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
|
#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
|
dd5aab2f
David Mayerich
fixed errors in s...
|
19
20
21
22
|
double* j_ka = (double*) malloc( (Nl + 2) * sizeof(double) );
double* y_ka = (double*) malloc( (Nl + 2) * sizeof(double) );
double* dj_ka= (double*) malloc( (Nl + 2) * sizeof(double) );
double* dy_ka= (double*) malloc( (Nl + 2) * sizeof(double) );
|
9339fbad
David Mayerich
implementing mie ...
|
23
|
|
dd5aab2f
David Mayerich
fixed errors in s...
|
24
25
26
27
|
stim::complex<double>* j_kna = (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* y_kna = (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* dj_kna= (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* dy_kna= (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
|
9339fbad
David Mayerich
implementing mie ...
|
28
29
30
31
32
33
34
35
36
37
|
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
|
}
|
dd5aab2f
David Mayerich
fixed errors in s...
|
48
49
50
|
//free memory
free(j_ka); free(y_ka); free(dj_ka); free(dy_ka); free(j_kna); free(y_kna); free(dj_kna); free(dy_kna);
|
9339fbad
David Mayerich
implementing mie ...
|
51
52
53
54
55
56
|
}
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
|
dd5aab2f
David Mayerich
fixed errors in s...
|
57
58
59
60
|
double* j_ka = (double*) malloc( (Nl + 2) * sizeof(double) );
double* y_ka = (double*) malloc( (Nl + 2) * sizeof(double) );
double* dj_ka= (double*) malloc( (Nl + 2) * sizeof(double) );
double* dy_ka= (double*) malloc( (Nl + 2) * sizeof(double) );
|
9339fbad
David Mayerich
implementing mie ...
|
61
|
|
dd5aab2f
David Mayerich
fixed errors in s...
|
62
63
64
65
|
stim::complex<double>* j_kna = (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* y_kna = (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* dj_kna= (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
stim::complex<double>* dy_kna= (stim::complex<double>*) malloc( (Nl + 2) * sizeof(stim::complex<double>) );
|
9339fbad
David Mayerich
implementing mie ...
|
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
|
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;
}
|
dd5aab2f
David Mayerich
fixed errors in s...
|
86
87
|
//free memory
free(j_ka); free(y_ka); free(dj_ka); free(dy_ka); free(j_kna); free(y_kna); free(dj_kna); free(dy_kna);
|
9339fbad
David Mayerich
implementing mie ...
|
88
89
|
}
|
31262e83
David Mayerich
GPU implementatio...
|
90
|
#define LOCAL_NL 16
|
6a53ac0c
David Mayerich
updated optics fi...
|
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
|
/// CUDA kernel for calculating the Mie scattering solution given a set of points (x, y, z), a list of plane waves, and a look-up table for Bl*hl
/// @param E (GPU) is the N x N destination scalar field
/// @param N is the number of sample points to evaluate
/// @param x (GPU) is the grid of X coordinates for each point in E
/// @param y (GPU) is the grid of Y coordinates for each point in E
/// @param z (GPU) is the grid of Z coordinates for each point in E
/// @param W (GPU) is an array of coherent scalar plane waves incident on the Mie scatterer
/// @param nW is the number of plane waves to evaluate (sum)
/// @param a is the radius of the Mie scatterer
/// @param n is the complex refractive index of the Mie scatterer
/// @param c is the position of the sphere in (x, y, z)
/// @param hB (GPU) is a look-up table of Hankel functions (equally spaced in distance from the sphere) pre-multiplied with scattering coefficients
/// @param kr_min is the minimum kr value in the hB look-up table (corresponding to the closest point to the sphere)
/// @param dkr is the spacing (in kr) between samples of the hB look-up table
/// @param N_hB is the number of samples in hB
/// @param Nl is the order of the calculation (number of Hankel function orders)
|
9339fbad
David Mayerich
implementing mie ...
|
108
|
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
109
|
__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::vec3<T> c, stim::complex<T>* hB, T r_min, T dr, size_t N_hB, int Nl){
|
6a53ac0c
David Mayerich
updated optics fi...
