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
|
/*
* (C) Copyright 2024- ECMWF.
*
* This software is licensed under the terms of the Apache Licence Version 2.0
* which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
* In applying this licence, ECMWF does not waive the privileges and immunities
* granted to it by virtue of its status as an intergovernmental organisation
* nor does it submit to any jurisdiction.
*/
#include <algorithm>
#include <chrono>
#include <iostream>
#include <vector>
#include "hic/hic.h"
#include "pluto/pluto.h"
// ---------------------------------------------------------------------------------------------------
// Helper array types to allocate and deallocate uninitialized memory using allocator.
// This could be in a separate file
template <typename T, typename Allocator = pluto::allocator<T>>
class array {
public:
using value_type = T;
using allocator_type = Allocator;
array(std::size_t size): size_{size} { data_ = alloc_.allocate(size_); }
array(std::size_t size, pluto::stream_view stream): size_{size}, stream_(stream) {
data_ = alloc_.allocate_async(size_, stream_);
}
template <typename Alloc>
array(std::size_t size, const Alloc& alloc): alloc_{alloc}, size_{size} {
data_ = alloc_.allocate(size_);
}
template <typename Alloc>
array(std::size_t size, pluto::stream_view stream, const Alloc& alloc):
alloc_{alloc}, size_{size}, stream_(stream) {
data_ = alloc_.allocate_async(size_, stream_);
}
~array() {
if (not stream_.empty()) {
alloc_.deallocate_async(data_, size_, stream_);
}
else {
alloc_.deallocate(data_, size_);
}
}
void set_stream(pluto::stream_view stream) { stream_ = stream; }
value_type& operator[](std::size_t i) { return data_[i]; }
const value_type& operator[](std::size_t i) const { return data_[i]; }
value_type* data() { return data_; };
const value_type* data() const { return data_; }
std::size_t size() const { return size_; }
private:
allocator_type alloc_;
std::size_t size_;
value_type* data_{nullptr};
pluto::stream_view stream_;
};
// ---------------------------------------------------------------------------------------------------
// Kernel to add +1 to device array. This could be in a separate CUDA/HIP source file
#if PLUTO_HAVE_HIC
template <typename T>
HIC_GLOBAL void kernel_plus_one_on_device(T* d, int n) {
const int idx{int(blockDim.x) * int(blockIdx.x) + int(threadIdx.x)};
const int stride{int(blockDim.x) * int(gridDim.x)};
for (int i{idx}; i < n; i += stride) {
d[i] += 1.;
}
}
template <typename T>
void plus_one_on_device(T* d, int n, pluto::stream_view stream) {
if (pluto::devices()) {
const int threads_per_block{1024};
const int blocks_per_grid{32};
kernel_plus_one_on_device<<<blocks_per_grid, threads_per_block, 0, stream.value<hicStream_t>()>>>(d, n);
}
else {
for (int i = 0; i < n; ++i) {
d[i] += 1.;
}
}
}
#else
template <typename T>
void plus_one_on_device(T* d, int n, pluto::stream_view) {
for (int i = 0; i < n; ++i) {
d[i] += 1.;
}
}
#endif
// ---------------------------------------------------------------------------------------------------
int main(int argc, char* argv[]) {
std::cerr << "devices = " << pluto::devices() << std::endl;
// Choose data type
using value_type = float;
// Total problem size
std::size_t size = 1'000'000'000;
if (argc > 1) {
size = std::atoi(argv[1]);
}
std::cerr << "size = " << size << " (" << size * sizeof(value_type) / 1024. / 1024. / 1024. << " Gb)"
<< std::endl;
// Number of streams to divide problem size in
std::size_t nb_streams = 7;
if (argc > 2) {
nb_streams = std::atoi(argv[2]);
}
std::cerr << "streams = " << nb_streams << std::endl;
// A stream pool
std::vector<pluto::stream> streams(nb_streams);
std::cerr << "host alloc" << std::endl;
auto host_resource = pluto::pinned_resource();
array<value_type> array_h1(size, host_resource);
array<value_type> array_h2(size, host_resource);
std::cerr << "device alloc" << std::endl;
auto device_resource = pluto::device_resource();
array<value_type> array_d1(size, device_resource);
std::cerr << "async loop start" << std::endl;
auto start = std::chrono::steady_clock::now();
for (std::size_t jstream = 0; jstream < streams.size(); ++jstream) {
const auto& stream = streams[jstream];
{
const auto& stream_offset = jstream * size / streams.size();
const auto& stream_size = (jstream < streams.size() - 1 ? size / streams.size() : size - stream_offset);
array<value_type> stream_tmp(stream_size, stream, device_resource);
auto* h1 = array_h1.data() + stream_offset;
auto* h2 = array_h2.data() + stream_offset;
auto* d1 = array_d1.data() + stream_offset;
[[maybe_unused]] auto* dtmp = stream_tmp.data();
h1[stream_size - 1] = 1.;
h2[stream_size - 1] = -1.;
pluto::copy_host_to_device(d1, h1, stream_size, stream);
plus_one_on_device(d1, stream_size, stream);
pluto::copy_device_to_host(h2, d1, stream_size, stream);
}
//stream.wait();
}
std::cerr << "async loop end" << std::endl;
pluto::wait();
auto end = std::chrono::steady_clock::now();
std::cout << "execution without allocations took " << std::chrono::duration<double>(end - start).count() << " s"
<< std::endl;
if (array_h2[size - 1] != 2.) {
std::cerr << "ERROR: asynchronous execution not successful, h2[size-1] = " << array_h2[size - 1] << std::endl;
}
}
|
