ReductionType map_reduce_vertices(GraphType& g,
MapFunctionType mapfunction,
const vertex_set& vset = GraphType::complete_set()) {
BOOST_CONCEPT_ASSERT((graphlab::Serializable<ReductionType>));
BOOST_CONCEPT_ASSERT((graphlab::OpPlusEq<ReductionType>));
typedef typename GraphType::vertex_type vertex_type;
if(!g.is_finalized()) {
logstream(LOG_FATAL)
<< "\n\tAttempting to run graph.map_reduce_vertices(...) "
<< "\n\tbefore calling graph.finalize()."
<< std::endl;
}
g.dc().barrier();
bool global_result_set = false;
ReductionType global_result = ReductionType();
#ifdef _OPENMP
#pragma omp parallel
#endif
{
bool result_set = false;
ReductionType result = ReductionType();
#ifdef _OPENMP
#pragma omp for
#endif
for (int i = 0; i < (int)g.num_local_vertices(); ++i) {
auto lvertex = g.l_vertex(i);
if (lvertex.owned() && vset.l_contains(lvid_type(i))) {
if (!result_set) {
vertex_type vtx(lvertex);
result = mapfunction(vtx);
result_set = true;
}
else if (result_set){
const vertex_type vtx(lvertex);
const ReductionType tmp = mapfunction(vtx);
result += tmp;
}
}
}
#ifdef _OPENMP
#pragma omp critical
#endif
{
if (result_set) {
if (!global_result_set) {
global_result = result;
global_result_set = true;
}
else {
global_result += result;
}
}
}
}
conditional_addition_wrapper<ReductionType>
wrapper(global_result, global_result_set);
g.dc().all_reduce(wrapper);
return wrapper.value;
} // end of map_reduce_vertices
bool load_binary(GraphType& g, const std::string& prefix) {
g.dc().full_barrier();
std::string fname = prefix + tostr(g.procid()) + ".bin";
logstream(LOG_INFO) << "Load graph from " << fname << std::endl;
general_ifstream fin(fname, true);
if(!fin.good()) {
logstream(LOG_ERROR) << "\n\tError opening file: " << fname << std::endl;
return false;
}
iarchive iarc(fin);
iarc >> g;
logstream(LOG_INFO) << "Finish loading graph from " << fname << std::endl;
g.dc().full_barrier();
return true;
} // end of load
void load(GraphType& g, std::string prefix,
line_parser_type<GraphType> line_parser) {
if (prefix.length() == 0)
return;
g.dc().full_barrier();
g.clear();
std::string directory_name; std::string original_path(prefix);
boost::filesystem::path path(prefix);
std::string search_prefix;
if (boost::filesystem::is_directory(path)) {
// if this is a directory
// force a "/" at the end of the path
// make sure to check that the path is non-empty. (you do not
// want to make the empty path "" the root path "/" )
directory_name = path.native();
} else {
directory_name = path.parent_path().native();
search_prefix = path.filename().native();
directory_name = (directory_name.empty() ? "." : directory_name);
}
std::vector<std::string> graph_files;
fs_util::list_files_with_prefix(directory_name, search_prefix, graph_files);
if (graph_files.size() == 0) {
logstream(LOG_WARNING) << "No files found matching " << original_path << std::endl;
}
parallel_for(0, graph_files.size(), [&](size_t i) {
if (i % g.numprocs() == g.procid()) {
logstream(LOG_EMPH) << "Loading graph from file: " << graph_files[i] << std::endl;
general_ifstream fin(graph_files[i]);
if(!fin.good()) {
log_and_throw_io_failure("Cannot open file: " + graph_files[i]);
}
const bool success = load_from_stream(g, graph_files[i], fin, line_parser);
if(!success) {
log_and_throw_io_failure("Fail parsing file: " + graph_files[i]);
}
}
});
g.dc().full_barrier();
g.finalize();
} // end of load
void parallel_for_edges(GraphType& g,
std::vector<EdgeFunctorType>& accfunction) {
typedef typename GraphType::local_edge_type local_edge_type;
typedef typename GraphType::edge_type edge_type;
ASSERT_TRUE(g.is_finalized());
g.dc().barrier();
int numaccfunctions = (int)accfunction.size();
ASSERT_GE(numaccfunctions, 1);
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < (int)accfunction.size(); ++i) {
for (int j = i;j < (int)g.num_local_vertices(); j+=numaccfunctions) {
for (const local_edge_type& e : g.l_vertex(j).in_edges()) {
accfunction[i](edge_type(e));
}
}
}
g.dc().barrier();
}
void parallel_for_vertices(GraphType& g,