|
110
|
extern __shared__ stim::complex<T> shared_hB[]; //declare the list of waves in shared memory
|
9339fbad
David Mayerich
implementing mie ...
|
111
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
112
|
size_t i = blockIdx.x * blockDim.x + threadIdx.x; //get the index into the sample array (sample point associated with this thread)
|
31262e83
David Mayerich
GPU implementatio...
|
113
|
if(i >= N) return; //exit if this thread is outside the array
|
9339fbad
David Mayerich
implementing mie ...
|
114
115
116
117
|
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];
|
2a10ecf4
David Mayerich
updated files and...
|
118
|
p = p - c;
|
31262e83
David Mayerich
GPU implementatio...
|
119
120
|
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)
|
6a53ac0c
David Mayerich
updated optics fi...
|
121
|
T fij = (r - r_min)/dr; //FP index into the spherical bessel LUT
|
31262e83
David Mayerich
GPU implementatio...
|
122
123
|
size_t ij = (size_t) fij; //convert to an integral index
T alpha = fij - ij; //calculate the fractional portion of the index
|
6a53ac0c
David Mayerich
updated optics fi...
|
124
125
|
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 ...
|
126
127
|
T cos_phi;
|
6a53ac0c
David Mayerich
updated optics fi...
|
128
|
T Pl_2, Pl_1, Pl; //declare registers to store the previous two Legendre polynomials
|
31262e83
David Mayerich
GPU implementatio...
|
129
130
|
stim::complex<T> hBl;
|
9339fbad
David Mayerich
implementing mie ...
|
131
132
|
stim::complex<T> Ei = 0; //create a register to store the result
int l;
|
31262e83
David Mayerich
GPU implementatio...
|
133
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
134
135
|
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...
|
136
|
|
ca41e27d
David Mayerich
Linux edits for B...
|
137
138
|
//unroll LOCAL_NL + 1
#pragma unroll 17 //copy the first LOCAL_NL+1 h_l * B_l components to registers
|
31262e83
David Mayerich
GPU implementatio...
|
139
140
141
|
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...
|
142
|
for(l = LOCAL_NL+1; l <= Nl; l++) //copy any additional h_l * B_l components to shared memory
|
31262e83
David Mayerich
GPU implementatio...
|
143
144
|
shared_hB[shared_start + (l - (LOCAL_NL+1))] = clerp<T>( hB[n0j + l], hB[n1j + l], alpha );
|
2a10ecf4
David Mayerich
updated files and...
|
145
|
complex<T> e, Ew;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
146
|
for(size_t w = 0; w < nW; w++){ //for each plane wave
|
9339fbad
David Mayerich
implementing mie ...
|
147
|
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...
|
148
|
Pl_2 = 1; //the Legendre polynomials will be calculated recursively, initialize the first two steps of the recursive relation
|
31262e83
David Mayerich
GPU implementatio...
|
149
|
Pl_1 = cos_phi;
|
2a10ecf4
David Mayerich
updated files and...
|
150
151
152
153
|
e = exp(complex<T>(0, W[w].kvec().dot(c)));
Ew = W[w].E() * e;
Ei += Ew * hlBl[0] * Pl_2; //unroll the first two orders using the initial steps of the Legendre recursive relation
Ei += Ew * hlBl[1] * Pl_1;
|
31262e83
David Mayerich
GPU implementatio...
|
154
|
|
ca41e27d
David Mayerich
Linux edits for B...
|
155
156
|
//LOCAL_NL - 1
#pragma unroll 15 //unroll the next LOCAL_NL-1 loops for speed (iterating through the components in the register file)
|
31262e83
David Mayerich
GPU implementatio...
|
157
|
for(l = 2; l <= LOCAL_NL; l++){
|
8309b07a
David Mayerich
fixed some vec3 e...
|
158
|
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)
|
2a10ecf4
David Mayerich
updated files and...
|
159
|
Ei += Ew * hlBl[l] * Pl; //calculate and sum the current field order
|
31262e83
David Mayerich
GPU implementatio...
|
160
|
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
|
9339fbad
David Mayerich
implementing mie ...
