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
|
#include "utest_helper.hpp"
#include "string.h"
static void cpu_exec (int n, float* src, float* dst)
{
int i = 0;
for (; i < n; i++) {
float f = src[i];
f = f < 0 ? -f : f;
dst[i] = f;
}
}
#define QUEUE_SECONDS_LIMIT 10
#define SUBMIT_SECONDS_LIMIT 20
#define COMMAND_SECONDS_LIMIT 10
static void check_profiling_time(cl_ulong queued, cl_ulong submit, cl_ulong start, cl_ulong end)
{
size_t profiling_resolution = 0;
OCL_CALL(clGetDeviceInfo, device, CL_DEVICE_PROFILING_TIMER_RESOLUTION,
sizeof(profiling_resolution), &profiling_resolution, NULL);
/* Convert the time to second. */
double queue_to_submit = (double)(submit - queued)*1e-9;
double submit_to_start = (double)(start - submit)*1e-9;
double start_to_end = (double)(end - start)*1e-9;
//printf("Profiling info:\n");
//printf("Time from queue to submit : %fms\n", (double)(queue_to_submit) * 1000.f );
//printf( "Time from submit to start : %fms\n", (double)(submit_to_start) * 1000.f );
//printf( "Time from start to end: %fms\n", (double)(start_to_end) * 1000.f );
OCL_ASSERTM(queued <= submit, "Enqueue time is later than submit time, invalid\n");
OCL_ASSERTM(submit <= start, "Submit time is later than start time, invalid\n");
OCL_ASSERTM(start <= end, "Start time is later than end time, invalid\n");
OCL_ASSERTM(queue_to_submit <= QUEUE_SECONDS_LIMIT, "Too large time from queue to submit\n");
OCL_ASSERTM(submit_to_start <= QUEUE_SECONDS_LIMIT, "Too large time from submit to start\n");
OCL_ASSERTM(start_to_end <= QUEUE_SECONDS_LIMIT, "Too large time from start to end\n");
}
static void profiling_exec(void)
{
const size_t n = 512;
cl_int status = CL_SUCCESS;
cl_command_queue profiling_queue = NULL;
float* cpu_src = (float *)malloc(n*sizeof(float));
float* cpu_dst = (float *)malloc(n*sizeof(float));
cl_event exec_event;
cl_ulong time_queue, time_submit, time_start, time_end;
/* Because the profiling prop, we can not use default queue. */
profiling_queue = clCreateCommandQueue(ctx, device, CL_QUEUE_PROFILING_ENABLE, &status);
OCL_ASSERT(status == CL_SUCCESS);
OCL_CREATE_KERNEL("compiler_fabs");
OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(float), NULL);
OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(float), NULL);
OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
globals[0] = n;
locals[0] = 256;
OCL_MAP_BUFFER(0);
for (int32_t i = 0; i < (int32_t) n; ++i)
cpu_src[i] = ((float*)buf_data[0])[i] = .1f * (rand() & 15) - .75f;
OCL_UNMAP_BUFFER(0);
cpu_exec(n, cpu_src, cpu_dst);
// Run the kernel on GPU
OCL_CALL(clEnqueueNDRangeKernel, profiling_queue, kernel, 1, NULL, globals, locals, 0, NULL, &exec_event);
OCL_CALL(clWaitForEvents, 1, &exec_event);
OCL_CALL(clGetEventProfilingInfo, exec_event, CL_PROFILING_COMMAND_QUEUED, sizeof(cl_ulong), &time_queue, NULL);
OCL_CALL(clGetEventProfilingInfo, exec_event, CL_PROFILING_COMMAND_SUBMIT, sizeof(cl_ulong), &time_submit, NULL);
OCL_CALL(clGetEventProfilingInfo, exec_event, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &time_start, NULL);
OCL_CALL(clGetEventProfilingInfo, exec_event, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &time_end, NULL);
clReleaseEvent(exec_event);
check_profiling_time(time_queue, time_submit, time_start, time_end);
// Compare
OCL_MAP_BUFFER(1);
for (int32_t i = 0; i < (int32_t) n; ++i)
OCL_ASSERT(((float *)buf_data[1])[i] == cpu_dst[i]);
OCL_UNMAP_BUFFER(1);
clReleaseCommandQueue(profiling_queue);
free(cpu_dst);
free(cpu_src);
}
MAKE_UTEST_FROM_FUNCTION(profiling_exec);
|
