int main(int argc, char* argv[])
{
try
{
size_t kmerSize = 11;
cost char* seq = "AGGCGCTAGGGTAGAGGATGATGA";
std::cout
<< "The initial sequence is '"
<< seq
<< "', kmer size is"
<< kmerSize
<< std::endl;
// We create the graph from a given sequence, and for a given kmer size
Graph graph = Graph::create(
new BankStrings( seq, NULL),
"-kmer-size %d -abundance-min 1 -verbose 0", kmerSize
);
// Get the first node
Node node = graph.buildNode(seq);
// Create a node iterator that iterates over all the simple nodes from
// the first node. Rememeber that a simple node has inDeg = outDeg = 1
Graph::Iterator<Node> path =
graph.simplePath<Node> (node, DIR_OUTCOMING);
// Iterate over the simple path:
for (path.start(); !path.isDone(); path.next())
{
std::cout
<< " ["
<< path.rank()
<< "] current item is "
<< graph.toString(path.item())
<< std::endl;
}
std::cout
<< "The simple path was "
<< path.rank()
<< " long"
<< std::endl;
}
catch(Exception& e)
{
std::cerr << "EXCEPTION: " << e.getMessage() << std::endl;
}
return EXIT_SUCCESS;
}
template<size_t span> struct debruijn_mphf_bench { void operator () (Parameter params)
{
typedef NodeFast<span> NodeFastT;
typedef GraphTemplate<NodeFastT,EdgeFast<span>,GraphDataVariantFast<span>> GraphFast;
size_t kmerSize = params.k;
Graph graph;
GraphFast graphFast;
if (params.seq == "")
{
graph = Graph::create (params.args.c_str());
graphFast = GraphFast::create (params.args.c_str());
}
else
{
graph = Graph::create (new BankStrings (params.seq.c_str(), 0), params.args.c_str());
graphFast = GraphFast::create (new BankStrings (params.seq.c_str(), 0), params.args.c_str());
}
cout << "graph built, benchmarking.." << endl;
int miniSize = 8;
int NB_REPETITIONS = 2000000;
double unit = 1000000000;
cout.setf(ios_base::fixed);
cout.precision(3);
Graph::Iterator<Node> nodes = graph.iterator();
typename GraphFast::template Iterator<NodeFastT> nodesFast = graphFast.iterator();
nodes.first ();
/** We get the first node. */
Node node = nodes.item();
typedef typename Kmer<span>::Type Type;
typedef typename Kmer<span>::ModelCanonical ModelCanonical;
typedef typename Kmer<span>::ModelDirect ModelDirect;
typedef typename Kmer<span>::template ModelMinimizer <ModelCanonical> ModelMini;
typedef typename ModelMini::Kmer KmerType;
ModelMini modelMini (kmerSize, miniSize);
ModelCanonical modelCanonical (kmerSize);
// for some reason.. if *compiled*, this code confuses makes later MPHF queries 3x slower. really? yes. try to replace "if (confuse_mphf)" by "if (confuse_mphf && 0)" and re-run, you will see.
{
bool confuse_mphf = false;
if (confuse_mphf)
{
//Type b; b.setVal(0);
//modelCanonical.emphf_hasher(modelCanonical.adaptor(b));
//typedef std::pair<u_int8_t const*, u_int8_t const*> byte_range_t;
//int c = 0;
//byte_range_t brange( reinterpret_cast <u_int8_t const*> (&c), reinterpret_cast <u_int8_t const*> (&c) + 2 );
//byte_range_t brange( (u_int8_t const*) 1,(u_int8_t const*)33);
//auto hashes = modelCanonical.empfh_hasher(brange);
}
for (int i = 0; i < 0; i++)
{
auto start_tt=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelCanonical.EMPHFhash(nodes.item().kmer.get<Type>());
auto end_tt=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computing EMPHFhash of kmers on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_tt, end_tt) / unit) << " seconds" << endl;
}
// it's slow. i don't understand why. see above for the "confuse mphf" part
//return; //FIXME
}
/** We get the value of the first node (just an example, it's not used later). */
Type kmer = node.kmer.get<Type>();
auto start_t=chrono::system_clock::now();
auto end_t=chrono::system_clock::now();
cout << "----\non all nodes of the graph\n-----\n";
/* disable node state (because we don't want to pay the price for overhea of checking whether a node is deleted or not in contain() */
std::cout<< "PAY ATTENTION: this neighbor() benchmark, in the Bloom flavor, is without performing a MPHF query for each found node" << std::endl;
graph.disableNodeState();
graphFast.disableNodeState();
/* compute baseline times (= overheads we're not interested in) */
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
{}
end_t=chrono::system_clock::now();
auto baseline_graph_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph nodes enumeration (" << nodes.size() << " nodes) : " << baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
{}
end_t=chrono::system_clock::now();
auto baseline_graphfast_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph NodeFast enumeration (" << nodes.size() << " nodes) : " << baseline_graphfast_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelMini.getMinimizerValueDummy(nodes.item().kmer.get<Type>());
end_t=chrono::system_clock::now();
auto baseline_minim_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph nodes enumeration and minimizer computation setup (" << nodes.size() << " nodes) : " << baseline_minim_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
nodes.item().getKmer<Type>();
end_t=chrono::system_clock::now();
