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
|
#include "./mixed_precision.hpp"
#include <forstio/codec/schema.hpp>
#include <sstream>
int main(int argc, char** argv){
using namespace saw;
uint64_t start_test_size = 1024ul * 1024ul;
if(argc <= 0 || argc >= 256){
std::cerr<<"Argument size being weird. Got "<<argc<<" args"<<std::endl;
return -1;
}
std::vector<std::string_view> args;
args.resize(static_cast<uint64_t>(argc));
for(uint64_t i = 0; i < args.size(); ++i){
args.at(i) = {argv[i]};
}
if(args.size() > 1){
auto& str = args.at(1);
auto ec = std::from_chars(str.data(), str.data() + str.size(), start_test_size);
if(ec.ec != std::errc{}){
std::cerr<<"Start size is not an int."<<std::endl;
return -1;
}
}
uint64_t max_test_size = start_test_size * 1024ul;
if(args.size() > 2){
auto& str = args.at(2);
auto ec = std::from_chars(str.data(), str.data() + str.size(), max_test_size);
if(ec.ec != std::errc{}){
std::cerr<<"Stop size is not an int."<<std::endl;
return -1;
}
}
if(start_test_size > max_test_size){
std::cerr<<"Invalid arguments. Stop size is smaller than Start size."<<std::endl;
return -1;
}
uint64_t runs = 128ul;
if(args.size() > 3){
auto& str = args.at(3);
auto ec = std::from_chars(str.data(), str.data() + str.size(), runs);
if(ec.ec != std::errc{}){
std::cerr<<"Run size is not an int."<<std::endl;
return -1;
}
}
uint64_t arithmetic_intensity = 1u;
if(args.size() > 4){
auto& str = args.at(4);
auto ec = std::from_chars(str.data(), str.data() + str.size(), arithmetic_intensity);
if(ec.ec != std::errc{}){
std::cerr<<"Arithmetic intensity is not an int."<<std::endl;
return -1;
}
}
std::random_device r;
std::default_random_engine e1{r()};
std::uniform_real_distribution<> dis{-3.0,-1.0};
saw::event_loop loop;
saw::wait_scope wait{loop};
remote<rmt::Sycl> rmt;
own<remote_address<rmt::Sycl>> rmt_addr{};
rmt.resolve_address().then([&](auto addr){
rmt_addr = std::move(addr);
}).detach();
wait.poll();
if(!rmt_addr){
return -1;
}
cl::sycl::event mixed_ev;
cl::sycl::event float32_ev;
cl::sycl::event float64_ev;
auto sycl_iface = listen_mixed_precision(mixed_ev, float64_ev, float32_ev, arithmetic_intensity);
data<sch::MixedArray> mixed_host_data;
data<sch::Float64Array> float64_host_data;
data<sch::Float32Array> float32_host_data;
auto time_eval = [](uint64_t & current_min_time, cl::sycl::event& ev){
auto end = ev.get_profiling_info<cl::sycl::info::event_profiling::command_end>();
auto start = ev.get_profiling_info<cl::sycl::info::event_profiling::command_start>();
uint64_t curr_time = (end-start);
current_min_time = std::min(curr_time, current_min_time);
};
auto our_device = share<device<rmt::Sycl>>();
auto& device = *our_device;
/**
* Warmup
*/
std::cout<<"Warming up ..."<<std::endl;
for(uint64_t test_size = 1ul; test_size < max_test_size; test_size *= 2ul){
mixed_host_data = {test_size};
float64_host_data = {test_size};
float32_host_data = {test_size};
for(uint64_t i = 0; i < test_size; ++i){
double gen_num = dis(e1);
mixed_host_data.at(i) = static_cast<double>(gen_num);
float64_host_data.at(i) = static_cast<double>(gen_num);
float32_host_data.at(i) = static_cast<float>(gen_num);
}
data<sch::MixedArray, encode::Sycl<encode::Native>> mixed_device_data{mixed_host_data};
data<sch::Float64Array, encode::Sycl<encode::Native>> float64_device_data{float64_host_data};
data<sch::Float32Array, encode::Sycl<encode::Native>> float32_device_data{float32_host_data};
sycl_iface.template call<"float64_32">(mixed_device_data, &(device.get_handle()));
sycl_iface.template call<"float64">(float64_device_data, &(device.get_handle()));
sycl_iface.template call<"float32">(float32_device_data, &(device.get_handle()));
device.get_handle().wait();
}
std::cout<<"Benchmark starting ...";
/**
* Benchmark
*/
std::stringstream sstr;
for(uint64_t test_size = start_test_size; test_size <= max_test_size; test_size *= 2ul){
uint64_t time_mixed = std::numeric_limits<uint64_t>::max();
uint64_t time_float64 = std::numeric_limits<uint64_t>::max();
uint64_t time_float32 = std::numeric_limits<uint64_t>::max();
for(uint64_t runs_i = 0u; runs_i < runs; ++runs_i){
(std::cout<<'.').flush();
data<sch::MixedArray> mixed_host_data;
data<sch::Float64Array> float64_host_data;
data<sch::Float32Array> float32_host_data;
mixed_host_data = {test_size};
float64_host_data = {test_size};
float32_host_data = {test_size};
for(uint64_t i = 0; i < test_size; ++i){
double gen_num = dis(e1);
mixed_host_data.at(i) = static_cast<double>(gen_num);
float64_host_data.at(i) = static_cast<double>(gen_num);
float32_host_data.at(i) = static_cast<float>(gen_num);
}
data<sch::MixedArray, encode::Sycl<encode::Native>> mixed_device_data{mixed_host_data};
data<sch::Float64Array, encode::Sycl<encode::Native>> float64_device_data{float64_host_data};
data<sch::Float32Array, encode::Sycl<encode::Native>> float32_device_data{float32_host_data};
sycl_iface.template call<"float64_32">(mixed_device_data, &(device.get_handle()));
device.get_handle().wait();
time_eval(time_mixed, mixed_ev);
sycl_iface.template call<"float64">(float64_device_data, &(device.get_handle()));
device.get_handle().wait();
time_eval(time_float64, float64_ev);
sycl_iface.template call<"float32">(float32_device_data, &(device.get_handle()));
device.get_handle().wait();
time_eval(time_float32, float32_ev);
}
sstr<<test_size;
sstr<<",\t";
sstr<<time_mixed / 1.0e9;
sstr<<",\t";
sstr<<time_float64 / 1.0e9;
sstr<<",\t";
sstr<<time_float32 / 1.0e9;
sstr<<"\n";
}
std::cout<<'\n'<<'\n'<<sstr.str()<<std::endl;
return 0;
}
|
