9339fbad
David Mayerich
implementing mie ...
|
1
2
|
#ifndef STIM_MIE_H
#define STIM_MIE_H
|
8309b07a
David Mayerich
fixed some vec3 e...
|
3
|
#include <boost/math/special_functions/bessel.hpp>
|
9339fbad
David Mayerich
implementing mie ...
|
4
5
6
|
#include "scalarwave.h"
#include "../math/bessel.h"
|
31262e83
David Mayerich
GPU implementatio...
|
7
|
#include "../cuda/cudatools/devices.h"
|
9339fbad
David Mayerich
implementing mie ...
|
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
|
#include <cmath>
namespace stim{
/// Calculate the scattering coefficients for a spherical scatterer
template<typename T>
void B_coefficients(stim::complex<T>* B, T a, T k, stim::complex<T> n, int Nl){
//temporary variables
double vm; //allocate space to store the return values for the bessel function calculation
double* j_ka = (double*) malloc( (Nl + 1) * sizeof(double) );
double* y_ka = (double*) malloc( (Nl + 1) * sizeof(double) );
double* dj_ka= (double*) malloc( (Nl + 1) * sizeof(double) );
double* dy_ka= (double*) malloc( (Nl + 1) * sizeof(double) );
stim::complex<double>* j_kna = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* y_kna = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dj_kna= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dy_kna= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
double ka = k * a; //store k*a (argument for spherical bessel and Hankel functions)
stim::complex<double> kna = k * n * a; //store k*n*a (argument for spherical bessel functions and derivatives)
stim::bessjyv_sph<double>(Nl, ka, vm, j_ka, y_ka, dj_ka, dy_ka); //calculate bessel functions and derivatives for k*a
stim::cbessjyva_sph<double>(Nl, kna, vm, j_kna, y_kna, dj_kna, dy_kna); //calculate complex bessel functions for k*n*a
stim::complex<double> h_ka, dh_ka;
stim::complex<double> numerator, denominator;
stim::complex<double> i(0, 1);
|
31262e83
David Mayerich
GPU implementatio...
|
38
|
for(int l = 0; l <= Nl; l++){
|
9339fbad
David Mayerich
implementing mie ...
|
39
40
41
42
43
44
45
46
|
h_ka.r = j_ka[l];
h_ka.i = y_ka[l];
dh_ka.r = dj_ka[l];
dh_ka.i = dy_ka[l];
numerator = j_ka[l] * dj_kna[l] * (stim::complex<double>)n - j_kna[l] * dj_ka[l];
denominator = j_kna[l] * dh_ka - h_ka * dj_kna[l] * (stim::complex<double>)n;
B[l] = (2 * l + 1) * pow(i, l) * numerator / denominator;
|
9339fbad
David Mayerich
implementing mie ...
|
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
|
}
}
template<typename T>
void A_coefficients(stim::complex<T>* A, T a, T k, stim::complex<T> n, int Nl){
//temporary variables
double vm; //allocate space to store the return values for the bessel function calculation
double* j_ka = (double*) malloc( (Nl + 1) * sizeof(double) );
double* y_ka = (double*) malloc( (Nl + 1) * sizeof(double) );
double* dj_ka= (double*) malloc( (Nl + 1) * sizeof(double) );
double* dy_ka= (double*) malloc( (Nl + 1) * sizeof(double) );
stim::complex<double>* j_kna = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* y_kna = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dj_kna= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dy_kna= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
double ka = k * a; //store k*a (argument for spherical bessel and Hankel functions)
stim::complex<double> kna = k * n * a; //store k*n*a (argument for spherical bessel functions and derivatives)
stim::bessjyv_sph<double>(Nl, ka, vm, j_ka, y_ka, dj_ka, dy_ka); //calculate bessel functions and derivatives for k*a
stim::cbessjyva_sph<double>(Nl, kna, vm, j_kna, y_kna, dj_kna, dy_kna); //calculate complex bessel functions for k*n*a
stim::complex<double> h_ka, dh_ka;
stim::complex<double> numerator, denominator;
stim::complex<double> i(0, 1);
for(size_t l = 0; l <= Nl; l++){
h_ka.r = j_ka[l];
h_ka.i = y_ka[l];
dh_ka.r = dj_ka[l];
dh_ka.i = dy_ka[l];
numerator = j_ka[l] * dh_ka - dj_ka[l] * h_ka;
denominator = j_kna[l] * dh_ka - h_ka * dj_kna[l] * (stim::complex<double>)n;
A[l] = (2 * l + 1) * pow(i, l) * numerator / denominator;
}
}
|
31262e83
David Mayerich
GPU implementatio...
|
85
|
#define LOCAL_NL 16
|
9339fbad
David Mayerich
implementing mie ...
|
86
|
template<typename T>
|
8309b07a
David Mayerich
fixed some vec3 e...
|
87
|
__global__ void cuda_scalar_mie_scatter(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T>* W, size_t nW, T a, stim::complex<T> n, stim::complex<T>* hB, T r_min, T dr, size_t N_hB, int Nl){
|
31262e83
David Mayerich
GPU implementatio...
|
88
|
extern __shared__ stim::complex<T> shared_hB[]; //declare the list of waves in shared memory
|
9339fbad
David Mayerich
implementing mie ...
|
89
90
|
size_t i = blockIdx.x * blockDim.x + threadIdx.x; //get the index into the array
|
31262e83
David Mayerich
GPU implementatio...
|
91
|
if(i >= N) return; //exit if this thread is outside the array
|
9339fbad
David Mayerich
implementing mie ...
|
92
93
94
95
96
|
stim::vec3<T> p;
(x == NULL) ? p[0] = 0 : p[0] = x[i]; // test for NULL values and set positions
(y == NULL) ? p[1] = 0 : p[1] = y[i];
(z == NULL) ? p[2] = 0 : p[2] = z[i];
|
31262e83
David Mayerich
GPU implementatio...
