CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
complex.h
Go to the documentation of this file.
1 /***************************************************************************************************
2  * Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without modification, are permitted
5  * provided that the following conditions are met:
6  * * Redistributions of source code must retain the above copyright notice, this list of
7  * conditions and the following disclaimer.
8  * * Redistributions in binary form must reproduce the above copyright notice, this list of
9  * conditions and the following disclaimer in the documentation and/or other materials
10  * provided with the distribution.
11  * * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used
12  * to endorse or promote products derived from this software without specific prior written
13  * permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
17  * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
19  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
20  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
21  * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
22  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
23  *
24  **************************************************************************************************/
25 #pragma once
26 
27 #include <cuComplex.h>
28 #include <cstdint>
29 
30 #include "cutlass/cutlass.h"
31 #include "cutlass/half.h"
32 #include "cutlass/real.h"
33 
34 #if !defined(__CUDACC_RTC__)
35 #include <iosfwd>
36 #endif
37 
38 namespace cutlass {
39 
41 
43 enum class ComplexTransform {
44  kNone,
46 };
47 
49 
50 //
51 // Accessors for CUDA complex types
52 //
53 
56 float const &real(cuFloatComplex const &z) { return z.x; }
57 
60 float &real(cuFloatComplex &z) { return z.x; }
61 
64 double const &real(cuDoubleComplex const &z) { return z.x; }
65 
68 double &real(cuDoubleComplex &z) { return z.x; }
69 
72 float const &imag(cuFloatComplex const &z) { return z.y; }
73 
76 float &imag(cuFloatComplex &z) { return z.y; }
77 
80 double const &imag(cuDoubleComplex const &z) { return z.y; }
81 
84 double &imag(cuDoubleComplex &z) { return z.y; }
85 
87 
90 
91 template <typename T>
92 class complex
93 {
94  public:
96 
97  private:
98  //
99  // Data members
100  //
101 
103  T _real;
104 
106  T _imag;
107 
108  public:
109 
110 //
111 // Methods
112 //
113 
116  complex(T r = T(0)) : _real(r), _imag(T(0)) {}
117 
120  complex(T r, T i) : _real(r), _imag(i) {}
121  //
123  template<typename A>
125  complex(complex<A> const &z) : _real(static_cast<T>(z.real())), _imag(static_cast<T>(z.imag())) {}
126 
129  complex(cuFloatComplex const &z) : _real(static_cast<T>(cuCrealf(z))), _imag(static_cast<T>(cuCimagf(z))) {}
130 
133  complex(cuDoubleComplex const &z) : _real(static_cast<T>(cuCreal(z))), _imag(static_cast<T>(cuCimag(z))) {}
134 
136  template<typename A>
139  {
140  _real = static_cast<T>(z.real());
141  _imag = static_cast<T>(z.imag());
142  return *this;
143  }
144 
146  CUTLASS_HOST_DEVICE bool operator==(complex<T> const &rhs) const {
147  return this->real() == rhs.real() && this->imag() == rhs.imag();
148  }
149 
151  CUTLASS_HOST_DEVICE bool operator!=(complex<T> const &rhs) const {
152  return !(*this == rhs);
153  }
154 
156  template <typename A>
158  return complex<T>(this->real() + rhs.real(), this->imag() + rhs.imag());
159  }
160 
162  template <typename A>
164  return complex<T>(this->real() - rhs.real(), this->imag() - rhs.imag());
165  }
166 
168  template <typename A>
170  return complex<T>(this->real() * rhs.real() - this->imag() * rhs.imag(),
171  this->real() * rhs.imag() + this->imag() * rhs.real());
172  }
173 
175  template <typename A>
177  return complex<T>(this->real() * s, this->imag() * s);
178  }
179 
181  template <typename A>
183  T d = (rhs.real() * (rhs) + rhs.imag() * rhs.imag());
184 
185  return complex<T>((this->real() * (rhs) + this->imag() * rhs.imag()) / d,
186  (this->imag() * (rhs)-this->real() * rhs.imag()) / d);
187  }
188 
190  template <typename A>
192  return complex<T>(this->real() / s, this->imag() / s);
193  }
194 
196  template <typename A>
198  *this = *this + rhs;
199  return *this;
200  }
201 
203  template <typename A>
205  *this = *this - rhs;
206  return *this;
207  }
208 
210  template <typename A>
212  *this = *this * rhs;
213  return *this;
214  }
215 
217  template <typename A>
219  *this = *this * s;
220  return *this;
221  }
222 
224  template <typename A>
226  *this = *this / rhs;
