-
Notifications
You must be signed in to change notification settings - Fork 37
/
test_fc_layer.cpp
205 lines (175 loc) · 6.79 KB
/
test_fc_layer.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
197
198
199
200
201
202
203
204
205
#include <stdio.h>
#include <iostream>
#include <algorithm>
#include "CL/opencl.h"
#include "AOCLUtils/aocl_utils.h"
#include "cnn_structs.h"
#include "data_utils.h"
#define FC_WG_SIZE 256
cl_platform_id platform = NULL;
unsigned num_devices = 0;
scoped_array<cl_device_id> devices;
cl_device_id target_device;
cl_context context = NULL;
cl_command_queue queue;
cl_program program = NULL;
cl_kernel kernel;
FcLayer fc;
aocl_utils::scoped_aligned_ptr<DTYPE> ref_output;
aocl_utils::scoped_aligned_ptr<DTYPE> h_input;
cl_mem d_input;
bool init_opencl();
void compute_reference();
void compare();
int main(int argc, char **argv) {
DataShape input_shape = {9216, 1, 1};
cl_int status;
size_t global_ws;
size_t local_ws;
init_opencl();
fc.bot_shape = &input_shape; fc.W = NULL; fc.b = NULL; fc.top_shape.z = 1;
fc.top_shape.x = 4096; fc.top_shape.y = 1;
fc.W = (WTYPE *)malloc(fc.bot_shape->x * fc.top_shape.x * sizeof(WTYPE));
fc.b = (WTYPE *)malloc(fc.top_shape.x * sizeof(WTYPE));
h_input.reset(fc.bot_shape->x);
fc.h_output.reset(fc.top_shape.x);
ref_output.reset(fc.top_shape.x);
CHECK_NEAR(1, 2);
// random input init
rand_init(h_input, fc.bot_shape->x, 0);
rand_init(fc.W, fc.top_shape.x * fc.bot_shape->x, 123);
rand_init(fc.b, fc.top_shape.x, 321);
d_input = clCreateBuffer(context, CL_MEM_WRITE_ONLY | CL_MEM_BANK_1_ALTERA | CL_MEM_COPY_HOST_PTR,
fc.bot_shape->x * sizeof(DTYPE), h_input, &status);
fc.d_input = &d_input;
fc.d_output = clCreateBuffer(context, CL_MEM_WRITE_ONLY | CL_MEM_BANK_1_ALTERA,
fc.top_shape.x * sizeof(DTYPE), NULL, &status);
fc.d_W = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_BANK_2_ALTERA | CL_MEM_COPY_HOST_PTR,
fc.bot_shape->x * fc.top_shape.x * sizeof(WTYPE), fc.W, &status);
fc.d_b = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_BANK_1_ALTERA | CL_MEM_COPY_HOST_PTR,
fc.top_shape.x * sizeof(WTYPE), fc.b, &status);
unsigned argi = 0;
status = clSetKernelArg(kernel, argi++, sizeof(cl_mem), fc.d_input);
checkError(status, "Failed to set argument %d", argi - 1);
status = clSetKernelArg(kernel, argi++, sizeof(cl_mem), &fc.d_W);
checkError(status, "Failed to set argument %d", argi - 1);
status = clSetKernelArg(kernel, argi++, sizeof(cl_mem), &fc.d_output);
checkError(status, "Failed to set argument %d", argi - 1);
status = clSetKernelArg(kernel, argi++, sizeof(int), &fc.bot_shape->x);
checkError(status, "Failed to set argument %d", argi - 1);
status = clSetKernelArg(kernel, argi++, sizeof(cl_mem), &fc.d_b);
checkError(status, "Failed to set argument %d", argi - 1);
const unsigned char act = 1;
status = clSetKernelArg(kernel, argi++, sizeof(unsigned char), &act);
checkError(status, "Failed to set argument %d", argi - 1);
global_ws = fc.top_shape.x;
local_ws = FC_WG_SIZE;
cl_event event;
std::cout << "Starting execution" << std::endl;
const double start_time = getCurrentTimestamp();
// launch the kernel
status = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global_ws, &local_ws, 0, NULL, &event);
checkError(status, "Failed to launch conv kernel");
clFinish(queue);
// Get kernel profiling info
cl_ulong start, end;
status = clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &start, NULL);
status |= clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &end, NULL);
