1/*
2 * VCluster_semantic_unit_tests_funcs.hpp
3 *
4 * Created on: Aug 18, 2018
5 * Author: i-bird
6 */
7
8#ifndef VCLUSTER_SEMANTIC_UNIT_TESTS_FUNCS_HPP_
9#define VCLUSTER_SEMANTIC_UNIT_TESTS_FUNCS_HPP_
10
11#include "VCluster/VCluster.hpp"
12
13template<typename Memory, template<typename> class layout_base>
14void test_different_layouts()
15{
16 for (size_t i = 0 ; i < 100 ; i++)
17 {
18 Vcluster<> & vcl = create_vcluster();
19
20 if (vcl.getProcessingUnits() >= 32)
21 return;
22
23 openfpm::vector<aggregate<int,float,size_t>,Memory,layout_base> v1;
24 v1.resize(vcl.getProcessUnitID());
25
26 for(size_t j = 0 ; j < vcl.getProcessUnitID() ; j++)
27 {
28 v1.template get<0>(j) = 5;
29 v1.template get<1>(j) = 10.0+1000.0;
30 v1.template get<2>(j) = 11.0+100000;
31 }
32
33 openfpm::vector<aggregate<int,float,size_t>,Memory,layout_base> v2;
34
35 vcl.SGather<decltype(v1),decltype(v2),layout_base>(v1,v2,(i%vcl.getProcessingUnits()));
36
37 if (vcl.getProcessUnitID() == (i%vcl.getProcessingUnits()))
38 {
39 size_t n = vcl.getProcessingUnits();
40 BOOST_REQUIRE_EQUAL(v2.size(),n*(n-1)/2);
41
42 bool is_correct = true;
43 for (size_t i = 0 ; i < v2.size() ; i++)
44 {
45 is_correct &= (v2.template get<0>(i) == 5);
46 is_correct &= (v2.template get<1>(i) == 10.0+1000.0);
47 is_correct &= (v2.template get<2>(i) == 11.0+100000.0);
48 }
49
50 BOOST_REQUIRE_EQUAL(is_correct,true);
51 }
52 if (vcl.getProcessUnitID() == 0 && i == 99)
53 std::cout << "Semantic gather test stop" << std::endl;
54 }
55}
56
57template<typename Memory>
58void test_ssend_recv_layout_switch(size_t opt)
59{
60 auto & v_cl = create_vcluster<Memory>();
61
62 if (v_cl.size() > 10) {return;}
63
64 openfpm::vector<openfpm::vector_gpu_single<aggregate<float,float[3]>>> vd;
65 openfpm::vector_gpu<aggregate<float,float[3]>> collect;
66 openfpm::vector_gpu<aggregate<float,float[3]>> collect2;
67 openfpm::vector<size_t> prc_send;
68 openfpm::vector<size_t> prc_recv;
69 openfpm::vector<size_t> sz_recv;
70
71 vd.resize(v_cl.size());
72
73 for (size_t i = 0 ; i < vd.size() ; i++)
74 {
75 vd.get(i).resize(100);
76
77 for (size_t j = 0 ; j < vd.get(i).size() ; j++)
78 {
79 vd.get(i).template get<0>(j) = 10000*i + v_cl.rank()*100 + j;
80
81 vd.get(i).template get<1>(j)[0] = 400000 + 10000*i + v_cl.rank()*100 + j;
82 vd.get(i).template get<1>(j)[1] = 400000 + 10000*i + v_cl.rank()*100 + j;
83 vd.get(i).template get<1>(j)[2] = 400000 + 10000*i + v_cl.rank()*100 + j;
84 }
85
86 prc_send.add(i);
87
88 if (opt & MPI_GPU_DIRECT)
89 {
90 vd.get(i).template hostToDevice<0,1>();
91
92 // Reset host
93
94 for (size_t j = 0 ; j < vd.get(i).size() ; j++)
95 {
96 vd.get(i).template get<0>(j) = 0.0;
97
98 vd.get(i).template get<1>(j)[0] = 0.0;
99 vd.get(i).template get<1>(j)[1] = 0.0;
100 vd.get(i).template get<1>(j)[2] = 0.0;
101 }
102 }
103 }
104
105 v_cl.template SSendRecv<openfpm::vector_gpu_single<aggregate<float,float[3]>>,decltype(collect),memory_traits_inte>
106 (vd,collect,prc_send, prc_recv,sz_recv,opt);
107
108 v_cl.template SSendRecvP<openfpm::vector_gpu_single<aggregate<float,float[3]>>,decltype(collect),memory_traits_inte,0,1>
109 (vd,collect2,prc_send, prc_recv,sz_recv,opt);
110
111 // collect must have 100 * v_cl.size()
112
113 BOOST_REQUIRE_EQUAL(collect.size(),100*v_cl.size());
114 BOOST_REQUIRE_EQUAL(collect2.size(),100*v_cl.size());
115
116 // we reset the host collected data if data must be on device
117
118 if (opt & MPI_GPU_DIRECT)
119 {
120 for (size_t j = 0 ; j < collect.size() ; j++)
121 {
122 collect.template get<0>(j) = 0.0;
123
124 collect.template get<1>(j)[0] = 0.0;
125 collect.template get<1>(j)[1] = 0.0;
126 collect.template get<1>(j)[2] = 0.0;
127
128 collect2.template get<0>(j) = 0.0;
129
130 collect2.template get<1>(j)[0] = 0.0;
131 collect2.template get<1>(j)[1] = 0.0;
132 collect2.template get<1>(j)[2] = 0.0;
133 }
134 }
135
136 // from device to host
137
138 if (opt & MPI_GPU_DIRECT)
139 {
140 collect.template deviceToHost<0,1>();
141 collect2.template deviceToHost<0,1>();
142 }
143
144 // now we check what we received
145
146 // check what we received
147
148 bool match = true;
149 for (size_t i = 0 ; i < v_cl.size() ; i++)
150 {
151 for (size_t j = 0 ; j < 100 ; j++)
152 {
153 match &= collect.template get<0>(i*100 +j) == v_cl.rank()*10000 + i*100 + j;
154
155 match &= collect.template get<1>(i*100 +j)[0] == 400000 + v_cl.rank()*10000 + i*100 + j;
156 match &= collect.template get<1>(i*100 +j)[1] == 400000 + v_cl.rank()*10000 + i*100 + j;
157 match &= collect.template get<1>(i*100 +j)[2] == 400000 + v_cl.rank()*10000 + i*100 + j;
158
159 match &= collect2.template get<0>(i*100 +j) == v_cl.rank()*10000 + i*100 + j;
160
161 match &= collect2.template get<1>(i*100 +j)[0] == 400000 + v_cl.rank()*10000 + i*100 + j;
162 match &= collect2.template get<1>(i*100 +j)[1] == 400000 + v_cl.rank()*10000 + i*100 + j;
163 match &= collect2.template get<1>(i*100 +j)[2] == 400000 + v_cl.rank()*10000 + i*100 + j;
164 }
165
166 if (match == false){break;}
167 }
168
169 BOOST_REQUIRE_EQUAL(match,true);
170}
171
172#endif /* VCLUSTER_SEMANTIC_UNIT_TESTS_FUNCS_HPP_ */
173