227  return *this;
228  }
229 
232  T const &real() const { return _real; }
233 
236  T &real() { return _real; }
237 
240  T const &imag() const { return _imag; }
241 
244  T &imag() { return _imag; }
245 
248  explicit operator cuFloatComplex() const { return make_cuFloatComplex(float(real()), float(imag())); }
249 
252  explicit operator cuDoubleComplex() const { return make_cuDoubleComplex(real(), imag()); }
253 };
254 
256 
257 //
258 // Accessors for complex template
259 //
260 
262 template <typename T>
263 CUTLASS_HOST_DEVICE T const &real(complex<T> const &z) {
264  return z.real();
265 }
266 
268 template <typename T>
270  return z.real();
271 }
272 
274 template <typename T>
275 CUTLASS_HOST_DEVICE T const &imag(complex<T> const &z) {
276  return z.imag();
277 }
278 
280 template <typename T>
282  return z.imag();
283 }
284 
285 //
286 // Output operators
287 //
288 
289 #if !defined(__CUDACC_RTC__)
290 template <typename T>
291 std::ostream &operator<<(std::ostream &out, complex<T> const &z) {
292  T _r = real(z);
293  T _i = imag(z);
294 
295  if (bool(_i)) {
296  return out << _r << "+i" << _i;
297  }
298  return out << _r;
299 }
300 #endif
301 
302 //
303 // Non-member operators defined for complex types
304 //
305 
306 
307 //
308 // Non-member functions defined for complex numbers
309 //
310 
312 template <typename T>
314  return sqrt(norm(z));
315 }
316 
318 template <typename T>
320  return atan2(imag(z), real(z));
321 }
322 
324 template <typename T>
325 CUTLASS_HOST_DEVICE T norm(T const &z) {
326  return z * z;
327 }
328 
330 template <>
331 CUTLASS_HOST_DEVICE int8_t norm(int8_t const &z) {
332  return static_cast<int8_t>(z * z);
333 }
334 
336 template <typename T>
338  return real(z) * real(z) + imag(z) * imag(z);
339 }
340 
342 template <typename T, typename R>
343 CUTLASS_HOST_DEVICE R norm_accumulate(T const &x, R const & accumulator) {
344  return accumulator + static_cast<R>(x) * static_cast<R>(x);
345 }
346 
348 template <typename T, typename R>
349 CUTLASS_HOST_DEVICE R norm_accumulate(complex<T> const &z, R const &accumulator) {
350  return accumulator + static_cast<R>(real(z)) * static_cast<R>(real(z)) +
351  static_cast<R>(imag(z)) * static_cast<R>(imag(z));
352 }
353 
355 template <typename T>
357  return complex<T>(real(z), -imag(z));
358 }
359 
361 template <typename T>
363  T d = real(z) * real(z) + imag(z) * imag(z) + T(1);
364  return complex<T>((T(2) * real(z)) / d, (T(2) * imag(z)) / d);
365 }
366 
368 template <typename T>
369 CUTLASS_HOST_DEVICE complex<T> polar(T const &r, T const &theta = T()) {
370  return complex<T>(r * cos(theta), r * sin(theta));
371 }
372 
374 template <typename T>
376  return complex<T>(real(z) * cos(imag(z)), real(z) * sin(imag(z)));
377 }
378 
380 template <typename T>
382  return complex<T>(log(abs(z)), arg(z));
383 }
384 
386 template <typename T>
388  return log(z) / T(log(T(10)));
389 }
390 
392 template <typename T>
394  return sqrt(T(2)) / T(2) *
395  complex<T>(sqrt(sqrt(norm(z)) + real(z)),
396  (imag(z) < 0 ? T(-1) : T(1)) * sqrt(sqrt(norm(z)) - real(z)));
397 }
398 
400 template <typename T>
402  return (exp(z) + exp(-z)) / T(2);
403 }
404 
406 template <typename T>
408  return (exp(-z) - exp(z)) * complex<T>(T(0), T(1) / T(2));
409 }
410 
412 
414 template <typename T>
415 struct RealType< complex<T> > {
416  using Type = T;
417 };
418 
420 
421 template <>
423 cutlass::complex<half_t> from_real<cutlass::complex<half_t> >(double r) {
424  return cutlass::complex<half_t>(half_t(r));
425 }
426 
427 template <>
429 cutlass::complex<float> from_real<cutlass::complex<float> >(double r) {
430  return cutlass::complex<float>(float(r));
431 }
432 
433 template <>
435 cutlass::complex<double> from_real<cutlass::complex<double> >(double r) {
436  return cutlass::complex<double>(r);
437 }
438 
440 
441 } // namespace cutlass
442 
CUTLASS_HOST_DEVICE complex< T > cos(complex< T > const &z)
Computes the cosine of complex z.
Definition: complex.h:401
CUTLASS_HOST_DEVICE complex(cuDoubleComplex const &z)
Conversion from cuDoubleComplex.
Definition: complex.h:133
CUTLASS_HOST_DEVICE complex< T > & operator-=(complex< A > const &rhs)
Subtraction.
Definition: complex.h:204
CUTLASS_HOST_DEVICE complex< T > operator-(complex< A > const &rhs) const
Subtraction.
Definition: complex.h:163
Definition: aligned_buffer.h:35
ComplexTransform
Enumeraed type describing a transformation on a complex value.
Definition: complex.h:43
CUTLASS_HOST_DEVICE complex< T > & operator=(complex< A > const &z)
Assignment.
