[3832f27] | 1 | #line 1 "kernel_template.c" |
---|
[f734e7d] | 2 | // GENERATED CODE --- DO NOT EDIT --- |
---|
| 3 | // Code is produced by sasmodels.gen from sasmodels/models/MODEL.c |
---|
| 4 | |
---|
| 5 | #ifdef __OPENCL_VERSION__ |
---|
| 6 | # define USE_OPENCL |
---|
| 7 | #endif |
---|
| 8 | |
---|
[e3a9733] | 9 | #define USE_KAHAN_SUMMATION 0 |
---|
| 10 | |
---|
[f734e7d] | 11 | // If opencl is not available, then we are compiling a C function |
---|
| 12 | // Note: if using a C++ compiler, then define kernel as extern "C" |
---|
| 13 | #ifndef USE_OPENCL |
---|
[5efe850] | 14 | // Use SAS_DOUBLE to force the use of double even for float kernels |
---|
| 15 | # define SAS_DOUBLE dou ## ble |
---|
[f734e7d] | 16 | # ifdef __cplusplus |
---|
[960cd80] | 17 | #include <cstdio> |
---|
| 18 | #include <cmath> |
---|
| 19 | using namespace std; |
---|
| 20 | #if defined(_MSC_VER) |
---|
[29f27df] | 21 | #include <limits> |
---|
[caf768d] | 22 | #include <float.h> |
---|
| 23 | #define kernel extern "C" __declspec( dllexport ) |
---|
[750ffa5] | 24 | inline double trunc(double x) { return x>=0?floor(x):-floor(-x); } |
---|
[960cd80] | 25 | inline double fmin(double x, double y) { return x>y ? y : x; } |
---|
| 26 | inline double fmax(double x, double y) { return x<y ? y : x; } |
---|
[2a55a6f] | 27 | #define isnan(x) _isnan(x) |
---|
| 28 | #define isinf(x) (!_finite(x)) |
---|
| 29 | #define isfinite(x) _finite(x) |
---|
[960cd80] | 30 | #define NAN (std::numeric_limits<double>::quiet_NaN()) // non-signalling NaN |
---|
[2a55a6f] | 31 | #define INFINITY (std::numeric_limits<double>::infinity()) |
---|
[98cb4d7] | 32 | #define NEED_EXPM1 |
---|
| 33 | #define NEED_TGAMMA |
---|
[6a0d6aa] | 34 | #define NEED_ERF |
---|
[f734e7d] | 35 | #else |
---|
[caf768d] | 36 | #define kernel extern "C" |
---|
[f734e7d] | 37 | #endif |
---|
[750ffa5] | 38 | inline void SINCOS(double angle, double &svar, double &cvar) { svar=sin(angle); cvar=cos(angle); } |
---|
[f734e7d] | 39 | # else |
---|
[95e861b] | 40 | #include <stdio.h> |
---|
[2a55a6f] | 41 | #if defined(__TINYC__) |
---|
| 42 | #include <math.h> |
---|
| 43 | // TODO: test isnan |
---|
| 44 | inline double _isnan(double x) { return x != x; } // hope this doesn't optimize away! |
---|
| 45 | #undef isnan |
---|
| 46 | #define isnan(x) _isnan(x) |
---|
[5efe850] | 47 | // Defeat the double->float conversion since we don't have tgmath |
---|
| 48 | inline SAS_DOUBLE trunc(SAS_DOUBLE x) { return x>=0?floor(x):-floor(-x); } |
---|
| 49 | inline SAS_DOUBLE fmin(SAS_DOUBLE x, SAS_DOUBLE y) { return x>y ? y : x; } |
---|
| 50 | inline SAS_DOUBLE fmax(SAS_DOUBLE x, SAS_DOUBLE y) { return x<y ? y : x; } |
---|
[2a55a6f] | 51 | #define NEED_EXPM1 |
---|
| 52 | #define NEED_TGAMMA |
---|
[6a0d6aa] | 53 | #define NEED_ERF |
---|
[2a55a6f] | 54 | #else |
---|
| 55 | #include <tgmath.h> // C99 type-generic math, so sin(float) => sinf |
---|
| 56 | #endif |
---|
[750ffa5] | 57 | // MSVC doesn't support C99, so no need for dllexport on C99 branch |
---|
[f734e7d] | 58 | #define kernel |
---|
[750ffa5] | 59 | #define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) |
---|
[f734e7d] | 60 | # endif |
---|
| 61 | # define global |
---|
| 62 | # define local |
---|
| 63 | # define constant const |
---|
[750ffa5] | 64 | // OpenCL powr(a,b) = C99 pow(a,b), b >= 0 |
---|