|
97
98
|
T r = p.len(); //calculate the distance from the sphere
if(r < a) return; //exit if the point is inside the sphere (we only calculate the internal field)
|
8309b07a
David Mayerich
fixed some vec3 e...
|
99
|
T fij = (r - r_min)/dr; //FP index into the spherical bessel LUT
|
31262e83
David Mayerich
GPU implementatio...
|
100
101
|
size_t ij = (size_t) fij; //convert to an integral index
T alpha = fij - ij; //calculate the fractional portion of the index
|
8309b07a
David Mayerich
fixed some vec3 e...
|
102
103
|
size_t n0j = ij * (Nl + 1); //start of the first entry in the LUT
size_t n1j = (ij+1) * (Nl + 1); //start of the second entry in the LUT
|
9339fbad
David Mayerich
implementing mie ...
|
104
105
|
T cos_phi;
|
31262e83
David Mayerich
GPU implementatio...
|
106
107
108
|
T Pl_2, Pl_1, Pl; //declare registers to store the previous two Legendre polynomials
stim::complex<T> hBl;
|
9339fbad
David Mayerich
implementing mie ...
|
109
110
|
stim::complex<T> Ei = 0; //create a register to store the result
int l;
|
31262e83
David Mayerich
GPU implementatio...
|
111
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
112
113
|
stim::complex<T> hlBl[LOCAL_NL+1]; //the first LOCAL_NL components are stored in registers for speed
int shared_start = threadIdx.x * (Nl - LOCAL_NL); //wrap up some operations so that they aren't done in the main loops
|
31262e83
David Mayerich
GPU implementatio...
|
114
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
115
|
#pragma unroll LOCAL_NL+1 //copy the first LOCAL_NL+1 h_l * B_l components to registers
|
31262e83
David Mayerich
GPU implementatio...
|
116
117
118
|
for(l = 0; l <= LOCAL_NL; l++)
hlBl[l] = clerp<T>( hB[n0j + l], hB[n1j + l], alpha );
|
8309b07a
David Mayerich
fixed some vec3 e...
|
119
|
for(l = LOCAL_NL+1; l <= Nl; l++) //copy any additional h_l * B_l components to shared memory
|
31262e83
David Mayerich
GPU implementatio...
|
120
121
|
shared_hB[shared_start + (l - (LOCAL_NL+1))] = clerp<T>( hB[n0j + l], hB[n1j + l], alpha );
|
8309b07a
David Mayerich
fixed some vec3 e...
|
122
|
for(size_t w = 0; w < nW; w++){ //for each plane wave
|
9339fbad
David Mayerich
implementing mie ...
|
123
|
cos_phi = p.norm().dot(W[w].kvec().norm()); //calculate the cosine of the angle between the k vector and the direction from the sphere
|
8309b07a
David Mayerich
fixed some vec3 e...
|
124
|
Pl_2 = 1; //the Legendre polynomials will be calculated recursively, initialize the first two steps of the recursive relation
|
31262e83
David Mayerich
GPU implementatio...
|
125
|
Pl_1 = cos_phi;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
126
|
Ei += W[w].E() * hlBl[0] * Pl_2; //unroll the first two orders using the initial steps of the Legendre recursive relation
|
31262e83
David Mayerich
GPU implementatio...
|
127
128
|
Ei += W[w].E() * hlBl[1] * Pl_1;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
129
|
#pragma unroll LOCAL_NL-1 //unroll the next LOCAL_NL-1 loops for speed (iterating through the components in the register file)
|
31262e83
David Mayerich
GPU implementatio...
|
130
|
for(l = 2; l <= LOCAL_NL; l++){
|
8309b07a
David Mayerich
fixed some vec3 e...
|
131
132
|
Pl = ( (2 * (l-1) + 1) * cos_phi * Pl_1 - (l-1) * Pl_2 ) / (l); //calculate the next step in the Legendre polynomial recursive relation (this is where most of the computation occurs)
Ei += W[w].E() * hlBl[l] * Pl; //calculate and sum the current field order
|
31262e83
David Mayerich
GPU implementatio...
|
133
|
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
|
9339fbad
David Mayerich
implementing mie ...
|
134
|
Pl_1 = Pl;
|
31262e83
David Mayerich
GPU implementatio...
|
135
|
}
|
9339fbad
David Mayerich
implementing mie ...
|
136
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
137
138
139
140
141
|
for(l = LOCAL_NL+1; l <= Nl; l++){ //do the same as above, except for any additional orders that are stored in shared memory (not registers)
Pl = ( (2 * (l-1) + 1) * cos_phi * Pl_1 - (l-1) * Pl_2 ) / (l); //again, this is where most computation in the kernel occurs
Ei += W[w].E() * shared_hB[shared_start + l - LOCAL_NL - 1] * Pl;
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
Pl_1 = Pl;
|
9339fbad
David Mayerich
implementing mie ...
|
142
|
}
|
9339fbad
David Mayerich
implementing mie ...
|
143
|
}
|
31262e83
David Mayerich
GPU implementatio...
|
144
|
E[i] += Ei; //copy the result to device memory
|
9339fbad
David Mayerich
implementing mie ...
|
145
146
147
|
}
template<typename T>
|
31262e83
David Mayerich
GPU implementatio...
|
148
149
150
|
void gpu_scalar_mie_scatter(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T>* W, size_t nW, T a, stim::complex<T> n, stim::complex<T>* hB, T kr_min, T dkr, size_t N_hB, size_t Nl){
size_t max_shared_mem = stim::sharedMemPerBlock();
|
8309b07a
David Mayerich
fixed some vec3 e...