Definition: complex.h:138
CUTLASS_HOST_DEVICE T abs(complex< T > const &z)
Returns the magnitude of the complex number.
Definition: complex.h:313
CUTLASS_HOST_DEVICE float const & imag(cuFloatComplex const &z)
Returns the imaginary part of the complex number.
Definition: complex.h:72
CUTLASS_HOST_DEVICE complex(T r=T(0))
Constructor.
Definition: complex.h:116
CUTLASS_HOST_DEVICE R norm_accumulate(T const &x, R const &accumulator)
Norm-accumulate calculation.
Definition: complex.h:343
Defines a class for using IEEE half-precision floating-point types in host or device code...
CUTLASS_HOST_DEVICE T norm(T const &z)
Returns the squared magnitude of a real number.
Definition: complex.h:325
CUTLASS_HOST_DEVICE complex< T > polar(T const &r, T const &theta=T())
Returns a complex number with magnitude r and phase theta.
Definition: complex.h:369
IEEE half-precision floating-point type.
Definition: half.h:126
CUTLASS_HOST_DEVICE bool operator!=(complex< T > const &rhs) const
Inequality operator.
Definition: complex.h:151
CUTLASS_HOST_DEVICE float const & real(cuFloatComplex const &z)
Returns the real part of the complex number.
Definition: complex.h:56
CUTLASS_HOST_DEVICE complex< T > exp(complex< T > const &z)
Computes the complex exponential of z.
Definition: complex.h:375
CUTLASS_HOST_DEVICE T const & imag() const
Accesses the imaginary part of the complex number.
Definition: complex.h:240
CUTLASS_HOST_DEVICE T arg(complex< T > const &z)
Returns the magnitude of the complex number.
Definition: complex.h:319
CUTLASS_HOST_DEVICE complex< T > log(complex< T > const &z)
Computes the complex exponential of z.
Definition: complex.h:381
CUTLASS_HOST_DEVICE complex(T r, T i)
Constructor.
Definition: complex.h:120
CUTLASS_HOST_DEVICE bool operator==(complex< T > const &rhs) const
Equality operator.
Definition: complex.h:146
CUTLASS_HOST_DEVICE complex< T > conj(complex< T > const &z)
Returns the complex conjugate.
Definition: complex.h:356
CUTLASS_HOST_DEVICE complex< T > & operator*=(complex< A > const &rhs)
Multiplication.
Definition: complex.h:211
#define CUTLASS_HOST_DEVICE
Definition: cutlass.h:89
CUTLASS_HOST_DEVICE T & imag()
Accesses the imaginary part of the complex number.
Definition: complex.h:244
CUTLASS_HOST_DEVICE complex< T > sin(complex< T > const &z)
Computes the sin of complex z.
Definition: complex.h:407
CUTLASS_HOST_DEVICE complex(cuFloatComplex const &z)
Conversion from cuFloatComplex.
Definition: complex.h:129
CUTLASS_HOST_DEVICE complex< T > operator+(complex< A > const &rhs) const
Addition.
Definition: complex.h:157
CUTLASS_HOST_DEVICE T const & real() const
Accesses the real part of the complex number.
Definition: complex.h:232
CUTLASS_HOST_DEVICE complex< T > proj(complex< T > const &z)
Projects the complex number z onto the Riemann sphere.
Definition: complex.h:362
Definition: complex.h:92
CUTLASS_HOST_DEVICE complex< T > operator*(complex< A > const &rhs) const
Multiplication.
Definition: complex.h:169
CUTLASS_HOST_DEVICE complex< T > & operator*=(A s)
Scalar multiplication.
Definition: complex.h:218
CUTLASS_HOST_DEVICE T & real()
Accesses the real part of the complex number.
Definition: complex.h:236
CUTLASS_HOST_DEVICE complex< T > operator*(A const &s) const
Scalar Multiplication.
Definition: complex.h:176
CUTLASS_HOST_DEVICE complex< T > & operator+=(complex< A > const &rhs)
Addition.
Definition: complex.h:197
CUTLASS_HOST_DEVICE complex< T > operator/(complex< A > const &rhs) const
Division.
Definition: complex.h:182
CUTLASS_HOST_DEVICE complex< T > & operator/=(complex< A > const &rhs)
Division.
Definition: complex.h:225
CUTLASS_HOST_DEVICE complex(complex< A > const &z)
Constructor.
Definition: complex.h:125
CUTLASS_HOST_DEVICE complex< T > log10(complex< T > const &z)
Computes the complex exponential of z.
Definition: complex.h:387
Used to determine the real-valued underlying type of a numeric type T.
Definition: real.h:31
Basic include for CUTLASS.
CUTLASS_HOST_DEVICE complex< T > operator/(A const &s) const
Scalar Division.
Definition: complex.h:191
CUTLASS_HOST_DEVICE complex< T > sqrt(complex< T > const &z)
Computes the square root of complex number z.
Definition: complex.h:393
T Type
Definition: complex.h:416