std::vector<VertexFunctorType>& accfunction) {
typedef typename GraphType::vertex_type vertex_type;
ASSERT_TRUE(g.is_finalized());
g.dc().barrier();
int numaccfunctions = (int)accfunction.size();
ASSERT_GE(numaccfunctions, 1);
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (int i = 0; i < (int)accfunction.size(); ++i) {
for (int j = i;j < (int)g.num_local_vertices(); j+=numaccfunctions) {
auto lvertex = g.l_vertex(j);
if (lvertex.owned()) {
accfunction[i](vertex_type(lvertex));
}
}
}
g.dc().barrier();
}
bool save_binary(const GraphType& g, const std::string& prefix) {
g.dc().full_barrier();
ASSERT_TRUE (g.is_finalized());
timer savetime; savetime.start();
std::string fname = prefix + tostr(g.procid()) + ".bin";
logstream(LOG_INFO) << "Save graph to " << fname << std::endl;
general_ofstream fout(fname, true);
if (!fout.good()) {
logstream(LOG_ERROR) << "\n\tError opening file: " << fname << std::endl;
return false;
}
oarchive oarc(fout);
oarc << g;
logstream(LOG_INFO) << "Finish saving graph to " << fname << std::endl
<< "Finished saving binary graph: "
<< savetime.current_time() << std::endl;
g.dc().full_barrier();
fout.close();
return true;
} // end of save
void transform_vertices(GraphType& g,
TransformType transform_functor,
const vertex_set vset = GraphType::complete_set()) {
typedef typename GraphType::vertex_type vertex_type;
if(!g.is_finalized()) {
logstream(LOG_FATAL)
<< "\n\tAttempting to call graph.transform_vertices(...)"
<< "\n\tbefore finalizing the graph."
<< std::endl;
}
g.dc().barrier();
size_t ibegin = 0;
size_t iend = g.num_local_vertices();
parallel_for (ibegin, iend, [&](size_t i) {
auto lvertex = g.l_vertex(i);
if (lvertex.owned() && vset.l_contains(lvid_type(i))) {
vertex_type vtx(lvertex);
transform_functor(vtx);
}
});
g.dc().barrier();
g.synchronize();
}
void save(GraphType& g,
const std::string& prefix,
Writer writer,
bool gzip = true,
bool save_vertex = true,
bool save_edge = true,
size_t files_per_machine = 4) {
typedef typename GraphType::vertex_type vertex_type;
typedef typename GraphType::edge_type edge_type;
typedef std::function<void(vertex_type)> vertex_function_type;
typedef std::function<void(edge_type)> edge_function_type;
typedef boost::iostreams::filtering_stream<boost::iostreams::output>
boost_fstream_type;
if(!g.is_finalized()) {
g.finalize();
}
g.dc().full_barrier();
// figure out the filenames
std::vector<std::string> graph_files;
std::vector<union_fstream*> outstreams;
std::vector<boost_fstream_type*> booststreams;
graph_files.resize(files_per_machine);
for(size_t i = 0; i < files_per_machine; ++i) {
graph_files[i] = prefix + "_" + tostr(1 + i + g.procid() * files_per_machine)
+ "_of_" + tostr(g.numprocs() * files_per_machine);
if (gzip) graph_files[i] += ".gz";
}
// create the vector of callbacks
std::vector<vertex_function_type> vertex_callbacks(graph_files.size());
std::vector<edge_function_type> edge_callbacks(graph_files.size());
for(size_t i = 0; i < graph_files.size(); ++i) {
logstream(LOG_INFO) << "Saving to file: " << graph_files[i] << std::endl;
union_fstream* out_file =
new union_fstream(graph_files[i], std::ios_base::out | std::ios_base::binary);
// attach gzip if the file is gzip
boost_fstream_type* fout = new boost_fstream_type;
// Using gzip filter
if (gzip) fout->push(boost::iostreams::gzip_compressor());
fout->push(*(out_file->get_ostream()));
outstreams.push_back(out_file);
booststreams.push_back(fout);
vertex_callbacks[i] = [&](vertex_type v) {*fout << writer.save_vertex(v);};
edge_callbacks[i] = [&](edge_type e) {*fout << writer.save_edge(e);};
}
if (save_vertex) parallel_for_vertices(g, vertex_callbacks);
if (save_edge) parallel_for_edges(g, edge_callbacks);
// cleanup
for(size_t i = 0; i < graph_files.size(); ++i) {
booststreams[i]->pop();
if (gzip) booststreams[i]->pop();
outstreams[i]->close();
delete booststreams[i];
delete outstreams[i];
}
vertex_callbacks.clear();
edge_callbacks.clear();
outstreams.clear();
booststreams.clear();
g.dc().full_barrier();
} // end of save to posixfs
