| 1 |
//===----------------------------------------------------------------------===// |
| 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 |
/// \file |
| 10 |
/// This file contains the implementation of the device kernels that wrap the |
| 11 |
/// math functions from the cuda-math provider. |
| 12 |
/// |
| 13 |
//===----------------------------------------------------------------------===// |
| 14 |
|
| 15 |
#ifdef CUDA_MATH_FOUND |
| 16 |
|
| 17 |
#include "Conformance/device_code/DeviceAPIs.hpp" |
| 18 |
#include "Conformance/device_code/KernelRunner.hpp" |
| 19 |
|
| 20 |
#include <gpuintrin.h> |
| 21 |
#include <stddef.h> |
| 22 |
|
| 23 |
using namespace kernels; |
| 24 |
|
| 25 |
//===----------------------------------------------------------------------===// |
| 26 |
// Helpers |
| 27 |
//===----------------------------------------------------------------------===// |
| 28 |
|
| 29 |
static inline float powfRoundedExponent(float Base, float Exponent) { |
| 30 |
return __nv_powf(Base, __nv_roundf(Exponent)); |
| 31 |
} |
| 32 |
|
| 33 |
static inline double sincosSin(double X) { |
| 34 |
double SinX, CosX; |
| 35 |
__nv_sincos(X, &SinX, &CosX); |
| 36 |
return SinX; |
| 37 |
} |
| 38 |
|
| 39 |
static inline double sincosCos(double X) { |
| 40 |
double SinX, CosX; |
| 41 |
__nv_sincos(X, &SinX, &CosX); |
| 42 |
return CosX; |
| 43 |
} |
| 44 |
|
| 45 |
static inline float sincosfSin(float X) { |
| 46 |
float SinX, CosX; |
| 47 |
__nv_sincosf(X, &SinX, &CosX); |
| 48 |
return SinX; |
| 49 |
} |
| 50 |
|
| 51 |
static inline float sincosfCos(float X) { |
| 52 |
float SinX, CosX; |
| 53 |
__nv_sincosf(X, &SinX, &CosX); |
| 54 |
return CosX; |
| 55 |
} |
| 56 |
|
| 57 |
//===----------------------------------------------------------------------===// |
| 58 |
// Kernels |
| 59 |
//===----------------------------------------------------------------------===// |
| 60 |
|
| 61 |
extern "C" { |
| 62 |
|
| 63 |
__gpu_kernel void acosKernel(const double *X, double *Out, |
| 64 |
size_t NumElements) noexcept { |
| 65 |
runKernelBody<__nv_acos>(NumElements, Out, X); |
| 66 |
} |
| 67 |
|
| 68 |
__gpu_kernel void acosfKernel(const float *X, float *Out, |
| 69 |
size_t NumElements) noexcept { |
| 70 |
runKernelBody<__nv_acosf>(NumElements, Out, X); |
| 71 |
} |
| 72 |
|
| 73 |
__gpu_kernel void acoshfKernel(const float *X, float *Out, |
| 74 |
size_t NumElements) noexcept { |
| 75 |
runKernelBody<__nv_acoshf>(NumElements, Out, X); |
| 76 |
} |
| 77 |
|
| 78 |
__gpu_kernel void asinKernel(const double *X, double *Out, |
| 79 |
size_t NumElements) noexcept { |
| 80 |
runKernelBody<__nv_asin>(NumElements, Out, X); |
| 81 |
} |
| 82 |
|
| 83 |
__gpu_kernel void asinfKernel(const float *X, float *Out, |
| 84 |
size_t NumElements) noexcept { |
| 85 |
runKernelBody<__nv_asinf>(NumElements, Out, X); |
| 86 |
} |
| 87 |
|
| 88 |
__gpu_kernel void asinhfKernel(const float *X, float *Out, |
| 89 |
size_t NumElements) noexcept { |
| 90 |
runKernelBody<__nv_asinhf>(NumElements, Out, X); |
| 91 |
} |
| 92 |
|
| 93 |
__gpu_kernel void atanfKernel(const float *X, float *Out, |
| 94 |
size_t NumElements) noexcept { |
| 95 |
runKernelBody<__nv_atanf>(NumElements, Out, X); |
| 96 |
} |
| 97 |
|
| 98 |
__gpu_kernel void atan2fKernel(const float *X, const float *Y, float *Out, |
| 99 |
size_t NumElements) noexcept { |
| 100 |
runKernelBody<__nv_atan2f>(NumElements, Out, X, Y); |
| 101 |
} |
| 102 |
|
| 103 |
__gpu_kernel void atanhfKernel(const float *X, float *Out, |
| 104 |
size_t NumElements) noexcept { |
| 105 |
runKernelBody<__nv_atanhf>(NumElements, Out, X); |
| 106 |
} |
| 107 |
|
| 108 |