| 65 | // OpenCL pown(a,b) = C99 pow(a,b), b integer |
---|
[f734e7d] | 66 | # define powr(a,b) pow(a,b) |
---|
| 67 | # define pown(a,b) pow(a,b) |
---|
| 68 | #else |
---|
[deac08c] | 69 | # if defined(USE_SINCOS) |
---|
[f734e7d] | 70 | # define SINCOS(angle,svar,cvar) svar=sincos(angle,&cvar) |
---|
| 71 | # else |
---|
[750ffa5] | 72 | # define SINCOS(angle,svar,cvar) do {const double _t_=angle; svar=sin(_t_);cvar=cos(_t_);} while (0) |
---|
[f734e7d] | 73 | # endif |
---|
| 74 | #endif |
---|
| 75 | |
---|
[98cb4d7] | 76 | #if defined(NEED_EXPM1) |
---|
[5efe850] | 77 | static SAS_DOUBLE expm1(SAS_DOUBLE x_in) { |
---|
| 78 | double x = (double)x_in; // go back to float for single precision kernels |
---|
[98cb4d7] | 79 | // Adapted from the cephes math library. |
---|
| 80 | // Copyright 1984 - 1992 by Stephen L. Moshier |
---|
| 81 | if (x != x || x == 0.0) { |
---|
| 82 | return x; // NaN and +/- 0 |
---|
| 83 | } else if (x < -0.5 || x > 0.5) { |
---|
| 84 | return exp(x) - 1.0; |
---|
| 85 | } else { |
---|
| 86 | const double xsq = x*x; |
---|
| 87 | const double p = ((( |
---|
| 88 | +1.2617719307481059087798E-4)*xsq |
---|
| 89 | +3.0299440770744196129956E-2)*xsq |
---|
| 90 | +9.9999999999999999991025E-1); |
---|
| 91 | const double q = (((( |
---|
| 92 | +3.0019850513866445504159E-6)*xsq |
---|
| 93 | +2.5244834034968410419224E-3)*xsq |
---|
| 94 | +2.2726554820815502876593E-1)*xsq |
---|
| 95 | +2.0000000000000000000897E0); |
---|
| 96 | double r = x * p; |
---|
| 97 | r = r / (q - r); |
---|
| 98 | return r+r; |
---|
| 99 | } |
---|
| 100 | } |
---|
| 101 | #endif |
---|
| 102 | |
---|
[f734e7d] | 103 | // Standard mathematical constants: |
---|
| 104 | // M_E, M_LOG2E, M_LOG10E, M_LN2, M_LN10, M_PI, M_PI_2=pi/2, M_PI_4=pi/4, |
---|
| 105 | // M_1_PI=1/pi, M_2_PI=2/pi, M_2_SQRTPI=2/sqrt(pi), SQRT2, SQRT1_2=sqrt(1/2) |
---|
| 106 | // OpenCL defines M_constant_F for float constants, and nothing if double |
---|
| 107 | // is not enabled on the card, which is why these constants may be missing |
---|
| 108 | #ifndef M_PI |
---|
| 109 | # define M_PI 3.141592653589793 |
---|
| 110 | #endif |
---|
| 111 | #ifndef M_PI_2 |
---|
| 112 | # define M_PI_2 1.570796326794897 |
---|
| 113 | #endif |
---|
| 114 | #ifndef M_PI_4 |
---|
| 115 | # define M_PI_4 0.7853981633974483 |
---|
| 116 | #endif |
---|
[e7678b2] | 117 | #ifndef M_E |
---|
| 118 | # define M_E 2.718281828459045091 |
---|
| 119 | #endif |
---|
[f734e7d] | 120 | |
---|
[deac08c] | 121 | // Non-standard function library |
---|
| 122 | // pi/180, used for converting between degrees and radians |
---|
| 123 | // 4/3 pi for computing sphere volumes |
---|
| 124 | // square and cube for computing squares and cubes |
---|
[f734e7d] | 125 | #ifndef M_PI_180 |
---|
| 126 | # define M_PI_180 0.017453292519943295 |
---|
| 127 | #endif |
---|
[deac08c] | 128 | #ifndef M_4PI_3 |
---|
| 129 | # define M_4PI_3 4.18879020478639 |
---|
| 130 | #endif |
---|
[73860b6] | 131 | //inline double square(double x) { return pow(x,2.0); } |
---|
[deac08c] | 132 | //inline double square(double x) { return pown(x,2); } |
---|
[cf85329] | 133 | inline double square(double x) { return x*x; } |
---|
[deac08c] | 134 | inline double cube(double x) { return x*x*x; } |
---|
[1e7b0db0] | 135 | inline double sas_sinx_x(double x) { return x==0 ? 1.0 : sin(x)/x; } |
---|
