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
|
#include <random>
#include <set>
#include <purify/AllToAllSparseVector.h>
#include <purify/DistributeSparseVector.h>
#include "catch2/catch_all.hpp"
#include <sopt/mpi/communicator.h>
using namespace purify;
TEST_CASE("Distribe Sparse Vector") {
auto const world = sopt::mpi::Communicator::World();
Vector<t_int> const grid = world.broadcast<Vector<t_int>>(
Vector<t_int>::Random(std::max(world.size() * 2, world.size() + 2)));
std::vector<t_int> const indices = {static_cast<t_int>(world.rank()),
static_cast<t_int>(world.size() + 1)};
DistributeSparseVector distributor(indices, grid.size(), world);
SECTION("Scatter") {
Vector<t_int> output;
distributor.scatter(grid, output);
REQUIRE(output.size() == 2);
CHECK(output(0) == grid(world.rank()));
CHECK(output(1) == grid(world.size() + 1));
}
SECTION("Gather") {
Vector<t_int> local(2);
local << world.rank(), world.size() + world.rank();
Vector<t_int> output;
if (world.is_root()) {
distributor.gather(local, output);
CHECK(output.size() == grid.size());
for (decltype(world.size()) i(0); i < world.size(); ++i) CHECK(output(i) == i);
CHECK(output(world.size()) == 0);
CHECK(output(world.size() + 1) ==
world.size() * world.size() + (world.size() * (world.size() - 1)) / 2);
CHECK(grid.size() - 2 >= world.size());
CHECK((output.tail(grid.size() - world.size() - 2).array() == 0).all());
} else {
distributor.gather(local);
}
}
}
TEST_CASE("All to one Sparse Vector") {
auto const world = sopt::mpi::Communicator::World();
Vector<t_int> const grid = world.broadcast<Vector<t_int>>(
Vector<t_int>::Random(std::max(world.size() * 2, world.size() + 2)));
std::vector<t_int> const indices = {static_cast<t_int>(world.rank()),
static_cast<t_int>(world.size() + 1)};
AllToAllSparseVector<t_int> distributor(indices, grid.size(), grid.size() * world.rank(), world);
SECTION("Scatter") {
Vector<t_int> output;
distributor.recv_grid(grid, output);
CAPTURE(world.rank());
REQUIRE(output.size() == 2);
CHECK(output(0) == grid(world.rank()));
CHECK(output(1) == grid(world.size() + 1));
}
SECTION("Gather") {
Vector<t_int> local(2);
local << world.rank(), world.size() + world.rank();
Vector<t_int> output;
distributor.send_grid(local, output);
CHECK(output.size() == grid.size());
if (world.is_root()) {
for (decltype(world.size()) i(0); i < world.size(); ++i) CHECK(output(i) == i);
CHECK(output(world.size()) == 0);
CHECK(output(world.size() + 1) ==
world.size() * world.size() + (world.size() * (world.size() - 1)) / 2);
CHECK(grid.size() - 2 >= world.size());
CHECK((output.tail(grid.size() - world.size() - 2).array() == 0).all());
} else {
for (t_int i = 0; i < output.size(); i++) CHECK(output(i) == 0);
}
}
}
TEST_CASE("All to All Sparse Vector") {
auto const world = sopt::mpi::Communicator::World();
CAPTURE(world.rank());
Vector<t_int> const grid = world.broadcast<Vector<t_int>>(
Vector<t_int>::Random(std::max(world.size() * 2, world.size() + 2)));
SECTION("Check throw when index is not ordered by node") {
std::vector<t_int> const indices = {
static_cast<t_int>(grid.size() * (world.rank() + 1) + world.size() + 1),
static_cast<t_int>(world.rank())};
CHECK_THROWS(
AllToAllSparseVector<t_int>(indices, grid.size(), grid.size() * world.rank(), world));
}
std::vector<t_int> const indices = {