|
151
|
size_t hBl_array = sizeof(stim::complex<T>) * (Nl + 1);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
152
153
|
//std::cout<<"hl*Bl array size: "<<hBl_array<<std::endl;
//std::cout<<"shared memory: "<<max_shared_mem<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
154
|
int threads = (int)((max_shared_mem / hBl_array) / 32 * 32);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
155
|
//std::cout<<"threads per block: "<<threads<<std::endl;
|
31262e83
David Mayerich
GPU implementatio...
|
156
157
158
159
160
|
dim3 blocks((unsigned)(N / threads + 1)); //calculate the optimal number of blocks
size_t shared_mem;
if(Nl <= LOCAL_NL) shared_mem = 0;
else shared_mem = threads * sizeof(stim::complex<T>) * (Nl - LOCAL_NL); //amount of shared memory to allocate
|
4252d827
David Mayerich
ivote3 fixes and ...
|
161
|
//std::cout<<"shared memory allocated: "<<shared_mem<<std::endl;
|
31262e83
David Mayerich
GPU implementatio...
|
162
|
cuda_scalar_mie_scatter<T><<< blocks, threads, shared_mem >>>(E, N, x, y, z, W, nW, a, n, hB, kr_min, dkr, N_hB, (int)Nl); //call the kernel
|
31262e83
David Mayerich
GPU implementatio...
|
163
164
165
166
167
168
169
170
171
172
173
174
175
|
}
template<typename T>
__global__ void cuda_dist(T* r, T* x, T* y, T* z, size_t N){
size_t i = blockIdx.x * blockDim.x + threadIdx.x; //get the index into the array
if(i >= N) return; //exit if this thread is outside the array
stim::vec3<T> p;
(x == NULL) ? p[0] = 0 : p[0] = x[i]; // test for NULL values and set positions
(y == NULL) ? p[1] = 0 : p[1] = y[i];
(z == NULL) ? p[2] = 0 : p[2] = z[i];
r[i] = p.len();
|
9339fbad
David Mayerich
implementing mie ...
|
176
|
}
|
9339fbad
David Mayerich
implementing mie ...
|
177
|
template<typename T>
|
4252d827
David Mayerich
ivote3 fixes and ...
|
178
179
|
void gpu_scalar_mie_scatter(stim::complex<T>* E, size_t N, T* x, T* y, T* z, std::vector<stim::scalarwave<T>> W, T a, stim::complex<T> n, T r_spacing = 0.1){
|
9339fbad
David Mayerich
implementing mie ...
|
180
181
182
|
//calculate the necessary number of orders required to represent the scattered field
T k = W[0].kmag();
|
31262e83
David Mayerich
GPU implementatio...
|
183
184
|
int Nl = (int)ceil(k*a + 4 * cbrt( k * a ) + 2);
if(Nl < LOCAL_NL) Nl = LOCAL_NL; //always do at least the minimum number of local operations (kernel optimization)
|
4252d827
David Mayerich
ivote3 fixes and ...
|
185
|
//std::cout<<"Nl: "<<Nl<<std::endl;
|
9339fbad
David Mayerich
implementing mie ...
|
186
187
188
|
//calculate the scattering coefficients for the sphere
stim::complex<T>* B = (stim::complex<T>*) malloc( sizeof(stim::complex<T>) * (Nl + 1) ); //allocate space for the scattering coefficients
|
4252d827
David Mayerich
ivote3 fixes and ...
|
189
190
|
B_coefficients(B, a, k, n, Nl);
|
9339fbad
David Mayerich
implementing mie ...
|
191
192
193
194
195
|
// PLANE WAVES
stim::scalarwave<T>* dev_W; //allocate space and copy plane waves
HANDLE_ERROR( cudaMalloc(&dev_W, sizeof(stim::scalarwave<T>) * W.size()) );
HANDLE_ERROR( cudaMemcpy(dev_W, &W[0], sizeof(stim::scalarwave<T>) * W.size(), cudaMemcpyHostToDevice) );
|
9339fbad
David Mayerich
implementing mie ...
|
196
|
// BESSEL FUNCTION LOOK-UP TABLE
|
31262e83
David Mayerich
GPU implementatio...
|
197
198
199
200
201
202
|
//calculate the distance from the sphere center
T* dev_r;
HANDLE_ERROR( cudaMalloc(&dev_r, sizeof(T) * N) );
int threads = stim::maxThreadsPerBlock();
dim3 blocks((unsigned)(N / threads + 1));
|
4252d827
David Mayerich
ivote3 fixes and ...
|
203
|
cuda_dist<T> <<< blocks, threads >>>(dev_r, x, y, z, N);
|
31262e83
David Mayerich
GPU implementatio...
|
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
|
//Find the minimum and maximum values of r
cublasStatus_t stat;
cublasHandle_t handle;
stat = cublasCreate(&handle); //create a cuBLAS handle
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS initialization failed\n");
exit(1);
}
int i_min, i_max;
stat = cublasIsamin(handle, (int)N, dev_r, 1, &i_min);
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS Error: failed to calculate minimum r value.\n");
exit(1);
}
stat = cublasIsamax(handle, (int)N, dev_r, 1, &i_max);
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS Error: failed to calculate maximum r value.\n");
exit(1);
}
|
8309b07a
David Mayerich
fixed some vec3 e...
|
227
228
|
i_min--; //cuBLAS uses 1-based indexing for Fortran compatibility
i_max--;
|
31262e83
David Mayerich
GPU implementatio...
|
229
230
231
232
|
T r_min, r_max; //allocate space to store the minimum and maximum values
HANDLE_ERROR( cudaMemcpy(&r_min, dev_r + i_min, sizeof(T), cudaMemcpyDeviceToHost) ); //copy the min and max values from the device to the CPU
HANDLE_ERROR( cudaMemcpy(&r_max, dev_r + i_max, sizeof(T), cudaMemcpyDeviceToHost) );
|
8309b07a
David Mayerich
fixed some vec3 e...