|
161
|
Pl_1 = Pl;
|
31262e83
David Mayerich
GPU implementatio...
|
162
|
}
|
9339fbad
David Mayerich
implementing mie ...
|
163
|
|
8309b07a
David Mayerich
fixed some vec3 e...
|
164
165
|
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
|
2a10ecf4
David Mayerich
updated files and...
|
166
|
Ei += Ew * shared_hB[shared_start + l - LOCAL_NL - 1] * Pl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
167
168
|
Pl_2 = Pl_1; //shift Pl_1 -> Pl_2 and Pl -> Pl_1
Pl_1 = Pl;
|
9339fbad
David Mayerich
implementing mie ...
|
169
|
}
|
9339fbad
David Mayerich
implementing mie ...
|
170
|
}
|
31262e83
David Mayerich
GPU implementatio...
|
171
|
E[i] += Ei; //copy the result to device memory
|
9339fbad
David Mayerich
implementing mie ...
|
172
173
|
}
|
6a53ac0c
David Mayerich
updated optics fi...
|
174
|
///Calculate the scalar Mie scattered field on the GPU when a list of GPU-based pre-multiplied Hankel functions are available
|
2a10ecf4
David Mayerich
updated files and...
|
175
176
177
178
179
|
/// @param E (GPU) is the N x N destination scalar field
/// @param N is the number fo elements of the scalar field in each direction
/// @param x (GPU) is the grid of X coordinates for each point in E
/// @param y (GPU) is the grid of Y coordinates for each point in E
/// @param z (GPU) is the grid of Z coordinates for each point in E
|
6a53ac0c
David Mayerich
updated optics fi...
|
180
181
|
/// @param W (GPU) is an array of coherent scalar plane waves incident on the Mie scatterer
/// @param nW is the number of plane waves to evaluate (sum)
|
2a10ecf4
David Mayerich
updated files and...
|
182
183
|
/// @param a is the radius of the Mie scatterer
/// @param n is the complex refractive index of the Mie scatterer
|
6a53ac0c
David Mayerich
updated optics fi...
|
184
185
186
187
188
189
|
/// @param c is the position of the sphere in (x, y, z)
/// @param hB (GPU) is a look-up table of Hankel functions (equally spaced in distance from the sphere) pre-multiplied with scattering coefficients
/// @param kr_min is the minimum kr value in the hB look-up table (corresponding to the closest point to the sphere)
/// @param dkr is the spacing (in kr) between samples of the hB look-up table
/// @param N_hB is the number of samples in hB
/// @param Nl is the order of the calculation (number of Hankel function orders)
|
9339fbad
David Mayerich
implementing mie ...
|
190
|
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
191
|
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::vec3<T> c, stim::complex<T>* hB, T kr_min, T dkr, size_t N_hB, size_t Nl){
|
31262e83
David Mayerich
GPU implementatio...
|
192
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
193
194
195
|
size_t max_shared_mem = stim::sharedMemPerBlock(); //get the amount of shared memory per block
size_t hBl_array = sizeof(stim::complex<T>) * (Nl + 1); //calculate the number of bytes required to store the LUT corresponding to a single sample in shared memory
int threads = (int)((max_shared_mem / hBl_array) / 32 * 32); //calculate the optimal number of threads per block (make sure it's divisible by the number of warps - 32)
|
31262e83
David Mayerich
GPU implementatio...
|
196
197
198
199
200
|
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 ...
|
201
|
//std::cout<<"shared memory allocated: "<<shared_mem<<std::endl;
|
2a10ecf4
David Mayerich
updated files and...
|
202
|
cuda_scalar_mie_scatter<T><<< blocks, threads, shared_mem >>>(E, N, x, y, z, W, nW, a, n, c, hB, kr_min, dkr, N_hB, (int)Nl); //call the kernel
|
31262e83
David Mayerich
GPU implementatio...
|
203
204
205
|
}
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
206
|
__global__ void cuda_dist(T* r, T* x, T* y, T* z, size_t N, stim::vec3<T> c = stim::vec3<T>(0, 0, 0)){
|
31262e83
David Mayerich
GPU implementatio...