static_cast<t_int>(world.rank()),
static_cast<t_int>(grid.size() * world.rank() + world.size() + 1)};
AllToAllSparseVector<t_int> distributor(indices, grid.size(), grid.size() * world.rank(), world);
SECTION("Scatter") {
Vector<t_int> output;
distributor.recv_grid(grid, output);
REQUIRE(output.size() == 2);
CHECK(output(0) == grid(world.rank()));
CHECK(output(1) == grid(world.size() + 1));
}
SECTION("Gather") {
Vector<t_int> local = Vector<t_int>::Ones(2);
Vector<t_int> output;
distributor.send_grid(local, output);
CHECK(output.size() == grid.size());
CAPTURE(world.rank());
CAPTURE(output);
if (world.is_root()) {
for (decltype(world.size()) i(0); i < world.size(); ++i) CHECK(output(i) == 1);
CHECK(output(world.size()) == 0);
CHECK(output(world.size() + 1) == 1);
CHECK(grid.size() - 2 >= world.size());
CHECK((output.tail(grid.size() - world.size() - 2).array() == 0).all());
} else {
for (t_int i = 0; i < output.size(); i++) {
if (i != world.size() + 1)
CHECK(output(i) == 0);
else
CHECK(output(i) == 1);
}
}
}
}
TEST_CASE("recv_sizes") {
for (t_int nodes : {1, 2, 5, 10, 20, 50, 100, 1000}) {
for (t_int imsize : {128, 1024, 2048, 4096, 8192, 16384, 32768}) {
const std::int32_t N = imsize * imsize;
CAPTURE(imsize);
CAPTURE(N);
// First create an instance of an engine.
std::random_device rnd_device;
// Specify the engine and distribution.
std::mt19937_64 mersenne_engine(rnd_device()); // Generates random integers
std::uniform_int_distribution<t_int> dist(0, nodes * N);
auto gen = [&dist, &mersenne_engine]() { return dist(mersenne_engine); };
std::vector<t_int> local_indices(10);
std::generate(local_indices.begin(), local_indices.end(), gen);
std::sort(local_indices.begin(), local_indices.end(),
[](t_int a, t_int b) { return (a < b); });
CAPTURE(local_indices);
CAPTURE(nodes);
if (static_cast<std::int64_t>(N) * static_cast<std::int64_t>(nodes) >
std::numeric_limits<t_int>::max())
CHECK_THROWS(all_to_all_recv_sizes<t_int>(local_indices, nodes, N));
else {
std::vector<t_int> recv = all_to_all_recv_sizes<t_int>(local_indices, nodes, N);
for (const auto& a : recv) CHECK(a >= 0);
CHECK(local_indices.size() == std::accumulate(recv.begin(), recv.end(), 0));
}
}
}
}
TEST_CASE("recv_sizes 64 bit") {
for (std::int64_t nodes : {1, 2, 5, 10, 20, 50, 100, 1000}) {
for (std::int64_t imsize : {128, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072}) {
const std::int64_t N = imsize * imsize;
CAPTURE(imsize);
// First create an instance of an engine.
std::random_device rnd_device;
// Specify the engine and distribution.
std::mt19937_64 mersenne_engine(rnd_device()); // Generates random integers
std::uniform_int_distribution<std::int64_t> dist(
0, static_cast<std::int64_t>(nodes) * static_cast<std::int64_t>(N));
auto gen = [&dist, &mersenne_engine]() -> std::int64_t { return dist(mersenne_engine); };
std::vector<std::int64_t> local_indices(10);
std::generate(local_indices.begin(), local_indices.end(), gen);
std::sort(local_indices.begin(), local_indices.end(),
[](std::int64_t a, std::int64_t b) { return (a < b); });
CAPTURE(local_indices);
CAPTURE(nodes);
std::vector<t_int> recv = all_to_all_recv_sizes<std::int64_t>(local_indices, nodes, N);
for (const auto& a : recv) CHECK(a >= 0);
CHECK(local_indices.size() == std::accumulate(recv.begin(), recv.end(), 0));
}
}
}
|