[f734e7d] | 136 | |
---|
| 137 | |
---|
| 138 | %(DEFINES)s |
---|
| 139 | |
---|
| 140 | %(SOURCES)s |
---|
| 141 | |
---|
| 142 | /* |
---|
| 143 | ########################################################## |
---|
| 144 | # # |
---|
| 145 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
| 146 | # !! !! # |
---|
| 147 | # !! KEEP THIS CODE CONSISTENT WITH KERNELPY.PY !! # |
---|
| 148 | # !! !! # |
---|
| 149 | # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # |
---|
| 150 | # # |
---|
| 151 | ########################################################## |
---|
| 152 | */ |
---|
| 153 | |
---|
| 154 | #ifdef IQ_KERNEL_NAME |
---|
| 155 | kernel void IQ_KERNEL_NAME( |
---|
| 156 | global const double *q, |
---|
| 157 | global double *result, |
---|
| 158 | const int Nq, |
---|
| 159 | #ifdef IQ_OPEN_LOOPS |
---|
| 160 | #ifdef USE_OPENCL |
---|
| 161 | global double *loops_g, |
---|
| 162 | #endif |
---|
| 163 | local double *loops, |
---|
| 164 | const double cutoff, |
---|
| 165 | IQ_DISPERSION_LENGTH_DECLARATIONS, |
---|
| 166 | #endif |
---|
| 167 | IQ_FIXED_PARAMETER_DECLARATIONS |
---|
| 168 | ) |
---|
| 169 | { |
---|
| 170 | #ifdef USE_OPENCL |
---|
| 171 | #ifdef IQ_OPEN_LOOPS |
---|
| 172 | // copy loops info to local memory |
---|
| 173 | event_t e = async_work_group_copy(loops, loops_g, (IQ_DISPERSION_LENGTH_SUM)*2, 0); |
---|
| 174 | wait_group_events(1, &e); |
---|
| 175 | #endif |
---|
| 176 | |
---|
| 177 | int i = get_global_id(0); |
---|
| 178 | if (i < Nq) |
---|
| 179 | #else |
---|
| 180 | #pragma omp parallel for |
---|
| 181 | for (int i=0; i < Nq; i++) |
---|
| 182 | #endif |
---|
| 183 | { |
---|
| 184 | const double qi = q[i]; |
---|
| 185 | #ifdef IQ_OPEN_LOOPS |
---|
| 186 | double ret=0.0, norm=0.0; |
---|
| 187 | IQ_OPEN_LOOPS |
---|
| 188 | //for (int radius_i=0; radius_i < Nradius; radius_i++) { |
---|
| 189 | // const double radius = loops[2*(radius_i)]; |
---|
| 190 | // const double radius_w = loops[2*(radius_i)+1]; |
---|
| 191 | |
---|
| 192 | const double weight = IQ_WEIGHT_PRODUCT; |
---|
| 193 | if (weight > cutoff) { |
---|
[750ffa5] | 194 | const double scattering = Iq(qi, IQ_PARAMETERS); |
---|
[c138211] | 195 | // allow kernels to exclude invalid regions by returning NaN |
---|
| 196 | if (!isnan(scattering)) { |
---|
[750ffa5] | 197 | ret += weight*scattering; |
---|
[f734e7d] | 198 | #ifdef VOLUME_PARAMETERS |
---|
[c4e7a5f] | 199 | norm += weight * form_volume(VOLUME_PARAMETERS); |
---|
| 200 | #else |
---|
| 201 | norm += weight; |
---|
[f734e7d] | 202 | #endif |
---|
[c138211] | 203 | } |
---|
[750ffa5] | 204 | //else { printf("exclude qx,qy,I:%%g,%%g,%%g\n",qi,scattering); } |
---|
[f734e7d] | 205 | } |
---|
| 206 | IQ_CLOSE_LOOPS |
---|
[c4e7a5f] | 207 | // norm can only be zero if volume is zero, so no scattering |
---|
| 208 | result[i] = (norm > 0. ? scale*ret/norm + background : background); |
---|
[f734e7d] | 209 | #else |
---|
| 210 | result[i] = scale*Iq(qi, IQ_PARAMETERS) + background; |
---|
| 211 | #endif |
---|
| 212 | } |
---|
| 213 | } |
---|
| 214 | #endif |
---|
| 215 | |
---|
| 216 | |
---|
| 217 | #ifdef IQXY_KERNEL_NAME |
---|
| 218 | kernel void IQXY_KERNEL_NAME( |
---|
| 219 | global const double *qx, |
---|
| 220 | global const double *qy, |
---|
| 221 | global double *result, |
---|
| 222 | const int Nq, |
---|
