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
206
207
208
|
#include <CL/sycl.hpp>
#include <omp.h>
#include <typeinfo>
#include <memory>
#include <cxxabi.h>
#define N 64
#define GZ 16
#define STRIDE (2*GZ + N)
#define TILEK 4
#define TILEJ 4
#define TILEI 16
#define ITER 10
std::string demangle(const char *name) {
int status = -4; // some arbitrary value to eliminate the compiler warning
// enable c++11 by passing the flag -std=c++11 to g++
std::unique_ptr<char, void (*)(void *)> res{
abi::__cxa_demangle(name, NULL, NULL, &status),
std::free
};
return (status == 0) ? res.get() : name;
}
using namespace cl::sycl;
template<typename T>
class subgr;
template<typename T>
class SGfunctor {
public:
SGfunctor(accessor<uint32_t, 1, access::mode::write, access::target::global_buffer> sg_sz,
accessor<T, 1, access::mode::write, access::target::global_buffer> sg_i) : sg_sz(sg_sz), sg_i(sg_i) {}
[[cl::intel_reqd_sub_group_size(16)]]
void operator()(nd_item<1> NdItem) {
intel::sub_group SG = NdItem.get_sub_group();
uint32_t wggid = NdItem.get_global_id(0);
uint32_t sgid = SG.get_local_id().get(0);
if (wggid == 0)
sg_sz[0] = SG.get_max_local_range()[0];
sgid = SG.shuffle_up(sgid, 2);
sg_i[wggid] = sgid;
}
private:
accessor<T, 1, access::mode::write, access::target::global_buffer> sg_i;
accessor<uint32_t, 1, access::mode::write, access::target::global_buffer> sg_sz;
};
template<typename T>
void subkrnl(device &Device, size_t G = 256, size_t L = 64) {
buffer<uint32_t> sg_buf{1};
buffer<T> sg_info{G};
queue Queue(Device);
nd_range<1> NumOfWorkItems(G, L);
auto kernel = [&](handler &cgh) -> void {
auto sg_sz = sg_buf.template get_access<access::mode::write>(cgh);
auto sg_i = sg_info.template get_access<access::mode::write>(cgh);
SGfunctor<T> sGfunctor(sg_sz, sg_i);
cgh.parallel_for<subgr<T>>(NumOfWorkItems, sGfunctor);
};
Queue.submit(kernel);
Queue.wait();
auto sg_sz = sg_buf.template get_access<access::mode::read>();
auto sg_i = sg_info.template get_access<access::mode::read>();
std::cout << "Subgroup size for " << demangle(typeid(T).name()) << ": " << sg_sz[0] << std::endl;
// for (int i = 0; i < sg_sz[0]; ++i)
// std::cout << sg_i[i] << " ";
// std::cout << std::endl;
}
void runSubgroups(device &Device) {
std::cout << "Running subgroup for different types" << std::endl;
subkrnl<int>(Device);
subkrnl<long>(Device);
subkrnl<float>(Device);
}
void run27pt(device &Device) {
// Creating buffer of 1024 ints to be used inside the kernel code
buffer<float, 3> in_buf{range<3>(STRIDE, STRIDE, STRIDE)};
buffer<float, 3> out_buf{range<3>(STRIDE, STRIDE, STRIDE)};
buffer<float, 1> c_buf{27};
// Creating SYCL queue
queue Queue(Device, {property::queue::enable_profiling()});
nd_range<3> NumOfWorkItems(range<3>(N, N, N), range<3>(TILEI, TILEJ, TILEK));
float st = omp_get_wtime();
auto kernel = [&](handler &cgh) {
// Getting write only access to the buffer on a device
auto in = in_buf.get_access<access::mode::read>(cgh);
auto out = out_buf.get_access<access::mode::write>(cgh);
auto c = c_buf.get_access<access::mode::read>(cgh);
// Executing kernel
cgh.parallel_for<class FillBuffer>(
NumOfWorkItems, [=](nd_item<3> WIid) {
uint32_t i = WIid.get_global_id(0) + GZ;
uint32_t j = WIid.get_global_id(1) + GZ;
uint32_t k = WIid.get_global_id(2) + GZ;
// Fill buffer with indexes
out[id<3>(i, j, k)] = c[0] * in[id<3>(i - 1, j - 1, k - 1)]
+ c[1] * in[id<3>(i, j - 1, k - 1)]
