source: sasmodels/sasmodels/kernel_header.c @ 3d6526d

core_shell_microgelsmagnetic_modelticket-1257-vesicle-productticket_1156ticket_1265_superballticket_822_more_unit_tests
Last change on this file since 3d6526d was 3d6526d, checked in by Paul Kienzle <pkienzle@…>, 6 years ago

tcc doesn't have the C99 cbrt function
  • Property mode set to 100644
File size: 8.4 KB
Line 
1#ifdef __OPENCL_VERSION__
2# define USE_OPENCL
3#elif defined(_OPENMP)
4# define USE_OPENMP
5#endif
6
7// If opencl is not available, then we are compiling a C function
8// Note: if using a C++ compiler, then define kernel as extern "C"
9#ifdef USE_OPENCL
10   typedef int int32_t;
11#  if defined(USE_SINCOS)
12#    define SINCOS(angle,svar,cvar) svar=sincos(angle,&cvar)
13#  else
14#    define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0)
15#  endif
16   // Intel CPU on Mac gives strange values for erf(); on the verified
17   // platforms (intel, nvidia, amd), the cephes erf() is significantly
18   // faster than that available in the native OpenCL.
19   #define NEED_ERF
20   // OpenCL only has type generic math
21   #define expf exp
22   #ifndef NEED_ERF
23   #  define erff erf
24   #  define erfcf erfc
25   #endif
26#else // !USE_OPENCL
27// Use SAS_DOUBLE to force the use of double even for float kernels
28#  define SAS_DOUBLE dou ## ble
29#  ifdef __cplusplus
30      #include <cstdio>
31      #include <cmath>
32      using namespace std;
33      #if defined(_MSC_VER)
34         #include <limits>
35         #include <float.h>
36         #define kernel extern "C" __declspec( dllexport )
37         inline double trunc(double x) { return x>=0?floor(x):-floor(-x); }
38         inline double fmin(double x, double y) { return x>y ? y : x; }
39         inline double fmax(double x, double y) { return x<y ? y : x; }
40         #define isnan(x) _isnan(x)
41         #define isinf(x) (!_finite(x))
42         #define isfinite(x) _finite(x)
43         #define NAN (std::numeric_limits<double>::quiet_NaN()) // non-signalling NaN
44         #define INFINITY (std::numeric_limits<double>::infinity())
45         #define NEED_ERF
46         #define NEED_EXPM1
47         #define NEED_TGAMMA
48     #else
49         #define kernel extern "C"
50         #include <cstdint>
51     #endif
52     inline void SINCOS(double angle, double &svar, double &cvar) { svar=sin(angle); cvar=cos(angle); }
53#  else // !__cplusplus
54     #include <inttypes.h>  // C99 guarantees that int32_t types is here
55     #include <stdio.h>
56     #if defined(__TINYC__)
57         typedef int int32_t;
58         #include <math.h>
59         // TODO: check isnan is correct
60         inline double _isnan(double x) { return x != x; } // hope this doesn't optimize away!
61         #undef isnan
62         #define isnan(x) _isnan(x)
63         // Defeat the double->float conversion since we don't have tgmath
64         inline SAS_DOUBLE trunc(SAS_DOUBLE x) { return x>=0?floor(x):-floor(-x); }
65         inline SAS_DOUBLE fmin(SAS_DOUBLE x, SAS_DOUBLE y) { return x>y ? y : x; }
66         inline SAS_DOUBLE fmax(SAS_DOUBLE x, SAS_DOUBLE y) { return x<y ? y : x; }
67         #define NEED_ERF
68         #define NEED_EXPM1
69         #define NEED_TGAMMA
70         #define NEED_CBRT
71         // expf missing from windows?
72         #define expf exp
73     #else
74         #include <tgmath.h> // C99 type-generic math, so sin(float) => sinf
75     #endif
76     // MSVC doesn't support C99, so no need for dllexport on C99 branch
77     #define kernel
78     #define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0)
79#  endif  // !__cplusplus
80#  define global
81#  define local
82#  define constant const
83// OpenCL powr(a,b) = C99 pow(a,b), b >= 0
84// OpenCL pown(a,b) = C99 pow(a,b), b integer
85#  define powr(a,b) pow(a,b)
86#  define pown(a,b) pow(a,b)
87#endif // !USE_OPENCL
88
89#if defined(NEED_CBRT)
90   #define cbrt(_x) (pow(_x, 0.33333333333333333333333))
91#endif
92
93#if defined(NEED_EXPM1)
94   // TODO: precision is a half digit lower than numpy on mac in [1e-7, 0.5]
95   // Run "explore/precision.py sas_expm1" to see this (may have to fiddle
96   // the xrange for log to see the complete range).
97   static SAS_DOUBLE expm1(SAS_DOUBLE x_in) {
98      double x = (double)x_in;  // go back to float for single precision kernels
99      // Adapted from the cephes math library.