checkError(status, "Error: could not get profile information");
clReleaseEvent(event);
std::cout << "Kernel time:" << (end-start) << std::endl;
const double end_time = getCurrentTimestamp();
const double total_time = (end_time - start_time)*1000;
std::cout << "Conv Layer Runtime(ms) = " << total_time << std::endl;
// read output from device buffer
status = clEnqueueReadBuffer(queue, fc.d_output, CL_TRUE, 0,
fc.top_shape.x * sizeof(DTYPE), fc.h_output, 0, NULL, NULL);
checkError(status, "Failed to read data from the device");
clFinish(queue);
std::cout << "Computing reference output" << std::endl;
// compute reference output and compare
//showMat<aocl_utils::scoped_aligned_ptr<DTYPE>& >(conv.h_output, conv.top_shape.z, conv.top_shape.y, conv.top_shape.x, 1);
compute_reference();
//showMat<aocl_utils::scoped_aligned_ptr<DTYPE>& >(ref_output, conv.top_shape.z, conv.top_shape.y, conv.top_shape.x, 1);
std::cout << "Comparing" << std::endl;
compare();
cleanup();
}
bool init_opencl() {
cl_int status;
printf("Initializing OpenCL\n");
if(!setCwdToExeDir()) {
return false;
}
// Get the OpenCL platform.
platform = findPlatform("Altera");
if(platform == NULL) {
printf("ERROR: Unable to find Altera OpenCL platform.\n");
return false;
}
// Query the available OpenCL device.
devices.reset(getDevices(platform, CL_DEVICE_TYPE_ALL, &num_devices));
printf("Platform: %s\n", getPlatformName(platform).c_str());
printf("Found %d devices in the board. Using only one device for this app\n", num_devices);
for(unsigned i = 0; i < num_devices; ++i) {
printf(" %s\n", getDeviceName(devices[i]).c_str());
}
target_device = devices[0];
// Create the context.
context = clCreateContext(NULL, num_devices, &target_device, &oclContextCallback, NULL, &status);
checkError(status, "Failed to create context");
std::string binary_file = getBoardBinaryFile("fc_kernel", target_device);
printf("Using AOCX: %s\n", binary_file.c_str());
program = createProgramFromBinary(context, binary_file.c_str(), &target_device, num_devices);
// Build the program that was just created.
status = clBuildProgram(program, 0, NULL, "", NULL, NULL);
checkError(status, "Failed to build program");
// Command queue.
queue = clCreateCommandQueue(context, target_device, CL_QUEUE_PROFILING_ENABLE, &status);
checkError(status, "Failed to create command queue");
// Kernel.
kernel = clCreateKernel(program, "fc_layer_relu", &status);
checkError(status, "Failed to create kernel");
std::cout << "OpenCL init done" << std::endl;
return true;
}
void compute_reference() {
DTYPE zero = 0.0f;
for(int out = 0; out < fc.top_shape.x; out++) {
DTYPE sum = 0.0f;
for(int in = 0; in < fc.bot_shape->x; in++) {
sum += h_input[in] * fc.W[out * fc.bot_shape->x + in];
}
sum += fc.b[out];
ref_output[out] = std::max(zero, sum);
}
}
void compare() {
float mse = 0.0f;
for(unsigned out = 0; out < fc.top_shape.x; out++) {
CHECK_NEAR(ref_output[out], fc.h_output[out]);
float diff = ref_output[out] - fc.h_output[out];
mse += diff * diff;
}
mse /= fc.top_shape.x;
std::cout << "MSE = " << mse << std::endl;
}
void cleanup() {
std::cout << "Releasing all OpenCL objects" << std::endl;
clReleaseKernel(kernel);
if(queue) {
clReleaseCommandQueue(queue);
}
if(program) {
clReleaseProgram(program);
}
if(context) {
clReleaseContext(context);
}
clReleaseMemObject(fc.d_output);
clReleaseMemObject(d_input);
clReleaseMemObject(fc.d_W);
clReleaseMemObject(fc.d_b);
free(fc.W);
free(fc.b);
}