|
233
|
r_min = max(r_min, a); //if the radius of the sphere is larger than r_min, change r_min to a (the scattered field doesn't exist inside the sphere)
|
31262e83
David Mayerich
GPU implementatio...
|
234
|
|
9339fbad
David Mayerich
implementing mie ...
|
235
|
//size_t Nlut_j = (size_t)((r_max - r_min) / r_spacing + 1); //number of values in the look-up table based on the user-specified spacing along r
|
31262e83
David Mayerich
GPU implementatio...
|
236
|
size_t N_hB_lut = (size_t)((r_max - r_min) / r_spacing + 1);
|
9339fbad
David Mayerich
implementing mie ...
|
237
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
238
239
|
//T kr_min = k * r_min;
//T kr_max = k * r_max;
|
9339fbad
David Mayerich
implementing mie ...
|
240
241
242
243
244
245
246
247
|
//temporary variables
double vm; //allocate space to store the return values for the bessel function calculation
double* jv = (double*) malloc( (Nl + 1) * sizeof(double) );
double* yv = (double*) malloc( (Nl + 1) * sizeof(double) );
double* djv= (double*) malloc( (Nl + 1) * sizeof(double) );
double* dyv= (double*) malloc( (Nl + 1) * sizeof(double) );
|
31262e83
David Mayerich
GPU implementatio...
|
248
249
|
size_t hB_bytes = sizeof(stim::complex<T>) * (Nl+1) * N_hB_lut;
stim::complex<T>* hB_lut = (stim::complex<T>*) malloc(hB_bytes); //pointer to the look-up table
|
8309b07a
David Mayerich
fixed some vec3 e...
|
250
|
T dr = (r_max - r_min) / (N_hB_lut-1); //distance between values in the LUT
|
4252d827
David Mayerich
ivote3 fixes and ...
|
251
|
//std::cout<<"LUT jl bytes: "<<hB_bytes<<std::endl;
|
31262e83
David Mayerich
GPU implementatio...
|
252
|
stim::complex<T> hl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
253
254
|
for(size_t ri = 0; ri < N_hB_lut; ri++){ //for each value in the LUT
stim::bessjyv_sph<double>(Nl, k * (r_min + ri * dr), vm, jv, yv, djv, dyv); //compute the list of spherical bessel functions from [0 Nl]
|
9339fbad
David Mayerich
implementing mie ...
|
255
|
for(size_t l = 0; l <= Nl; l++){ //for each order
|
31262e83
David Mayerich
GPU implementatio...
|
256
257
258
|
hl.r = (T)jv[l];
hl.i = (T)yv[l];
|
8309b07a
David Mayerich
fixed some vec3 e...
|
259
260
|
hB_lut[ri * (Nl + 1) + l] = hl * B[l]; //store the bessel function result
//std::cout<<hB_lut[ri * (Nl + 1) + l]<<std::endl;
|
9339fbad
David Mayerich
implementing mie ...
|
261
262
|
}
}
|
4252d827
David Mayerich
ivote3 fixes and ...
|
263
264
265
|
//T* real_lut = (T*) malloc(hB_bytes/2);
//stim::real(real_lut, hB_lut, N_hB_lut);
//stim::cpu2image<T>(real_lut, "hankel_B.bmp", Nl+1, N_hB_lut, stim::cmBrewer);
|
9339fbad
David Mayerich
implementing mie ...
|
266
267
|
//Allocate device memory and copy everything to the GPU
|
31262e83
David Mayerich
GPU implementatio...
|
268
269
270
|
stim::complex<T>* dev_hB_lut;
HANDLE_ERROR( cudaMalloc(&dev_hB_lut, hB_bytes) );
HANDLE_ERROR( cudaMemcpy(dev_hB_lut, hB_lut, hB_bytes, cudaMemcpyHostToDevice) );
|
9339fbad
David Mayerich
implementing mie ...
|
271
|
|
4252d827
David Mayerich
ivote3 fixes and ...
|
272
|
gpu_scalar_mie_scatter<T>(E, N, x, y, z, dev_W, W.size(), a, n, dev_hB_lut, r_min, dr, N_hB_lut, Nl);
|
9339fbad
David Mayerich
implementing mie ...
|
273
|
|
4252d827
David Mayerich
ivote3 fixes and ...
|
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
|
cudaMemcpy(E, E, N * sizeof(stim::complex<T>), cudaMemcpyDeviceToHost); //copy the field from device memory
}
/// Calculate the scalar Mie solution for the scattered field produced by a single plane wave
/// @param E is a pointer to the destination field values
/// @param N is the number of points used to calculate the field
/// @param x is an array of x coordinates for each point, specified relative to the sphere (x = NULL assumes all zeros)
/// @param y is an array of y coordinates for each point, specified relative to the sphere (y = NULL assumes all zeros)
/// @param z is an array of z coordinates for each point, specified relative to the sphere (z = NULL assumes all zeros)
/// @param W is an array of planewaves that will be scattered
/// @param a is the radius of the sphere
/// @param n is the complex refractive index of the sphere
template<typename T>
void cpu_scalar_mie_scatter(stim::complex<T>* E, size_t N, T* x, T* y, T* z, std::vector<stim::scalarwave<T>> W, T a, stim::complex<T> n, T r_spacing = 0.1){
#ifdef CUDA_FOUND
stim::complex<T>* dev_E; //allocate space for the field
cudaMalloc(&dev_E, N * sizeof(stim::complex<T>));
cudaMemcpy(dev_E, E, N * sizeof(stim::complex<T>), cudaMemcpyHostToDevice);
//cudaMemset(dev_F, 0, N * sizeof(stim::complex<T>)); //set the field to zero (necessary because a sum is used)
// COORDINATES
T* dev_x = NULL; //allocate space and copy the X coordinate (if specified)
if(x != NULL){
HANDLE_ERROR(cudaMalloc(&dev_x, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_x, x, N * sizeof(T), cudaMemcpyHostToDevice));
}
T* dev_y = NULL; //allocate space and copy the Y coordinate (if specified)
if(y != NULL){
HANDLE_ERROR(cudaMalloc(&dev_y, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_y, y, N * sizeof(T), cudaMemcpyHostToDevice));
}
T* dev_z = NULL; //allocate space and copy the Z coordinate (if specified)
if(z != NULL){
HANDLE_ERROR(cudaMalloc(&dev_z, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_z, z, N * sizeof(T), cudaMemcpyHostToDevice));
}
gpu_scalar_mie_scatter(dev_E, N, dev_x, dev_y, dev_z, W, a, n, r_spacing);
|
9339fbad
David Mayerich
implementing mie ...
