24aab7c9
David Mayerich
added field and c...
|
1
2
3
4
5
6
7
|
#ifndef RTS_FIELD_CUH
#define RTS_FIELD_CUH
#include <vector>
#include <string>
#include <sstream>
|
180d7f3a
David Mayerich
added binary file...
|
8
9
10
|
#include "cublas_v2.h"
#include <cuda_runtime.h>
|
24aab7c9
David Mayerich
added field and c...
|
11
12
13
14
|
#include "../math/rect.h"
#include "../cuda/threads.h"
#include "../cuda/error.h"
#include "../cuda/devices.h"
|
180d7f3a
David Mayerich
added binary file...
|
15
|
#include "../visualization/colormap.h"
|
24aab7c9
David Mayerich
added field and c...
|
16
17
|
|
8a86bd56
David Mayerich
changed rts names...
|
18
|
namespace stim{
|
24aab7c9
David Mayerich
added field and c...
|
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
|
//multiply R = X * Y
template<typename T>
__global__ void gpu_field_multiply(T* R, T* X, T* Y, unsigned int r0, unsigned int r1){
int iu = blockIdx.x * blockDim.x + threadIdx.x;
int iv = blockIdx.y * blockDim.y + threadIdx.y;
//make sure that the thread indices are in-bounds
if(iu >= r0 || iv >= r1) return;
//compute the index into the field
int i = iv*r0 + iu;
//calculate and store the result
R[i] = X[i] * Y[i];
}
//assign a constant value to all points
template<typename T>
__global__ void gpu_field_assign(T* ptr, T val, unsigned int r0, unsigned int r1){
int iu = blockIdx.x * blockDim.x + threadIdx.x;
int iv = blockIdx.y * blockDim.y + threadIdx.y;
//make sure that the thread indices are in-bounds
if(iu >= r0 || iv >= r1) return;
//compute the index into the field
int i = iv*r0 + iu;
//calculate and store the result
ptr[i] = val;
}
|
7a2d0012
David Mayerich
mirst1d updates
|
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
|
//crop the field to the new dimensions (width x height)
template<typename T>
__global__ void gpu_field_crop(T* dest, T* source,
unsigned int r0, unsigned int r1,
unsigned int width, unsigned int height){
int iu = blockIdx.x * blockDim.x + threadIdx.x;
int iv = blockIdx.y * blockDim.y + threadIdx.y;
//make sure that the thread indices are in-bounds
if(iu >= width || iv >= height) return;
//compute the index into the field
int is = iv*r0 + iu;
int id = iv*width + iu;
//calculate and store the result
dest[id] = source[is];
}
template<typename T, unsigned int D = 1>
|
24aab7c9
David Mayerich
added field and c...
|
75
76
77
78
79
80
|
class field{
protected:
T* X[D]; //pointer to the field data
unsigned int R[2]; //field resolution
|
ef70330e
David Mayerich
changed rts to stim
|
81
|
stim::rect<T> shape; //position and shape of the field slice
|
24aab7c9
David Mayerich
added field and c...
|
82
|
|
7a2d0012
David Mayerich
mirst1d updates
|
83
84
|
//calculates the optimal block and grid sizes using information from the GPU
void cuda_params(dim3& grids, dim3& blocks){
|
ef70330e
David Mayerich
changed rts to stim
|
85
|
int maxThreads = stim::maxThreadsPerBlock(); //compute the optimal block size
|
7a2d0012
David Mayerich
mirst1d updates
|
86
87
88
89
90
91
92
|
int SQRT_BLOCK = (int)std::sqrt((float)maxThreads);
//create one thread for each detector pixel
blocks = dim3(SQRT_BLOCK, SQRT_BLOCK);
grids = dim3((R[0] + SQRT_BLOCK -1)/SQRT_BLOCK, (R[1] + SQRT_BLOCK - 1)/SQRT_BLOCK);
}
|
180d7f3a
David Mayerich
added binary file...
|
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
|
//find the maximum value of component n
T find_max(unsigned int n){
cublasStatus_t stat;
cublasHandle_t handle;
//create a CUBLAS handle
stat = cublasCreate(&handle);
if(stat != CUBLAS_STATUS_SUCCESS){
std::cout<<"CUBLAS Error: initialization failed"<<std::endl;
exit(1);
}
int L = R[0] * R[1]; //compute the number of discrete points in a slice
int index; //result of the max operation
T result;
if(sizeof(T) == 4)
stat = cublasIsamax(handle, L, (const float*)X[n], 1, &index);
else
stat = cublasIdamax(handle, L, (const double*)X[n], 1, &index);
index -= 1; //adjust for 1-based indexing
//if there was a GPU error, terminate
if(stat != CUBLAS_STATUS_SUCCESS){
std::cout<<"CUBLAS Error: failure finding maximum value."<<std::endl;
exit(1);
}
//retrieve the maximum value for this slice and store it in the maxVal array
|
180d7f3a
David Mayerich
added binary file...