__gpu_kernel void cbrtKernel(const double *X, double *Out, |
| 109 |
size_t NumElements) noexcept { |
| 110 |
runKernelBody<__nv_cbrt>(NumElements, Out, X); |
| 111 |
} |
| 112 |
|
| 113 |
__gpu_kernel void cbrtfKernel(const float *X, float *Out, |
| 114 |
size_t NumElements) noexcept { |
| 115 |
runKernelBody<__nv_cbrtf>(NumElements, Out, X); |
| 116 |
} |
| 117 |
|
| 118 |
__gpu_kernel void cosKernel(const double *X, double *Out, |
| 119 |
size_t NumElements) noexcept { |
| 120 |
runKernelBody<__nv_cos>(NumElements, Out, X); |
| 121 |
} |
| 122 |
|
| 123 |
__gpu_kernel void cosfKernel(const float *X, float *Out, |
| 124 |
size_t NumElements) noexcept { |
| 125 |
runKernelBody<__nv_cosf>(NumElements, Out, X); |
| 126 |
} |
| 127 |
|
| 128 |
__gpu_kernel void coshfKernel(const float *X, float *Out, |
| 129 |
size_t NumElements) noexcept { |
| 130 |
runKernelBody<__nv_coshf>(NumElements, Out, X); |
| 131 |
} |
| 132 |
|
| 133 |
__gpu_kernel void cospifKernel(const float *X, float *Out, |
| 134 |
size_t NumElements) noexcept { |
| 135 |
runKernelBody<__nv_cospif>(NumElements, Out, X); |
| 136 |
} |
| 137 |
|
| 138 |
__gpu_kernel void erffKernel(const float *X, float *Out, |
| 139 |
size_t NumElements) noexcept { |
| 140 |
runKernelBody<__nv_erff>(NumElements, Out, X); |
| 141 |
} |
| 142 |
|
| 143 |
__gpu_kernel void expKernel(const double *X, double *Out, |
| 144 |
size_t NumElements) noexcept { |
| 145 |
runKernelBody<__nv_exp>(NumElements, Out, X); |
| 146 |
} |
| 147 |
|
| 148 |
__gpu_kernel void expfKernel(const float *X, float *Out, |
| 149 |
size_t NumElements) noexcept { |
| 150 |
runKernelBody<__nv_expf>(NumElements, Out, X); |
| 151 |
} |
| 152 |
|
| 153 |
__gpu_kernel void exp10Kernel(const double *X, double *Out, |
| 154 |
size_t NumElements) noexcept { |
| 155 |
runKernelBody<__nv_exp10>(NumElements, Out, X); |
| 156 |
} |
| 157 |
|
| 158 |
__gpu_kernel void exp10fKernel(const float *X, float *Out, |
| 159 |
size_t NumElements) noexcept { |
| 160 |
runKernelBody<__nv_exp10f>(NumElements, Out, X); |
| 161 |
} |
| 162 |
|
| 163 |
__gpu_kernel void exp2Kernel(const double *X, double *Out, |
| 164 |
size_t NumElements) noexcept { |
| 165 |
runKernelBody<__nv_exp2>(NumElements, Out, X); |
| 166 |
} |
| 167 |
|
| 168 |
__gpu_kernel void exp2fKernel(const float *X, float *Out, |
| 169 |
size_t NumElements) noexcept { |
| 170 |
runKernelBody<__nv_exp2f>(NumElements, Out, X); |
| 171 |
} |
| 172 |
|
| 173 |
__gpu_kernel void expm1Kernel(const double *X, double *Out, |
| 174 |
size_t NumElements) noexcept { |
| 175 |
runKernelBody<__nv_expm1>(NumElements, Out, X); |
| 176 |
} |
| 177 |
|
| 178 |
__gpu_kernel void expm1fKernel(const float *X, float *Out, |
| 179 |
size_t NumElements) noexcept { |
| 180 |
runKernelBody<__nv_expm1f>(NumElements, Out, X); |
| 181 |
} |
| 182 |
|
| 183 |
__gpu_kernel void hypotKernel(const double *X, const double *Y, double *Out, |
| 184 |
size_t NumElements) noexcept { |
| 185 |
runKernelBody<__nv_hypot>(NumElements, Out, X, Y); |
| 186 |
} |
| 187 |
|
| 188 |
__gpu_kernel void hypotfKernel(const float *X, const float *Y, float *Out, |
| 189 |
size_t NumElements) noexcept { |
| 190 |
runKernelBody<__nv_hypotf>(NumElements, Out, X, Y); |
| 191 |
} |
| 192 |
|
| 193 |
__gpu_kernel void logKernel(const double *X, double *Out, |
| 194 |
size_t NumElements) noexcept { |
| 195 |
runKernelBody<__nv_log>(NumElements, Out, X); |
| 196 |
} |
| 197 |
|
| 198 |
__gpu_kernel void logfKernel(const float *X, float *Out, |
| 199 |
size_t NumElements) noexcept { |
| 200 |
runKernelBody<__nv_logf>(NumElements, Out, X); |
| 201 |
} |
| 202 |
|
| 203 |
__gpu_kernel void log10Kernel(const double *X, double *Out, |
| 204 |