|
314
315
316
317
318
319
320
|
if(x != NULL) cudaFree(dev_x); //free everything
if(y != NULL) cudaFree(dev_y);
if(z != NULL) cudaFree(dev_z);
cudaFree(dev_E);
#else
|
4252d827
David Mayerich
ivote3 fixes and ...
|
321
322
323
324
325
326
327
328
329
330
|
//calculate the necessary number of orders required to represent the scattered field
T k = W[0].kmag();
int Nl = (int)ceil(k*a + 4 * cbrt( k * a ) + 2);
if(Nl < LOCAL_NL) Nl = LOCAL_NL; //always do at least the minimum number of local operations (kernel optimization)
//std::cout<<"Nl: "<<Nl<<std::endl;
//calculate the scattering coefficients for the sphere
stim::complex<T>* B = (stim::complex<T>*) malloc( sizeof(stim::complex<T>) * (Nl + 1) ); //allocate space for the scattering coefficients
B_coefficients(B, a, k, n, Nl);
|
9339fbad
David Mayerich
implementing mie ...
|
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
|
//allocate space to store the bessel function call results
double vm;
double* j_kr = (double*) malloc( (Nl + 1) * sizeof(double) );
double* y_kr = (double*) malloc( (Nl + 1) * sizeof(double) );
double* dj_kr= (double*) malloc( (Nl + 1) * sizeof(double) );
double* dy_kr= (double*) malloc( (Nl + 1) * sizeof(double) );
T* P = (T*) malloc( (Nl + 1) * sizeof(T) );
T r, kr, cos_phi;
stim::complex<T> h;
for(size_t i = 0; i < N; i++){
stim::vec3<T> p; //declare a 3D point
(x == NULL) ? p[0] = 0 : p[0] = x[i]; // test for NULL values and set positions
(y == NULL) ? p[1] = 0 : p[1] = y[i];
(z == NULL) ? p[2] = 0 : p[2] = z[i];
r = p.len();
if(r >= a){
for(size_t w = 0; w < W.size(); w++){
kr = p.len() * W[w].kmag(); //calculate k*r
stim::bessjyv_sph<double>(Nl, kr, vm, j_kr, y_kr, dj_kr, dy_kr);
cos_phi = p.norm().dot(W[w].kvec().norm()); //calculate the cosine of the angle from the propagating direction
stim::legendre<T>(Nl, cos_phi, P);
for(size_t l = 0; l <= Nl; l++){
h.r = j_kr[l];
h.i = y_kr[l];
E[i] += W[w].E() * B[l] * h * P[l];
}
}
}
}
#endif
}
template<typename T>
|
8309b07a
David Mayerich
fixed some vec3 e...
|
369
|
void cpu_scalar_mie_scatter(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T> w, T a, stim::complex<T> n, T r_spacing = 0.1){
|
9339fbad
David Mayerich
implementing mie ...
|
370
|
std::vector< stim::scalarwave<T> > W(1, w);
|
8309b07a
David Mayerich
fixed some vec3 e...
|
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
|
cpu_scalar_mie_scatter(E, N, x, y, z, W, a, n, r_spacing);
}
template<typename T>
__global__ void cuda_scalar_mie_internal(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T>* W, size_t nW, T a, stim::complex<T> n, stim::complex<T>* jA, T r_min, T dr, size_t N_jA, int Nl){
extern __shared__ stim::complex<T> shared_jA[]; //declare the list of waves in shared memory
size_t i = blockIdx.x * blockDim.x + threadIdx.x; //get the index into the array
if(i >= N) return; //exit if this thread is outside the array
stim::vec3<T> p;
(x == NULL) ? p[0] = 0 : p[0] = x[i]; // test for NULL values and set positions
(y == NULL) ? p[1] = 0 : p[1] = y[i];
(z == NULL) ? p[2] = 0 : p[2] = z[i];
T r = p.len(); //calculate the distance from the sphere
|
963d0676
David Mayerich
bug fixes related...
|
386
|
if(r >= a) return; //exit if the point is inside the sphere (we only calculate the internal field)
|
8309b07a
David Mayerich
fixed some vec3 e...
