __DEVICE__ float tgammaf(float __a) { return __nv_tgammaf(__a); } 321 __DEVICE__ double trunc(double __a) { return __nv_trunc(__a); } 322 __DEVICE__ float truncf(float __a) { return __nv_truncf(__a); } 323 __DEVICE__ unsigned long long ullmax(unsigned long long __a, 324 unsigned long long __b) { 325 return __nv_ullmax(__a, __b); 326 } 327 __DEVICE__ unsigned long long ullmin(unsigned long long __a, 328 unsigned long long __b) { 329 return __nv_ullmin(__a, __b); 330 } 331 __DEVICE__ unsigned int umax(unsigned int __a, unsigned int __b) { 332 return __nv_umax(__a, __b); 333 } 334 __DEVICE__ unsigned int umin(unsigned int __a, unsigned int __b) { 335 return __nv_umin(__a, __b); 336 } 337 __DEVICE__ double y0(double __a) { return __nv_y0(__a); } 338 __DEVICE__ float y0f(float __a) { return __nv_y0f(__a); } 339 __DEVICE__ double y1(double __a) { return __nv_y1(__a); } 340 __DEVICE__ float y1f(float __a) { return __nv_y1f(__a); } 341 __DEVICE__ double yn(int __a, double __b) { return __nv_yn(__a, __b); } 342 __DEVICE__ float ynf(int __a, float __b) { return __nv_ynf(__a, __b); } 343 344 #pragma pop_macro("__DEVICE__") 345 #pragma pop_macro("__DEVICE_VOID__") 346 #pragma pop_macro("__FAST_OR_SLOW") 347 348 #endif // __CLANG_CUDA_MATH_H__ 349