100      // Copyright 1984 - 1992 by Stephen L. Moshier
101      if (x != x || x == 0.0) {
102         return x; // NaN and +/- 0
103      } else if (x < -0.5 || x > 0.5) {
104         return exp(x) - 1.0;
105      } else {
106         const double xsq = x*x;
107         const double p = (((
108            +1.2617719307481059087798E-4)*xsq
109            +3.0299440770744196129956E-2)*xsq
110            +9.9999999999999999991025E-1);
111         const double q = ((((
112            +3.0019850513866445504159E-6)*xsq
113            +2.5244834034968410419224E-3)*xsq
114            +2.2726554820815502876593E-1)*xsq
115            +2.0000000000000000000897E0);
116         double r = x * p;
117         r =  r / (q - r);
118         return r+r;
119       }
120   }
121#endif
122
123// Standard mathematical constants:
124//   M_E, M_LOG2E, M_LOG10E, M_LN2, M_LN10, M_PI, M_PI_2=pi/2, M_PI_4=pi/4,
125//   M_1_PI=1/pi, M_2_PI=2/pi, M_2_SQRTPI=2/sqrt(pi), SQRT2, SQRT1_2=sqrt(1/2)
126// OpenCL defines M_constant_F for float constants, and nothing if double
127// is not enabled on the card, which is why these constants may be missing
128#ifndef M_PI
129#  define M_PI 3.141592653589793
130#endif
131#ifndef M_PI_2
132#  define M_PI_2 1.570796326794897
133#endif
134#ifndef M_PI_4
135#  define M_PI_4 0.7853981633974483
136#endif
137#ifndef M_E
138#  define M_E 2.718281828459045091
139#endif
140#ifndef M_SQRT1_2
141#  define M_SQRT1_2 0.70710678118654746
142#endif
143
144// Non-standard function library
145// pi/180, used for converting between degrees and radians
146// 4/3 pi for computing sphere volumes
147// square and cube for computing squares and cubes
148#ifndef M_PI_180
149#  define M_PI_180 0.017453292519943295
150#endif
151#ifndef M_4PI_3
152#  define M_4PI_3 4.18879020478639
153#endif
154inline double square(double x) { return x*x; }
155inline double cube(double x) { return x*x*x; }
156inline double sas_sinx_x(double x) { return x==0 ? 1.0 : sin(x)/x; }
157
158// CRUFT: support old style models with orientation received qx, qy and angles
159
160// To rotate from the canonical position to theta, phi, psi, first rotate by
161// psi about the major axis, oriented along z, which is a rotation in the
162// detector plane xy. Next rotate by theta about the y axis, aligning the major
163// axis in the xz plane. Finally, rotate by phi in the detector plane xy.
164// To compute the scattering, undo these rotations in reverse order:
165//     rotate in xy by -phi, rotate in xz by -theta, rotate in xy by -psi
166// The returned q is the length of the q vector and (xhat, yhat, zhat) is a unit
167// vector in the q direction.
168// To change between counterclockwise and clockwise rotation, change the
169// sign of phi and psi.
170
171#if 1
172//think cos(theta) should be sin(theta) in new coords, RKH 11Jan2017
173#define ORIENT_SYMMETRIC(qx, qy, theta, phi, q, sn, cn) do { \
174    SINCOS(phi*M_PI_180, sn, cn); \
175    q = sqrt(qx*qx + qy*qy); \
176    cn  = (q==0. ? 1.0 : (cn*qx + sn*qy)/q * sin(theta*M_PI_180));  \
177    sn = sqrt(1 - cn*cn); \
178    } while (0)
179#else
180// SasView 3.x definition of orientation
181#define ORIENT_SYMMETRIC(qx, qy, theta, phi, q, sn, cn) do { \
182    SINCOS(theta*M_PI_180, sn, cn); \
183    q = sqrt(qx*qx + qy*qy);\
184    cn = (q==0. ? 1.0 : (cn*cos(phi*M_PI_180)*qx + sn*qy)/q); \
185    sn = sqrt(1 - cn*cn); \
186    } while (0)
187#endif
188
189#if 1
190#define ORIENT_ASYMMETRIC(qx, qy, theta, phi, psi, q, xhat, yhat, zhat) do { \
191    q = sqrt(qx*qx + qy*qy); \
192    const double qxhat = qx/q; \
193    const double qyhat = qy/q; \
194    double sin_theta, cos_theta; \
195    double sin_phi, cos_phi; \
196    double sin_psi, cos_psi; \
197    SINCOS(theta*M_PI_180, sin_theta, cos_theta); \
198    SINCOS(phi*M_PI_180, sin_phi, cos_phi); \
199    SINCOS(psi*M_PI_180, sin_psi, cos_psi); \
200    xhat = qxhat*(-sin_phi*sin_psi + cos_theta*cos_phi*cos_psi) \
201         + qyhat*( cos_phi*sin_psi + cos_theta*sin_phi*cos_psi); \
202    yhat = qxhat*(-sin_phi*cos_psi - cos_theta*cos_phi*sin_psi) \
203         + qyhat*( cos_phi*cos_psi - cos_theta*sin_phi*sin_psi); \
204    zhat = qxhat*(-sin_theta*cos_phi) \
205         + qyhat*(-sin_theta*sin_phi); \
206    } while (0)
207#else
208// SasView 3.x definition of orientation
209#define ORIENT_ASYMMETRIC(qx, qy, theta, phi, psi, q, cos_alpha, cos_mu, cos_nu) do { \
210    q = sqrt(qx*qx + qy*qy); \
211    const double qxhat = qx/q; \
212    const double qyhat = qy/q; \
213    double sin_theta, cos_theta; \
214    double sin_phi, cos_phi; \
215    double sin_psi, cos_psi; \
216    SINCOS(theta*M_PI_180, sin_theta, cos_theta); \
217    SINCOS(phi*M_PI_180, sin_phi, cos_phi); \
218    SINCOS(psi*M_PI_180, sin_psi, cos_psi); \
219    cos_alpha = cos_theta*cos_phi*qxhat + sin_theta*qyhat; \
220    cos_mu = (-sin_theta*cos_psi*cos_phi - sin_psi*sin_phi)*qxhat + cos_theta*cos_psi*qyhat; \
221    cos_nu = (-cos_phi*sin_psi*sin_theta + sin_phi*cos_psi)*qxhat + sin_psi*cos_theta*qyhat; \
222    } while (0)
223#endif
Note: See TracBrowser for help on using the repository browser.