|
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
|
T fij = (r - r_min)/dr; //FP index into the spherical bessel LUT
size_t ij = (size_t) fij; //convert to an integral index
T alpha = fij - ij; //calculate the fractional portion of the index
size_t n0j = ij * (Nl + 1); //start of the first entry in the LUT
size_t n1j = (ij+1) * (Nl + 1); //start of the second entry in the LUT
T cos_phi;
T Pl_2, Pl_1, Pl; //declare registers to store the previous two Legendre polynomials
stim::complex<T> jAl;
stim::complex<T> Ei = 0; //create a register to store the result
int l;
stim::complex<T> jlAl[LOCAL_NL+1]; //the first LOCAL_NL components are stored in registers for speed
int shared_start = threadIdx.x * (Nl - LOCAL_NL); //wrap up some operations so that they aren't done in the main loops
#pragma unroll LOCAL_NL+1 //copy the first LOCAL_NL+1 h_l * B_l components to registers
for(l = 0; l <= LOCAL_NL; l++)
jlAl[l] = clerp<T>( jA[n0j + l], jA[n1j + l], alpha );
for(l = LOCAL_NL+1; l <= Nl; l++) //copy any additional h_l * B_l components to shared memory
shared_jA[shared_start + (l - (LOCAL_NL+1))] = clerp<T>( jA[n0j + l], jA[n1j + l], alpha );
for(size_t w = 0; w < nW; w++){ //for each plane wave
if(r == 0) cos_phi = 0;
else
cos_phi = p.norm().dot(W[w].kvec().norm()); //calculate the cosine of the angle between the k vector and the direction from the sphere
Pl_2 = 1; //the Legendre polynomials will be calculated recursively, initialize the first two steps of the recursive relation
Pl_1 = cos_phi;
Ei += W[w].E() * jlAl[0] * Pl_2; //unroll the first two orders using the initial steps of the Legendre recursive relation
Ei += W[w].E() * jlAl[1] * Pl_1;
#pragma unroll LOCAL_NL-1 //unroll the next LOCAL_NL-1 loops for speed (iterating through the components in the register file)
for(l = 2; l <= LOCAL_NL; l++){
Pl = ( (2 * (l-1) + 1) * cos_phi * Pl_1 - (l-1) * Pl_2 ) / (l); //calculate the next step in the Legendre polynomial recursive relation (this is where most of the computation occurs)
Ei += W[w].E() * jlAl[l] * Pl; //calculate and sum the current field order
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
Pl_1 = Pl;
}
for(l = LOCAL_NL+1; l <= Nl; l++){ //do the same as above, except for any additional orders that are stored in shared memory (not registers)
Pl = ( (2 * (l-1) + 1) * cos_phi * Pl_1 - (l-1) * Pl_2 ) / (l); //again, this is where most computation in the kernel occurs
Ei += W[w].E() * shared_jA[shared_start + l - LOCAL_NL - 1] * Pl;
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
Pl_1 = Pl;
}
}
E[i] = Ei; //copy the result to device memory
}
template<typename T>
void gpu_scalar_mie_internal(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T>* W, size_t nW, T a, stim::complex<T> n, stim::complex<T>* jA, T r_min, T dr, size_t N_jA, size_t Nl){
size_t max_shared_mem = stim::sharedMemPerBlock();
size_t hBl_array = sizeof(stim::complex<T>) * (Nl + 1);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
442
443
|
//std::cout<<"hl*Bl array size: "<<hBl_array<<std::endl;
//std::cout<<"shared memory: "<<max_shared_mem<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
444
|
int threads = (int)((max_shared_mem / hBl_array) / 32 * 32);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
445
|
//std::cout<<"threads per block: "<<threads<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
446
447
448
449
450
|
dim3 blocks((unsigned)(N / threads + 1)); //calculate the optimal number of blocks
size_t shared_mem;
if(Nl <= LOCAL_NL) shared_mem = 0;
else shared_mem = threads * sizeof(stim::complex<T>) * (Nl - LOCAL_NL); //amount of shared memory to allocate
|
4252d827
David Mayerich
ivote3 fixes and ...
|
451
|
//std::cout<<"shared memory allocated: "<<shared_mem<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
452
|
cuda_scalar_mie_internal<T><<< blocks, threads, shared_mem >>>(E, N, x, y, z, W, nW, a, n, jA, r_min, dr, N_jA, (int)Nl); //call the kernel
|
9339fbad
David Mayerich
implementing mie ...
|
453
454
455
456
457
458
459
460
461
462
463
464
465
|
}
/// Calculate the scalar Mie solution for the internal field produced by a single plane wave scattered by a sphere
/// @param E is a pointer to the destination field values
/// @param N is the number of points used to calculate the field
/// @param x is an array of x coordinates for each point, specified relative to the sphere (x = NULL assumes all zeros)
/// @param y is an array of y coordinates for each point, specified relative to the sphere (y = NULL assumes all zeros)
/// @param z is an array of z coordinates for each point, specified relative to the sphere (z = NULL assumes all zeros)
/// @param w is a planewave that will be scattered
/// @param a is the radius of the sphere
/// @param n is the complex refractive index of the sphere
template<typename T>
|
8309b07a
David Mayerich
fixed some vec3 e...
|
466
467
|
void cpu_scalar_mie_internal(stim::complex<T>* E, size_t N, T* x, T* y, T* z, std::vector< stim::scalarwave<T> > W, T a, stim::complex<T> n, T r_spacing = 0.1){
//calculate the necessary number of orders required to represent the scattered field
|
9339fbad
David Mayerich
implementing mie ...
|
468
469
|
T k = W[0].kmag();
|
8309b07a
David Mayerich
fixed some vec3 e...
|
470
471
|
int Nl = (int)ceil(k*a + 4 * cbrt( k * a ) + 2);
if(Nl < LOCAL_NL) Nl = LOCAL_NL; //always do at least the minimum number of local operations (kernel optimization)
|
4252d827
David Mayerich
ivote3 fixes and ...
|
472
|
//std::cout<<"Nl: "<<Nl<<std::endl;
|
9339fbad
David Mayerich
implementing mie ...
|
473
474
475
476
477
|
//calculate the scattering coefficients for the sphere
stim::complex<T>* A = (stim::complex<T>*) malloc( sizeof(stim::complex<T>) * (Nl + 1) ); //allocate space for the scattering coefficients
A_coefficients(A, a, k, n, Nl);
|
8309b07a
David Mayerich
fixed some vec3 e...