|
207
208
209
210
211
212
213
214
|
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];
|
2a10ecf4
David Mayerich
updated files and...
|
215
|
r[i] = (p - c).len();
|
9339fbad
David Mayerich
implementing mie ...
|
216
|
}
|
2a10ecf4
David Mayerich
updated files and...
|
217
218
219
|
///Calculate the scalar Mie scattered field on the GPU
/// @param E (GPU) is the N x N destination scalar field
|
6a53ac0c
David Mayerich
updated optics fi...
|
220
|
/// @param N is the number of sample points of the scalar field
|
2a10ecf4
David Mayerich
updated files and...
|
221
222
223
224
225
226
227
|
/// @param x (GPU) is the grid of X coordinates for each point in E
/// @param y (GPU) is the grid of Y coordinates for each point in E
/// @param z (GPU) is the grid of Z coordinates for each point in E
/// @param W (CPU) is an array of coherent scalar plane waves incident on the Mie scatterer
/// @param a is the radius of the Mie scatterer
/// @param n is the complex refractive index of the Mie scatterer
/// @param r_spacing is the minimum distance between r values of the sample points in E (used to calculate look-up tables)
|
9339fbad
David Mayerich
implementing mie ...
|
228
|
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
229
|
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, stim::vec3<T> c = stim::vec3<T>(0, 0, 0), T r_spacing = 0.1){
|
4252d827
David Mayerich
ivote3 fixes and ...
|
230
|
|
9339fbad
David Mayerich
implementing mie ...
|
231
232
233
|
//calculate the necessary number of orders required to represent the scattered field
T k = W[0].kmag();
|
6a53ac0c
David Mayerich
updated optics fi...
|
234
|
int Nl = (int)ceil(k*a + 4 * cbrt( k * a ) + 2); //calculate the number of orders required to represent the sphere
|
31262e83
David Mayerich
GPU implementatio...
|
235
|
if(Nl < LOCAL_NL) Nl = LOCAL_NL; //always do at least the minimum number of local operations (kernel optimization)
|
9339fbad
David Mayerich
implementing mie ...
|
236
237
238
|
//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
|
6a53ac0c
David Mayerich
updated optics fi...
|
239
|
B_coefficients(B, a, k, n, Nl); //calculate the scattering coefficients
|
4252d827
David Mayerich
ivote3 fixes and ...
|
240
|
|
9339fbad
David Mayerich
implementing mie ...
|
241
|
// PLANE WAVES
|
6a53ac0c
David Mayerich
updated optics fi...
|
242
|
stim::scalarwave<T>* dev_W; //allocate space and copy plane waves
|
9339fbad
David Mayerich
implementing mie ...
|
243
244
245
|
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 ...
|
246
|
// BESSEL FUNCTION LOOK-UP TABLE
|
6a53ac0c
David Mayerich
updated optics fi...
|
247
248
249
|
//calculate the distance from the sphere center at each sample point and store the result in dev_r
T* dev_r; //declare the device pointer to hold the distance from the sphere center
HANDLE_ERROR( cudaMalloc(&dev_r, sizeof(T) * N) ); //allocate space for the array of distances
|
31262e83
David Mayerich
GPU implementatio...
|
250
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
251
252
253
|
int threads = stim::maxThreadsPerBlock(); //query the device to find the maximum number of threads per block
dim3 blocks((unsigned)(N / threads + 1)); //calculate the number of blocks necessary to evaluate the total number of sample points N
cuda_dist<T> <<< blocks, threads >>>(dev_r, x, y, z, N, c); //calculate the distance
|
31262e83
David Mayerich
GPU implementatio...
|
254
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
255
|
//Use the cuBLAS library to find the minimum and maximum distances from the sphere center. This will be used to create a look-up table for the Hankel functions
|
31262e83
David Mayerich
GPU implementatio...
|
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
|
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...
|
277
278
|
i_min--; //cuBLAS uses 1-based indexing for Fortran compatibility
i_max--;
|
31262e83
David Mayerich
GPU implementatio...