+ c[2] * in[id<3>(i + 1, j - 1, k - 1)]
+ c[3] * in[id<3>(i - 1, j, k - 1)]
+ c[4] * in[id<3>(i, j, k - 1)]
+ c[5] * in[id<3>(i + 1, j, k - 1)]
+ c[6] * in[id<3>(i - 1, j + 1, k - 1)]
+ c[7] * in[id<3>(i, j + 1, k - 1)]
+ c[8] * in[id<3>(i + 1, j + 1, k - 1)]
+ c[9] * in[id<3>(i - 1, j - 1, k)]
+ c[10] * in[id<3>(i, j - 1, k)]
+ c[11] * in[id<3>(i + 1, j - 1, k)]
+ c[12] * in[id<3>(i - 1, j, k)]
+ c[13] * in[id<3>(i, j, k)]
+ c[14] * in[id<3>(i + 1, j, k)]
+ c[15] * in[id<3>(i - 1, j + 1, k)]
+ c[16] * in[id<3>(i, j + 1, k)]
+ c[17] * in[id<3>(i + 1, j + 1, k)]
+ c[18] * in[id<3>(i - 1, j - 1, k + 1)]
+ c[19] * in[id<3>(i, j - 1, k + 1)]
+ c[20] * in[id<3>(i + 1, j - 1, k + 1)]
+ c[21] * in[id<3>(i - 1, j, k + 1)]
+ c[22] * in[id<3>(i, j, k + 1)]
+ c[23] * in[id<3>(i + 1, j, k + 1)]
+ c[24] * in[id<3>(i - 1, j + 1, k + 1)]
+ c[25] * in[id<3>(i, j + 1, k + 1)]
+ c[26] * in[id<3>(i + 1, j + 1, k + 1)];
});
};
auto st_event = Queue.submit(kernel);
for (int i = 0; i < ITER - 2; ++i) {
// Submitting command group(work) to queue
Queue.submit(kernel);
}
auto ed_event = Queue.submit(kernel);
ed_event.wait();
// Getting read only access to the buffer on the host.
// Implicit barrier waiting for queue to complete the work.
const auto out_h = out_buf.get_access<access::mode::read>();
const auto in_h = in_buf.get_access<access::mode::read>();
double ed = omp_get_wtime();
double elapsed = (ed_event.get_profiling_info<info::event_profiling::command_end>() -
st_event.get_profiling_info<info::event_profiling::command_start>()) * 1e-9;
std::cout << "elapsed: " << (ed - st) / ITER << std::endl;
std::cout << "elapsed: " << elapsed << std::endl;
std::cout << "flops: " << N * N * N * 53.0 * ITER / elapsed * 1e-9 << std::endl;
}
void printInfo(device &Device) {
std::cout << "Using OpenCL device {" << Device.get_info<info::device::name>() <<
"} from {" << Device.get_info<info::device::vendor>() << "}" << std::endl;
std::cout << "I'm a " << (Device.is_cpu() ? "CPU" : "GPU") << std::endl;
auto dot_num_groups = Device.get_info<info::device::max_compute_units>();
auto dot_wgsize = Device.get_info<info::device::max_work_group_size>();
auto max_num_sg = Device.get_info<info::device::max_num_sub_groups>();
std::cout << "Compute units: " << dot_num_groups << std::endl;
std::cout << "Workgroup size: " << dot_wgsize << std::endl;
std::cout << "Maximum subgroup size: " << max_num_sg << std::endl;
}
int main() {
// This is an example of using a device_selector
// When instantiate a device use device(NEOGPUDeviceSelector)
class NEOGPUDeviceSelector : public device_selector {
public:
int operator()(const device &Device) const override {
const std::string DeviceName = Device.get_info<info::device::name>();
const std::string DeviceVendor = Device.get_info<info::device::vendor>();
return Device.is_gpu() && DeviceName.find("HD Graphics NEO") ? 1 : -1;
}
};
int DeviceNumber = 0;
for (device &Device : device::get_devices()) {
// Host device is not for compute
if (Device.is_host())
continue;
std::cout << "Device [" << DeviceNumber << "]:" << std::endl;
try {
printInfo(Device);
if (Device.is_gpu()) {
run27pt(Device);
runSubgroups(Device);
} else {
std::cout << "CPU devices possibly not supported" << std::endl;
}
} catch (invalid_parameter_error &E) {
std::cout << E.what() << std::endl;
}
std::cout << "==================" << std::endl;
++DeviceNumber;
}
}
|