|
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
|
#ifdef CUDA_FOUND
stim::complex<T>* dev_E; //allocate space for the field
cudaMalloc(&dev_E, N * sizeof(stim::complex<T>));
cudaMemcpy(dev_E, E, N * sizeof(stim::complex<T>), cudaMemcpyHostToDevice);
//cudaMemset(dev_F, 0, N * sizeof(stim::complex<T>)); //set the field to zero (necessary because a sum is used)
// COORDINATES
T* dev_x = NULL; //allocate space and copy the X coordinate (if specified)
if(x != NULL){
HANDLE_ERROR(cudaMalloc(&dev_x, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_x, x, N * sizeof(T), cudaMemcpyHostToDevice));
}
T* dev_y = NULL; //allocate space and copy the Y coordinate (if specified)
if(y != NULL){
HANDLE_ERROR(cudaMalloc(&dev_y, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_y, y, N * sizeof(T), cudaMemcpyHostToDevice));
}
T* dev_z = NULL; //allocate space and copy the Z coordinate (if specified)
if(z != NULL){
HANDLE_ERROR(cudaMalloc(&dev_z, N * sizeof(T)));
HANDLE_ERROR(cudaMemcpy(dev_z, z, N * sizeof(T), cudaMemcpyHostToDevice));
}
// PLANE WAVES
stim::scalarwave<T>* dev_W; //allocate space and copy plane waves
HANDLE_ERROR( cudaMalloc(&dev_W, sizeof(stim::scalarwave<T>) * W.size()) );
HANDLE_ERROR( cudaMemcpy(dev_W, &W[0], sizeof(stim::scalarwave<T>) * W.size(), cudaMemcpyHostToDevice) );
// BESSEL FUNCTION LOOK-UP TABLE
//calculate the distance from the sphere center
T* dev_r;
HANDLE_ERROR( cudaMalloc(&dev_r, sizeof(T) * N) );
int threads = stim::maxThreadsPerBlock();
dim3 blocks((unsigned)(N / threads + 1));
cuda_dist<T> <<< blocks, threads >>>(dev_r, dev_x, dev_y, dev_z, N);
//Find the minimum and maximum values of r
cublasStatus_t stat;
cublasHandle_t handle;
stat = cublasCreate(&handle); //create a cuBLAS handle
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS initialization failed\n");
exit(1);
}
int i_min, i_max;
stat = cublasIsamin(handle, (int)N, dev_r, 1, &i_min);
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS Error: failed to calculate minimum r value.\n");
exit(1);
}
stat = cublasIsamax(handle, (int)N, dev_r, 1, &i_max);
if (stat != CUBLAS_STATUS_SUCCESS){ //test for failure
printf ("CUBLAS Error: failed to calculate maximum r value.\n");
exit(1);
}
i_min--; //cuBLAS uses 1-based indexing for Fortran compatibility
i_max--;
T r_min, r_max; //allocate space to store the minimum and maximum values
HANDLE_ERROR( cudaMemcpy(&r_min, dev_r + i_min, sizeof(T), cudaMemcpyDeviceToHost) ); //copy the min and max values from the device to the CPU
HANDLE_ERROR( cudaMemcpy(&r_max, dev_r + i_max, sizeof(T), cudaMemcpyDeviceToHost) );
r_max = min(r_max, a); //the internal field doesn't exist outside of the sphere
size_t N_jA_lut = (size_t)((r_max - r_min) / r_spacing + 1);
//temporary variables
double vm; //allocate space to store the return values for the bessel function calculation
stim::complex<double>* jv = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* yv = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* djv= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dyv= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
size_t jA_bytes = sizeof(stim::complex<T>) * (Nl+1) * N_jA_lut;
stim::complex<T>* jA_lut = (stim::complex<T>*) malloc(jA_bytes); //pointer to the look-up table
T dr = (r_max - r_min) / (N_jA_lut-1); //distance between values in the LUT
|
4252d827
David Mayerich
ivote3 fixes and ...
|
557
|
//std::cout<<"LUT jl bytes: "<<jA_bytes<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
|
stim::complex<T> hl;
stim::complex<double> nd = (stim::complex<double>)n;
for(size_t ri = 0; ri < N_jA_lut; ri++){ //for each value in the LUT
stim::cbessjyva_sph<double>(Nl, nd * k * (r_min + ri * dr), vm, jv, yv, djv, dyv); //compute the list of spherical bessel functions from [0 Nl]
for(size_t l = 0; l <= Nl; l++){ //for each order
jA_lut[ri * (Nl + 1) + l] = (stim::complex<T>)(jv[l] * (stim::complex<double>)A[l]); //store the bessel function result
}
}
//Allocate device memory and copy everything to the GPU
stim::complex<T>* dev_jA_lut;
HANDLE_ERROR( cudaMalloc(&dev_jA_lut, jA_bytes) );
HANDLE_ERROR( cudaMemcpy(dev_jA_lut, jA_lut, jA_bytes, cudaMemcpyHostToDevice) );
gpu_scalar_mie_internal<T>(dev_E, N, dev_x, dev_y, dev_z, dev_W, W.size(), a, n, dev_jA_lut, r_min, dr, N_jA_lut, Nl);
cudaMemcpy(E, dev_E, N * sizeof(stim::complex<T>), cudaMemcpyDeviceToHost); //copy the field from device memory
if(x != NULL) cudaFree(dev_x); //free everything
if(y != NULL) cudaFree(dev_y);
if(z != NULL) cudaFree(dev_z);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
579
580
581
582
|
HANDLE_ERROR( cudaFree(dev_jA_lut) );
HANDLE_ERROR( cudaFree(dev_E) );
HANDLE_ERROR( cudaFree(dev_W) );
HANDLE_ERROR( cudaFree(dev_r) );
|
8309b07a
David Mayerich
fixed some vec3 e...
|
583
584
585
|
cudaFree(dev_E);
#else
|
9339fbad
David Mayerich
implementing mie ...