| 223 | #ifdef IQXY_OPEN_LOOPS |
---|
| 224 | #ifdef USE_OPENCL |
---|
| 225 | global double *loops_g, |
---|
| 226 | #endif |
---|
| 227 | local double *loops, |
---|
| 228 | const double cutoff, |
---|
| 229 | IQXY_DISPERSION_LENGTH_DECLARATIONS, |
---|
| 230 | #endif |
---|
| 231 | IQXY_FIXED_PARAMETER_DECLARATIONS |
---|
| 232 | ) |
---|
| 233 | { |
---|
| 234 | #ifdef USE_OPENCL |
---|
| 235 | #ifdef IQXY_OPEN_LOOPS |
---|
| 236 | // copy loops info to local memory |
---|
| 237 | event_t e = async_work_group_copy(loops, loops_g, (IQXY_DISPERSION_LENGTH_SUM)*2, 0); |
---|
| 238 | wait_group_events(1, &e); |
---|
| 239 | #endif |
---|
| 240 | |
---|
| 241 | int i = get_global_id(0); |
---|
| 242 | if (i < Nq) |
---|
| 243 | #else |
---|
| 244 | #pragma omp parallel for |
---|
| 245 | for (int i=0; i < Nq; i++) |
---|
| 246 | #endif |
---|
| 247 | { |
---|
| 248 | const double qxi = qx[i]; |
---|
| 249 | const double qyi = qy[i]; |
---|
[e3a9733] | 250 | #if USE_KAHAN_SUMMATION |
---|
| 251 | double accumulated_error = 0.0; |
---|
| 252 | #endif |
---|
[f734e7d] | 253 | #ifdef IQXY_OPEN_LOOPS |
---|
| 254 | double ret=0.0, norm=0.0; |
---|
| 255 | IQXY_OPEN_LOOPS |
---|
| 256 | //for (int radius_i=0; radius_i < Nradius; radius_i++) { |
---|
| 257 | // const double radius = loops[2*(radius_i)]; |
---|
| 258 | // const double radius_w = loops[2*(radius_i)+1]; |
---|
[c4e7a5f] | 259 | double weight = IQXY_WEIGHT_PRODUCT; |
---|
[f734e7d] | 260 | if (weight > cutoff) { |
---|
| 261 | |
---|
[750ffa5] | 262 | const double scattering = Iqxy(qxi, qyi, IQXY_PARAMETERS); |
---|
[9c79c32] | 263 | if (!isnan(scattering)) { // if scattering is bad, exclude it from sum |
---|
[c4e7a5f] | 264 | #if defined(IQXY_HAS_THETA) |
---|
| 265 | // Force a nominal value for the spherical correction even when |
---|
[0d6e865] | 266 | // theta is +0/180 so that there are no divide by zero problems. |
---|
| 267 | // For sin(theta) fixed at 0 and 180, we effectively multiply top and bottom |
---|
[c4e7a5f] | 268 | // by 1e-6, so the effect cancels. |
---|
[0278e3f] | 269 | const double spherical_correction = fmax(fabs(cos(M_PI_180*theta)), 1.e-6); |
---|
[c4e7a5f] | 270 | weight *= spherical_correction; |
---|
[e3a9733] | 271 | #endif |
---|
[c4e7a5f] | 272 | const double next = weight * scattering; |
---|
[e3a9733] | 273 | #if USE_KAHAN_SUMMATION |
---|
| 274 | const double y = next - accumulated_error; |
---|
| 275 | const double t = ret + y; |
---|
| 276 | accumulated_error = (t - ret) - y; |
---|
| 277 | ret = t; |
---|
[f734e7d] | 278 | #else |
---|
[e3a9733] | 279 | ret += next; |
---|
[f734e7d] | 280 | #endif |
---|
| 281 | #ifdef VOLUME_PARAMETERS |
---|
[c4e7a5f] | 282 | norm += weight*form_volume(VOLUME_PARAMETERS); |
---|
| 283 | #else |
---|
| 284 | norm += weight; |
---|
[718514b] | 285 | #endif |
---|
[9c79c32] | 286 | } |
---|
[750ffa5] | 287 | //else { printf("exclude qx,qy,I:%%g,%%g,%%g\n",qi,scattering); } |
---|
[f734e7d] | 288 | } |
---|
| 289 | IQXY_CLOSE_LOOPS |
---|
[c4e7a5f] | 290 | // norm can only be zero if volume is zero, so no scattering |
---|
| 291 | result[i] = (norm>0. ? scale*ret/norm + background : background); |
---|
[f734e7d] | 292 | #else |
---|
| 293 | result[i] = scale*Iqxy(qxi, qyi, IQXY_PARAMETERS) + background; |
---|
| 294 | #endif |
---|
| 295 | } |
---|
| 296 | } |
---|
| 297 | #endif |
---|