|
123
124
125
126
|
HANDLE_ERROR(cudaMemcpy(&result, X[n] + index, sizeof(T), cudaMemcpyDeviceToHost));
return result;
}
|
24aab7c9
David Mayerich
added field and c...
|
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
|
public:
//returns a list of file names given an input string with wild cards
std::vector<std::string> process_filename(std::string name){
std::stringstream ss(name);
std::string item;
std::vector<std::string> elems;
while(std::getline(ss, item, '.')) //split the string at the '.' character (filename and extension)
{
elems.push_back(item);
}
std::string prefix = elems[0]; //prefix contains the filename (with wildcard '?' characters)
std::string ext = elems[1]; //file extension (ex. .bmp, .png)
ext = std::string(".") + ext; //add a period back into the extension
size_t i0 = prefix.find_first_of("?"); //find the positions of the first and last wildcard ('?'')
size_t i1 = prefix.find_last_of("?");
std::string postfix = prefix.substr(i1+1);
prefix = prefix.substr(0, i0);
unsigned int digits = i1 - i0 + 1; //compute the number of wildcards
std::vector<std::string> flist; //create a vector of file names
//fill the list
for(unsigned int d=0; d<D; d++){
std::stringstream ss; //assemble the file name
ss<<prefix<<std::setfill('0')<<std::setw(digits)<<d<<postfix<<ext;
flist.push_back(ss.str());
}
return flist;
}
void init(){
for(unsigned int n=0; n<D; n++)
X[n] = NULL;
}
void destroy(){
for(unsigned int n=0; n<D; n++)
if(X[n] != NULL)
HANDLE_ERROR(cudaFree(X[n]));
}
|
24aab7c9
David Mayerich
added field and c...
|
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
|
public:
//field constructor
field(){
R[0] = R[1] = 0;
init();
}
field(unsigned int x, unsigned int y){
//set the resolution
R[0] = x;
R[1] = y;
//allocate memory on the GPU
for(unsigned int n=0; n<D; n++){
HANDLE_ERROR(cudaMalloc( (void**)&X[n], sizeof(T) * R[0] * R[1] ));
}
clear(); //zero the field
}
///copy constructor
field(const field &rhs){
//first make a shallow copy
R[0] = rhs.R[0];
R[1] = rhs.R[1];
for(unsigned int n=0; n<D; n++){
//do we have to make a deep copy?
if(rhs.X[n] == NULL)
X[n] = NULL; //no
else{
//allocate the necessary memory
HANDLE_ERROR(cudaMalloc(&X[n], sizeof(T) * R[0] * R[1]));
//copy the slice
HANDLE_ERROR(cudaMemcpy(X[n], rhs.X[n], sizeof(T) * R[0] * R[1], cudaMemcpyDeviceToDevice));
}
}
}
~field(){
destroy();
}
//assignment operator
field & operator= (const field & rhs){
//de-allocate any existing GPU memory
destroy();
//copy the slice resolution
R[0] = rhs.R[0];
R[1] = rhs.R[1];
for(unsigned int n=0; n<D; n++)
{
//allocate the necessary memory
HANDLE_ERROR(cudaMalloc(&X[n], sizeof(T) * R[0] * R[1]));
//copy the slice
HANDLE_ERROR(cudaMemcpy(X[n], rhs.X[n], sizeof(T) * R[0] * R[1], cudaMemcpyDeviceToDevice));
}
return *this;
}
field & operator= (const T rhs){
|
ef70330e
David Mayerich
changed rts to stim
|
236
|
int maxThreads = stim::maxThreadsPerBlock(); //compute the optimal block size
|
24aab7c9
David Mayerich
added field and c...
|
237
238
239
240
241
242
243
244
|
int SQRT_BLOCK = (int)std::sqrt((float)maxThreads);
//create one thread for each detector pixel
dim3 dimBlock(SQRT_BLOCK, SQRT_BLOCK);
dim3 dimGrid((R[0] + SQRT_BLOCK -1)/SQRT_BLOCK, (R[1] + SQRT_BLOCK - 1)/SQRT_BLOCK);
//assign the constant value to all positions and dimensions
for(int n=0; n<D; n++)
|
ef70330e
David Mayerich
changed rts to stim
|
245
|
stim::gpu_field_assign <<<dimGrid, dimBlock>>> (X[n], rhs, R[0], R[1]);
|
24aab7c9
David Mayerich
added field and c...
|
246
247
248
249
250
251
252
|
return *this;
}
//assignment of vector component
field & operator= (const vec<T, D> rhs){
|
ef70330e
David Mayerich
changed rts to stim
|
253
|
int maxThreads = stim::maxThreadsPerBlock(); //compute the optimal block size
|
24aab7c9
David Mayerich
added field and c...