|
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
|
//allocate space to store the bessel function call results
double vm;
stim::complex<double>* j_knr = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* y_knr = (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dj_knr= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
stim::complex<double>* dy_knr= (stim::complex<double>*) malloc( (Nl + 1) * sizeof(stim::complex<double>) );
T* P = (T*) malloc( (Nl + 1) * sizeof(T) );
T r, cos_phi;
stim::complex<double> knr;
stim::complex<T> h;
for(size_t i = 0; i < N; i++){
stim::vec3<T> p; //declare a 3D point
(x == NULL) ? p[0] = 0 : p[0] = x[i]; // test for NULL values and set positions
(y == NULL) ? p[1] = 0 : p[1] = y[i];
(z == NULL) ? p[2] = 0 : p[2] = z[i];
r = p.len();
if(r < a){
E[i] = 0;
for(size_t w = 0; w < W.size(); w++){
knr = (stim::complex<double>)n * p.len() * W[w].kmag(); //calculate k*n*r
stim::cbessjyva_sph<double>(Nl, knr, vm, j_knr, y_knr, dj_knr, dy_knr);
if(r == 0)
cos_phi = 0;
else
cos_phi = p.norm().dot(W[w].kvec().norm()); //calculate the cosine of the angle from the propagating direction
stim::legendre<T>(Nl, cos_phi, P);
for(size_t l = 0; l <= Nl; l++){
E[i] += W[w].E() * A[l] * (stim::complex<T>)j_knr[l] * P[l];
}
}
}
}
|
8309b07a
David Mayerich
fixed some vec3 e...
|
623
|
#endif
|
9339fbad
David Mayerich
implementing mie ...
|
624
625
626
|
}
template<typename T>
|
8309b07a
David Mayerich
fixed some vec3 e...
|
627
|
void cpu_scalar_mie_internal(stim::complex<T>* E, size_t N, T* x, T* y, T* z, stim::scalarwave<T> w, T a, stim::complex<T> n, T r_spacing = 0.1){
|
9339fbad
David Mayerich
implementing mie ...
|
628
|
std::vector< stim::scalarwave<T> > W(1, w);
|
8309b07a
David Mayerich
fixed some vec3 e...
|
629
|
cpu_scalar_mie_internal(E, N, x, y, z, W, a, n, r_spacing);
|
9339fbad
David Mayerich
implementing mie ...
|
630
631
|
}
|
963d0676
David Mayerich
bug fixes related...
|
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
|
/// Class stim::scalarmie represents a scalar Mie scattering model that can be used to calculate the fields produced by a scattering sphere.
template<typename T>
class scalarmie
{
private:
T radius; //radius of the scattering sphere
stim::complex<T> n; //refractive index of the scattering sphere
public:
scalarmie(T r, stim::complex<T> ri){
radius = r;
n = ri;
}
|
4252d827
David Mayerich
ivote3 fixes and ...
|
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
|
void sum_scat(stim::scalarfield<T>& E, T* X, T* Y, T* Z, stim::scalarbeam<T> b, int samples = 1000){
std::vector< stim::scalarwave<float> > wave_array = b.mc(samples); //decompose the beam into an array of plane waves
stim::cpu_scalar_mie_scatter<float>(E.ptr(), E.size(), X, Y, Z, wave_array, radius, n, E.spacing());
}
void sum_intern(stim::scalarfield<T>& E, T* X, T* Y, T* Z, stim::scalarbeam<T> b, int samples = 1000){
std::vector< stim::scalarwave<float> > wave_array = b.mc(samples); //decompose the beam into an array of plane waves
stim::cpu_scalar_mie_internal<float>(E.ptr(), E.size(), X, Y, Z, wave_array, radius, n, E.spacing());
}
void eval(stim::scalarfield<T>& E, T* X, T* Y, T* Z, stim::scalarbeam<T> b, int order = 500, int samples = 1000){
b.eval(E, X, Y, Z, order); //evaluate the incident field using a plane wave expansion
std::vector< stim::scalarwave<float> > wave_array = b.mc(samples); //decompose the beam into an array of plane waves
sum_scat(E, X, Y, Z, b, samples);
sum_intern(E, X, Y, Z, b, samples);
}
|
963d0676
David Mayerich
bug fixes related...
|
665
666
|
void eval(stim::scalarfield<T>& E, stim::scalarbeam<T> b, int order = 500, int samples = 1000){
|
4252d827
David Mayerich
ivote3 fixes and ...
|
667
|
/*size_t array_size = E.grid_bytes(); //calculate the number of bytes in the scalar grid
|
963d0676
David Mayerich
bug fixes related...
|
668
669
670
671
|
float* X = (float*) malloc( array_size ); //allocate space for the coordinate meshes
float* Y = (float*) malloc( array_size );
float* Z = (float*) malloc( array_size );
E.meshgrid(X, Y, Z, stim::CPUmem); //calculate the coordinate meshes
|
4252d827
David Mayerich
ivote3 fixes and ...
|
672
673
674
|
*/
E.meshgrid();
b.eval(E, order);
|
963d0676
David Mayerich
bug fixes related...
|
675
676
|
std::vector< stim::scalarwave<float> > wave_array = b.mc(samples); //decompose the beam into an array of plane waves
|
4252d827
David Mayerich
ivote3 fixes and ...
|
677
678
679
680
681
682
683
684
685
|
if(E.gpu()){
stim::gpu_scalar_mie_scatter<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, E.spacing());
}
else{
stim::cpu_scalar_mie_scatter<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, E.spacing());
stim::cpu_scalar_mie_internal<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, E.spacing());
}
//eval(E, X, Y, Z, b, order, samples); //evaluate the field
|
963d0676
David Mayerich
bug fixes related...
|
686
687
688
689
690
|
}
}; //end stim::scalarmie
} //end namespace stim
|
9339fbad
David Mayerich
implementing mie ...
|
691
692
|
#endif
|