|
279
280
281
282
|
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) );
|
6a53ac0c
David Mayerich
updated optics fi...
|
283
|
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...
|
284
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
285
|
size_t N_hB_lut = (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
|
9339fbad
David Mayerich
implementing mie ...
|
286
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
287
288
|
//Declare and evaluate variables used to calculate the spherical Bessel functions and store them temporarily on the CPU
double vm; //allocate space to store the return values for the bessel function calculation
|
9339fbad
David Mayerich
implementing mie ...
|
289
290
291
292
293
|
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) );
|
6a53ac0c
David Mayerich
updated optics fi...
|
294
295
296
297
298
|
size_t hB_bytes = sizeof(stim::complex<T>) * (Nl+1) * N_hB_lut; //calculate the number of bytes necessary to store the Hankel function LUT
stim::complex<T>* hB_lut = (stim::complex<T>*) malloc(hB_bytes); //pointer to the look-up table
T dr = (r_max - r_min) / (N_hB_lut-1); //calculate the optimal distance between values in the LUT
stim::complex<T> hl; //declare a complex value for the Hankel function result
for(size_t ri = 0; ri < N_hB_lut; ri++){ //for each value in the LUT
|
8309b07a
David Mayerich
fixed some vec3 e...
|
299
|
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]
|
6a53ac0c
David Mayerich
updated optics fi...
|
300
301
|
for(size_t l = 0; l <= Nl; l++){ //for each order
hl.r = (T)jv[l]; //generate the spherical Hankel function from the Bessel functions
|
31262e83
David Mayerich
GPU implementatio...
|
302
303
|
hl.i = (T)yv[l];
|
6a53ac0c
David Mayerich
updated optics fi...
|
304
|
hB_lut[ri * (Nl + 1) + l] = hl * B[l]; //pre-multiply the Hankel function by the scattering coefficients
|
9339fbad
David Mayerich
implementing mie ...
|
305
306
|
}
}
|
9339fbad
David Mayerich
implementing mie ...
|
307
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
308
|
//Copy the pre-multiplied Hankel function look-up table to the GPU - this LUT gives a list of uniformly spaced Hankel function values pre-multiplied by scattering coefficients
|
31262e83
David Mayerich
GPU implementatio...
|
309
310
311
|
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) );
|
6a53ac0c
David Mayerich
updated optics fi...
|
312
313
|
//calculate the Mie scattering solution on the GPU
|
2a10ecf4
David Mayerich
updated files and...
|
314
|
gpu_scalar_mie_scatter<T>(E, N, x, y, z, dev_W, W.size(), a, n, c, dev_hB_lut, r_min, dr, N_hB_lut, Nl);
|
9339fbad
David Mayerich
implementing mie ...
|
315
|
|
6a53ac0c
David Mayerich
updated optics fi...
|
316
|
//HANDLE_ERROR(cudaMemcpy(E, E, N * sizeof(stim::complex<T>), cudaMemcpyDeviceToHost)); //copy the field from device memory
|
2a10ecf4
David Mayerich
updated files and...
|
317
318
319
320
|
HANDLE_ERROR(cudaFree(dev_hB_lut));
HANDLE_ERROR(cudaFree(dev_r));
HANDLE_ERROR(cudaFree(dev_W));
|
9339fbad
David Mayerich
implementing mie ...
|
321
|
|
4252d827
David Mayerich
ivote3 fixes and ...
|
322
323
324
325
326
327
328
329
330
331
332
333
|
}
/// 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>
|
2a10ecf4
David Mayerich
updated files and...
|
334
|
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, stim::vec3<T> c = stim::vec3<T>(0, 0, 0)){
|
4252d827
David Mayerich
ivote3 fixes and ...
|
335
336
|
|
2a10ecf4
David Mayerich
updated files and...
|
337
|
/*#ifdef CUDA_FOUND
|
4252d827
David Mayerich
ivote3 fixes and ...
|
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
|
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));
}
|
2a10ecf4
David Mayerich
updated files and...
|
360
|
gpu_scalar_mie_scatter(dev_E, N, dev_x, dev_y, dev_z, W, a, n, c, r_spacing);
|
9339fbad
David Mayerich
implementing mie ...