|
254
255
256
257
258
259
260
261
|
int SQRT_BLOCK = (int)std::sqrt((float)maxThreads);
//create one thread for each detector pixel
dim3 dimBlock(SQRT_BLOCK, SQRT_BLOCK);
dim3 dimGrid((R[0] + SQRT_BLOCK -1)/SQRT_BLOCK, (R[1] + SQRT_BLOCK - 1)/SQRT_BLOCK);
//assign the constant value to all positions and dimensions
for(unsigned int n=0; n<D; n++)
|
ef70330e
David Mayerich
changed rts to stim
|
262
|
stim::gpu_field_assign <<<dimGrid, dimBlock>>> (X[n], rhs.v[n], R[0], R[1]);
|
24aab7c9
David Mayerich
added field and c...
|
263
264
265
266
267
268
269
270
|
return *this;
}
//multiply two fields (element-wise multiplication)
field<T, D> operator* (const field & rhs){
|
ef70330e
David Mayerich
changed rts to stim
|
271
|
int maxThreads = stim::maxThreadsPerBlock(); //compute the optimal block size
|
24aab7c9
David Mayerich
added field and c...
|
272
273
274
275
276
277
278
279
280
281
|
int SQRT_BLOCK = (int)std::sqrt((float)maxThreads);
//create one thread for each detector pixel
dim3 dimBlock(SQRT_BLOCK, SQRT_BLOCK);
dim3 dimGrid((R[0] + SQRT_BLOCK -1)/SQRT_BLOCK, (R[1] + SQRT_BLOCK - 1)/SQRT_BLOCK);
//create a scalar field to store the result
field<T, D> result(R[0], R[1]);
for(int n=0; n<D; n++)
|
ef70330e
David Mayerich
changed rts to stim
|
282
|
stim::gpu_field_multiply <<<dimGrid, dimBlock>>> (result.X[n], X[n], rhs.X[n], R[0], R[1]);
|
24aab7c9
David Mayerich
added field and c...
|
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
|
return result;
}
T* ptr(unsigned int n = 0){
if(n < D)
return X[n];
else return NULL;
}
//return the vector component at position (u, v)
vec<T, D> get(unsigned int u, unsigned int v){
vec<T, D> result;
for(unsigned int d=0; d<D; d++){
HANDLE_ERROR(cudaMemcpy(&result[d], X[d] + v*R[0] + u, sizeof(T), cudaMemcpyDeviceToHost));
}
return result;
}
//set all components of the field to zero
void clear(){
for(unsigned int n=0; n<D; n++)
if(X[n] != NULL)
HANDLE_ERROR(cudaMemset(X[n], 0, sizeof(T) * R[0] * R[1]));
}
|
7a2d0012
David Mayerich
mirst1d updates
|
311
312
|
//crop the field
field<T, D> crop(unsigned int width, unsigned int height){
|
ef70330e
David Mayerich
changed rts to stim
|
313
|
int maxThreads = stim::maxThreadsPerBlock(); //compute the optimal block size
|
7a2d0012
David Mayerich
mirst1d updates
|
314
315
316
317
318
319
320
321
322
323
|
int SQRT_BLOCK = (int)std::sqrt((float)maxThreads);
//create one thread for each detector pixel
dim3 dimBlock(SQRT_BLOCK, SQRT_BLOCK);
dim3 dimGrid((width + SQRT_BLOCK -1)/SQRT_BLOCK, (height + SQRT_BLOCK - 1)/SQRT_BLOCK);
//create a scalar field to store the result
field<T, D> result(width, height);
for(int n=0; n<D; n++)
|
ef70330e
David Mayerich
changed rts to stim
|
324
|
stim::gpu_field_crop <<<dimGrid, dimBlock>>> (result.X[n], X[n], R[0], R[1], width, height);
|
7a2d0012
David Mayerich
mirst1d updates
|
325
326
|
return result;
|
180d7f3a
David Mayerich
added binary file...
|
327
328
329
330
|
}
//save an image representing component n
void toImage(std::string filename, unsigned int n = 0,
|
ef70330e
David Mayerich
changed rts to stim
|
331
|
bool positive = false, stim::colormapType cmap = stim::cmBrewer){
|
180d7f3a
David Mayerich
added binary file...
|
332
|
T max_val = find_max(n); //find the maximum value
|
7a2d0012
David Mayerich
mirst1d updates
|
333
|
|
180d7f3a
David Mayerich
added binary file...
|
334
|
if(positive) //if the field is positive, use the range [0 max_val]
|
ef70330e
David Mayerich
changed rts to stim
|
335
|
stim::gpu2image<T>(X[n], filename, R[0], R[1], 0, max_val, cmap);
|
180d7f3a
David Mayerich
added binary file...
|
336
|
else
|
ef70330e
David Mayerich
changed rts to stim
|
337
|
stim::gpu2image<T>(X[n], filename, R[0], R[1], -max_val, max_val, cmap);
|
7a2d0012
David Mayerich
mirst1d updates
|
338
339
|
}
|
24aab7c9
David Mayerich
added field and c...
|
340
341
342
|
};
} //end namespace rts
|
180d7f3a
David Mayerich
added binary file...
|
343
|
#endif
|