-
Notifications
You must be signed in to change notification settings - Fork 0
/
opencl_setup.cpp
196 lines (165 loc) · 5.53 KB
/
opencl_setup.cpp
1
2
3
4
5
6
7
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
38
39
40
41
42
43
44
45
46
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
85
86
87
88
89
90
91
92
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
123
124
125
126
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
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
#include <cstring>
#include <cstdio>
#include <cstdlib>
#include <string>
#include <cassert>
#include <cmath>
#include <unistd.h>
#include "clhelp.h"
#include "kernel_struct.h"
struct cl_runtime_env
{
cl_vars_t cv;
std::map<std::string, cl_mem_data> gpu_data;
std::map<std::string, cl_kernel> cl_kernels;
kernel* kernels;
int num_kerns;
};
cl_vars_t setupRuntime(kernel*, std::map<std::string, cl_kernel>*, int);
void releaseClMem(std::map<std::string, cl_mem_data>);
void moveDataToGpu(cl_vars_t,
std::map<std::string, data_array>,
std::map<std::string, cl_mem_data>*,
kernel*, int);
void runKernel(cl_vars_t cv,
cl_kernel cl_kern,
kernel kern,
std::map<std::string, cl_mem_data> gpu_data,
double* vars);
void runKernel(cl_runtime_env env,
std::string kernel_name,
double* vars,
double* out,
int start_index,
int out_len)
{
kernel kern;
for (int i = 1; i < env.num_kerns; i++)
{
if (env.kernels[i].name == kernel_name)
kern = env.kernels[i];
}
runKernel(env.cv, env.cl_kernels[kernel_name],
kern,
env.gpu_data,
vars);
cl_int err;
err = clEnqueueReadBuffer(env.cv.commands, env.gpu_data["out"].array,
true,
start_index,
sizeof(double)*out_len,
out,
0, NULL, NULL);
CHK_ERR(err);
err = clFlush(env.cv.commands);
CHK_ERR(err);
}
void setupEnvironment(cl_runtime_env* env,
kernel* kernels,
std::map<std::string, data_array> cpu_data,
int num_kerns)
{
(*env).cv = setupRuntime(kernels, &(*env).cl_kernels, num_kerns);
moveDataToGpu((*env).cv, cpu_data, &(*env).gpu_data, kernels, num_kerns);
(*env).kernels = kernels;
(*env).num_kerns = num_kerns;
}
cl_vars_t setupRuntime(kernel* kernels, std::map<std::string, cl_kernel>*
kernel_map, int num_kerns)
{
std::string kernel_source_str[num_kerns-1];
std::string arraycompact_kernel_file[num_kerns-1];
cl_vars_t cv;
std::list<std::string> kernel_names;
//get the names of the kernel files
for (int i = 1; i<num_kerns; i++) {
arraycompact_kernel_file[i-1] = kernels[i].name + ".cl";
kernel_names.push_back(kernels[i].name);
}
cl_int err = CL_SUCCESS;
//read the kernel files
readFile(arraycompact_kernel_file,
kernel_source_str, num_kerns-1);
initialize_ocl(cv);
compile_ocl_program(*kernel_map, cv,
kernel_source_str,
num_kerns-1,
kernel_names);
return cv;
}
void moveDataToGpu(cl_vars_t cv,
std::map<std::string, data_array> cpu_data,
std::map<std::string, cl_mem_data>* gpu_data,
kernel* kernels, int num_kerns)
{
std::map<std::string, data_array>::iterator it;
for (it = cpu_data.begin(); it!=cpu_data.end(); ++it)
{
int len = (*it).second.len;
double* arr = (*it).second.array;
cl_mem g_arr;
cl_int err = CL_SUCCESS;
g_arr = clCreateBuffer(cv.context,CL_MEM_READ_WRITE,
sizeof(double)*len, NULL,&err);
CHK_ERR(err);
err = clEnqueueWriteBuffer(cv.commands, g_arr, true, 0, sizeof(double)*len,
arr, 0, NULL, NULL);
CHK_ERR(err);
std::pair<std::string, cl_mem_data> map_elem;
map_elem.first = (*it).first;
cl_mem_data gpu_array = {g_arr, len};
map_elem.second = gpu_array;
(*gpu_data).insert(map_elem);
}
}
void runKernel(cl_vars_t cv,
cl_kernel cl_kern,
kernel kern,
std::map<std::string, cl_mem_data> gpu_data,
double* vars)
{
int num_arrays = kern.num_arrays;
string* arrays = kern.arrays;
int num_vars = kern.num_vars;
cl_int err = CL_SUCCESS;
//set up the out array
err = clSetKernelArg(cl_kern, 0, sizeof(cl_mem), &(gpu_data["out"].array));
CHK_ERR(err);
err = clSetKernelArg(cl_kern, 0 + num_arrays, sizeof(int), &(gpu_data["out"].len));
CHK_ERR(err);
for (int i = 1; i < num_arrays; i++)
{
std::string array_name = arrays[i];
err = clSetKernelArg(cl_kern, i, sizeof(cl_mem), &(gpu_data[array_name].array));
CHK_ERR(err);
err = clSetKernelArg(cl_kern, i + num_arrays, sizeof(int), &(gpu_data[array_name].len));
CHK_ERR(err);
}
for (int i = 0; i < num_vars; i++)
{
err = clSetKernelArg(cl_kern, i + 2*num_arrays, sizeof(double), &(vars[i]));
CHK_ERR(err);
}
size_t global_work_size[1] = {gpu_data[arrays[0]].len};
size_t local_work_size[1] = {256};
adjustWorkSize(global_work_size[0], local_work_size[0]);//pad work groups
global_work_size[0] = std::max(local_work_size[0], global_work_size[0]);
err = clEnqueueNDRangeKernel(cv.commands,
cl_kern,
1,
NULL,
global_work_size,
local_work_size,
0,
NULL,
NULL
);
CHK_ERR(err);
}
void releaseClMem(std::map<std::string, cl_mem> gpu_data) {
std::map<std::string, cl_mem>::iterator it;
for (it = gpu_data.begin(); it!=gpu_data.end(); ++it)
{
clReleaseMemObject((*it).second);
}
}