|
361
362
363
364
365
366
|
if(x != NULL) cudaFree(dev_x); //free everything
if(y != NULL) cudaFree(dev_y);
if(z != NULL) cudaFree(dev_z);
cudaFree(dev_E);
#else
|
2a10ecf4
David Mayerich
updated files and...
|
367
|
*/
|
9339fbad
David Mayerich
implementing mie ...
|
368
|
|
4252d827
David Mayerich
ivote3 fixes and ...
|
369
370
371
372
373
374
375
376
377
378
|
//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 ...
|
379
380
381
382
383
384
385
386
387
388
389
390
|
//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;
|
2a10ecf4
David Mayerich
updated files and...
|
391
|
stim::complex<T> Ew;
|
9339fbad
David Mayerich
implementing mie ...
|
392
393
394
395
396
397
|
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];
|
2a10ecf4
David Mayerich
updated files and...
|
398
|
p = p - c;
|
9339fbad
David Mayerich
implementing mie ...
|
399
400
401
|
r = p.len();
if(r >= a){
for(size_t w = 0; w < W.size(); w++){
|
176a9b1c
Sebastian
fixed type mismat...
|
402
|
Ew = W[w].E() * exp(stim::complex<float>(0, W[w].kvec().dot(c)));
|
9339fbad
David Mayerich
implementing mie ...
|
403
404
405
406
407
408
409
410
|
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];
|
2a10ecf4
David Mayerich
updated files and...
|
411
|
E[i] += Ew * B[l] * h * P[l];
|
9339fbad
David Mayerich
implementing mie ...
|
412
413
414
415
|
}
}
}
}
|
2a10ecf4
David Mayerich
updated files and...
|
416
|
//#endif
|
9339fbad
David Mayerich
implementing mie ...
|
417
418
419
|
}
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
420
|
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, stim::vec3<T> c = stim::vec3<T>(0, 0, 0), T r_spacing = 0.1){
|
9339fbad
David Mayerich
implementing mie ...
|
421
|
std::vector< stim::scalarwave<T> > W(1, w);
|
2a10ecf4
David Mayerich
updated files and...
|
422
|
cpu_scalar_mie_scatter(E, N, x, y, z, W, a, n, c, r_spacing);
|
8309b07a
David Mayerich
fixed some vec3 e...
|
423
424
425
426
427
428
429
430
431
432
433
434
435
436
|
}
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...
|
437
|
if(r >= a) return; //exit if the point is inside the sphere (we only calculate the internal field)
|
2a10ecf4
David Mayerich
updated files and...
|
438
|
T fij = (r - r_min)/dr; //FP index into the spherical bessel LUT
|
8309b07a
David Mayerich
fixed some vec3 e...
|
439
440
|
size_t ij = (size_t) fij; //convert to an integral index
T alpha = fij - ij; //calculate the fractional portion of the index
|
2a10ecf4
David Mayerich
updated files and...
|
441
442
|
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
|
8309b07a
David Mayerich
fixed some vec3 e...
|
443
444
|
T cos_phi;
|
2a10ecf4
David Mayerich
updated files and...
|
445
|
T Pl_2, Pl_1, Pl; //declare registers to store the previous two Legendre polynomials
|
8309b07a
David Mayerich
fixed some vec3 e...
|
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
|
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
|
2a10ecf4
David Mayerich
updated files and...
|
464
|
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...
|
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
|
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 ...
|
493
494
|
//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...
|
495
|
int threads = (int)((max_shared_mem / hBl_array) / 32 * 32);
|
4252d827
David Mayerich
ivote3 fixes and ...
|
496
|
//std::cout<<"threads per block: "<<threads<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
497
498
499
500
501
|
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 ...
|
502
|
//std::cout<<"shared memory allocated: "<<shared_mem<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
503
|
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 ...
|
504
505
506
507
508
509
510
511
512
513
514
515
516
|
}
/// 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>
|
2a10ecf4
David Mayerich
updated files and...
|
517
|
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, stim::vec3<T> c = stim::vec3<T>(0, 0, 0)){
|
8309b07a
David Mayerich
fixed some vec3 e...