size_t NumElements) noexcept { |
| 205 |
runKernelBody<__nv_log10>(NumElements, Out, X); |
| 206 |
} |
| 207 |
|
| 208 |
__gpu_kernel void log10fKernel(const float *X, float *Out, |
| 209 |
size_t NumElements) noexcept { |
| 210 |
runKernelBody<__nv_log10f>(NumElements, Out, X); |
| 211 |
} |
| 212 |
|
| 213 |
__gpu_kernel void log1pKernel(const double *X, double *Out, |
| 214 |
size_t NumElements) noexcept { |
| 215 |
runKernelBody<__nv_log1p>(NumElements, Out, X); |
| 216 |
} |
| 217 |
|
| 218 |
__gpu_kernel void log1pfKernel(const float *X, float *Out, |
| 219 |
size_t NumElements) noexcept { |
| 220 |
runKernelBody<__nv_log1pf>(NumElements, Out, X); |
| 221 |
} |
| 222 |
|
| 223 |
__gpu_kernel void log2Kernel(const double *X, double *Out, |
| 224 |
size_t NumElements) noexcept { |
| 225 |
runKernelBody<__nv_log2>(NumElements, Out, X); |
| 226 |
} |
| 227 |
|
| 228 |
__gpu_kernel void log2fKernel(const float *X, float *Out, |
| 229 |
size_t NumElements) noexcept { |
| 230 |
runKernelBody<__nv_log2f>(NumElements, Out, X); |
| 231 |
} |
| 232 |
|
| 233 |
__gpu_kernel void powfKernel(const float *X, float *Y, float *Out, |
| 234 |
size_t NumElements) noexcept { |
| 235 |
runKernelBody<__nv_powf>(NumElements, Out, X, Y); |
| 236 |
} |
| 237 |
|
| 238 |
__gpu_kernel void powfRoundedExponentKernel(const float *X, float *Y, |
| 239 |
float *Out, |
| 240 |
size_t NumElements) noexcept { |
| 241 |
runKernelBody<powfRoundedExponent>(NumElements, Out, X, Y); |
| 242 |
} |
| 243 |
|
| 244 |
__gpu_kernel void sinKernel(const double *X, double *Out, |
| 245 |
size_t NumElements) noexcept { |
| 246 |
runKernelBody<__nv_sin>(NumElements, Out, X); |
| 247 |
} |
| 248 |
|
| 249 |
__gpu_kernel void sinfKernel(const float *X, float *Out, |
| 250 |
size_t NumElements) noexcept { |
| 251 |
runKernelBody<__nv_sinf>(NumElements, Out, X); |
| 252 |
} |
| 253 |
|
| 254 |
__gpu_kernel void sincosSinKernel(const double *X, double *Out, |
| 255 |
size_t NumElements) noexcept { |
| 256 |
runKernelBody<sincosSin>(NumElements, Out, X); |
| 257 |
} |
| 258 |
|
| 259 |
__gpu_kernel void sincosCosKernel(const double *X, double *Out, |
| 260 |
size_t NumElements) noexcept { |
| 261 |
runKernelBody<sincosCos>(NumElements, Out, X); |
| 262 |
} |
| 263 |
|
| 264 |
__gpu_kernel void sincosfSinKernel(const float *X, float *Out, |
| 265 |
size_t NumElements) noexcept { |
| 266 |
runKernelBody<sincosfSin>(NumElements, Out, X); |
| 267 |
} |
| 268 |
|
| 269 |
__gpu_kernel void sincosfCosKernel(const float *X, float *Out, |
| 270 |
size_t NumElements) noexcept { |
| 271 |
runKernelBody<sincosfCos>(NumElements, Out, X); |
| 272 |
} |
| 273 |
|
| 274 |
__gpu_kernel void sinhfKernel(const float *X, float *Out, |
| 275 |
size_t NumElements) noexcept { |
| 276 |
runKernelBody<__nv_sinhf>(NumElements, Out, X); |
| 277 |
} |
| 278 |
|
| 279 |
__gpu_kernel void sinpifKernel(const float *X, float *Out, |
| 280 |
size_t NumElements) noexcept { |
| 281 |
runKernelBody<__nv_sinpif>(NumElements, Out, X); |
| 282 |
} |
| 283 |
|
| 284 |
__gpu_kernel void tanKernel(const double *X, double *Out, |
| 285 |
size_t NumElements) noexcept { |
| 286 |
runKernelBody<__nv_tan>(NumElements, Out, X); |
| 287 |
} |
| 288 |
|
| 289 |
__gpu_kernel void tanfKernel(const float *X, float *Out, |
| 290 |
size_t NumElements) noexcept { |
| 291 |
runKernelBody<__nv_tanf>(NumElements, Out, X); |
| 292 |
} |
| 293 |
|
| 294 |
__gpu_kernel void tanhfKernel(const float *X, float *Out, |
| 295 |
size_t NumElements) noexcept { |
| 296 |
runKernelBody<__nv_tanhf>(NumElements, Out, X); |
| 297 |
} |
| 298 |
} // extern "C" |
| 299 |
|
| 300 |
#endif // CUDA_MATH_FOUND |