|
518
|
//calculate the necessary number of orders required to represent the scattered field
|
9339fbad
David Mayerich
implementing mie ...
|
519
520
|
T k = W[0].kmag();
|
8309b07a
David Mayerich
fixed some vec3 e...
|
521
522
|
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 ...
|
523
|
//std::cout<<"Nl: "<<Nl<<std::endl;
|
9339fbad
David Mayerich
implementing mie ...
|
524
525
526
527
528
|
//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);
|
2a10ecf4
David Mayerich
updated files and...
|
529
|
/*#ifdef CUDA_FOUND
|
8309b07a
David Mayerich
fixed some vec3 e...
|
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
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
|
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);
}
|
2a10ecf4
David Mayerich
updated files and...
|
587
|
cublasDestroy(handle); //destroy the CUBLAS handle
|
8309b07a
David Mayerich
fixed some vec3 e...
|
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
|
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 ...
|
609
|
//std::cout<<"LUT jl bytes: "<<jA_bytes<<std::endl;
|
8309b07a
David Mayerich
fixed some vec3 e...
|
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
|
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 ...
|
631
632
633
634
|
HANDLE_ERROR( cudaFree(dev_jA_lut) );
HANDLE_ERROR( cudaFree(dev_E) );
HANDLE_ERROR( cudaFree(dev_W) );
HANDLE_ERROR( cudaFree(dev_r) );
|
2a10ecf4
David Mayerich
updated files and...
|
635
|
HANDLE_ERROR( cudaFree(dev_E) );
|
8309b07a
David Mayerich
fixed some vec3 e...
|
636
|
#else
|
2a10ecf4
David Mayerich
updated files and...
|
637
|
*/
|
9339fbad
David Mayerich
implementing mie ...
|
638
639
640
641
642
643
644
645
646
647
648
649
|
//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;
|
2a10ecf4
David Mayerich
updated files and...
|
650
|
stim::complex<T> Ew;
|
9339fbad
David Mayerich
implementing mie ...
|
651
652
653
654
655
656
|
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];
|
2a10ecf4
David Mayerich
updated files and...
|
657
|
p = p - c;
|
9339fbad
David Mayerich
implementing mie ...
|
658
659
660
661
|
r = p.len();
if(r < a){
E[i] = 0;
for(size_t w = 0; w < W.size(); w++){
|
176a9b1c
Sebastian
fixed type mismat...
|
662
|
Ew = W[w].E() * exp(stim::complex<float>(0, W[w].kvec().dot(c)));
|
9339fbad
David Mayerich
implementing mie ...
|
663
664
665
666
667
668
669
670
671
672
|
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++){
|
2a10ecf4
David Mayerich
updated files and...
|
673
|
E[i] += Ew * A[l] * (stim::complex<T>)j_knr[l] * P[l];
|
9339fbad
David Mayerich
implementing mie ...
|
674
675
676
677
|
}
}
}
}
|
2a10ecf4
David Mayerich
updated files and...
|
678
|
//#endif
|
9339fbad
David Mayerich
implementing mie ...
|
679
680
681
|
}
template<typename T>
|
2a10ecf4
David Mayerich
updated files and...
|
682
|
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, stim::vec3<T> c = stim::vec3<T>(0, 0, 0)){
|
9339fbad
David Mayerich
implementing mie ...
|
683
|
std::vector< stim::scalarwave<T> > W(1, w);
|
2a10ecf4
David Mayerich
updated files and...
|
684
|
cpu_scalar_mie_internal(E, N, x, y, z, W, a, n, c);
|
9339fbad
David Mayerich
implementing mie ...
|
685
686
|
}
|
963d0676
David Mayerich
bug fixes related...
|
687
688
689
690
691
|
/// 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
{
|
2a10ecf4
David Mayerich
updated files and...
|
692
|
public:
|
963d0676
David Mayerich
bug fixes related...
|
693
694
|
T radius; //radius of the scattering sphere
stim::complex<T> n; //refractive index of the scattering sphere
|
2a10ecf4
David Mayerich
updated files and...
|
695
|
vec3<T> c; //position of the sphere in space
|
963d0676
David Mayerich
bug fixes related...
|
696
697
698
|
public:
|
2a10ecf4
David Mayerich
updated files and...
|
699
700
701
702
703
704
705
|
scalarmie() { //default constructor
radius = 0.5;
n = stim::complex<T>(1.4, 0.0);
c = stim::vec3<T>(0, 0, 0);
}
scalarmie(T r, stim::complex<T> ri, stim::vec3<T> center = stim::vec3<T>(0, 0, 0)){
|
963d0676
David Mayerich
bug fixes related...
|
706
707
|
radius = r;
n = ri;
|
2a10ecf4
David Mayerich
updated files and...
|
708
709
|
c = center;
//c = stim::vec3<T>(2, 1, 0);
|
963d0676
David Mayerich
bug fixes related...
|
710
711
|
}
|
2a10ecf4
David Mayerich
updated files and...
|
712
713
714
715
|
//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());
//}
|
4252d827
David Mayerich
ivote3 fixes and ...
|
716
|
|
2a10ecf4
David Mayerich
updated files and...
|
717
718
719
720
|
//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());
//}
|
4252d827
David Mayerich
ivote3 fixes and ...
|
721
|
|
2a10ecf4
David Mayerich
updated files and...
|
722
723
724
725
726
727
|
//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);
//}
|
4252d827
David Mayerich
ivote3 fixes and ...
|
728
|
|
963d0676
David Mayerich
bug fixes related...
|
729
730
|
void eval(stim::scalarfield<T>& E, stim::scalarbeam<T> b, int order = 500, int samples = 1000){
|
4252d827
David Mayerich
ivote3 fixes and ...
|
731
732
|
E.meshgrid();
b.eval(E, order);
|
963d0676
David Mayerich
bug fixes related...
|
733
734
|
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 ...
|
735
736
|
if(E.gpu()){
|
2a10ecf4
David Mayerich
updated files and...
|
737
|
stim::gpu_scalar_mie_scatter<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, c, E.spacing());
|
4252d827
David Mayerich
ivote3 fixes and ...
|
738
739
740
741
742
|
}
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());
}
|
963d0676
David Mayerich
bug fixes related...
|
743
744
745
746
|
}
}; //end stim::scalarmie
|
2a10ecf4
David Mayerich
updated files and...
|
747
748
|
template<typename T>
class scalarcluster : public std::vector< scalarmie<T> > {
|
ca41e27d
David Mayerich
Linux edits for B...
|
749
|
|
2a10ecf4
David Mayerich
updated files and...
|
750
751
752
753
754
755
756
757
758
759
760
|
public:
void eval(stim::scalarfield<T>& E, stim::scalarbeam<T> b, int order = 500, int samples = 1000) {
E.meshgrid();
b.eval(E, order);
std::vector< stim::scalarwave<float> > wave_array = b.mc(samples); //decompose the beam into an array of plane waves
T radius;
stim::complex<T> n;
stim::vec3<T> c;
|
ca41e27d
David Mayerich
Linux edits for B...
|
761
762
763
764
|
for (size_t si = 0; si < std::vector< scalarmie<T> >::size(); si++) { //for each sphere in the cluster
radius = std::vector< scalarmie<T> >::at(si).radius;
n = std::vector< scalarmie<T> >::at(si).n;
c = std::vector< scalarmie<T> >::at(si).c;
|
2a10ecf4
David Mayerich
updated files and...
|
765
766
767
768
769
770
771
772
773
774
775
|
if (E.gpu()) {
stim::gpu_scalar_mie_scatter<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, c, E.spacing());
}
else {
stim::cpu_scalar_mie_scatter<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, c);
stim::cpu_scalar_mie_internal<float>(E.ptr(), E.size(), E.x(), E.y(), E.z(), wave_array, radius, n, c);
}
}
}
};
|
963d0676
David Mayerich
bug fixes related...
|
776
|
} //end namespace stim
|
9339fbad
David Mayerich
implementing mie ...
|
777
778
|
#endif
|