auto baseline_hash_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph nodes enumeration and hash computation setup (" << nodes.size() << " nodes) : " << baseline_hash_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
nodesFast.item().kmer;
end_t=chrono::system_clock::now();
auto baseline_hashfast_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph NodeFast enumeration and hash computation setup (" << nodes.size() << " nodes) : " << baseline_hashfast_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.nodeMPHFIndexDummy(nodes.item());
end_t=chrono::system_clock::now();
auto baseline_mphf_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph nodes enumeration and mphf query setup (" << nodes.size() << " nodes) : " << baseline_mphf_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.nodeMPHFIndexDummy(nodesFast.item());
end_t=chrono::system_clock::now();
auto baseline_mphffast_time = diff_wtime(start_t, end_t) / unit;
cout << "baseline overhead for graph NodeFast enumeration and mphf query setup (" << nodes.size() << " nodes) : " << baseline_mphffast_time << " seconds" << endl;
/* do actual benchmark */
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelMini.getMinimizerValue(nodes.item().kmer.get<Type>(), true);
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computations of minimizers (fast method) of length " << miniSize << " on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_minim_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.nodeMPHFIndex(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computations of MPHF index on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_mphf_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.nodeMPHFIndex(nodesFast.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computations of MPHF index on all NodeFast (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_mphffast_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelCanonical.getHash(nodes.item().kmer.get<Type>());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computing hash1 of kmers on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_hash_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelCanonical.getHash2(nodes.item().kmer.get<Type>());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computing hash2 of kmers on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_hash_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
modelCanonical.getHash2(nodesFast.item().kmer);
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computing hash2 of kmers on all NodeFast (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_hashfast_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
modelCanonical.EMPHFhash(nodes.item().kmer.get<Type>());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " computing EMPHFhash of kmers on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_hash_time << " seconds" << endl;
// it's slow. i don't understand why. see above for the "confuse mphf" part
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.neighborsDummy(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " dummy neighbors() query on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.neighbors(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " neighbors() query on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.neighbors(nodesFast.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " neighbors() query on all NodeFast (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_graphfast_time << " seconds" << endl;
/* isBranching */
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.isBranching(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " isBranching() query on all nodes (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.isBranching(nodesFast.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " isBranching() query on all NodeFast (" << kmerSize << "-mers) : " << (diff_wtime(start_t, end_t) / unit) - baseline_graphfast_time << " seconds" << endl;
/* now, compute adjacency! */
graph.precomputeAdjacency();
graphFast.precomputeAdjacency();
cout << "adjacency precomputed" << endl;
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.neighbors(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " neighbors() query on all nodes (" << kmerSize << "-mers) using adjacency : " << (diff_wtime(start_t, end_t) / unit) - baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.neighbors(nodesFast.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " fast neighbors() query on all NodeFast (" << kmerSize << "-mers) using adjacency : " << (diff_wtime(start_t, end_t) / unit) - baseline_graphfast_time << " seconds" << endl;
/* isBranching */
start_t=chrono::system_clock::now();
for (nodes.first(); !nodes.isDone(); nodes.next())
graph.isBranching(nodes.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " isBranching() query on all nodes (" << kmerSize << "-mers) using adjacency : " << (diff_wtime(start_t, end_t) / unit) - baseline_graph_time << " seconds" << endl;
start_t=chrono::system_clock::now();
for (nodesFast.first(); !nodesFast.isDone(); nodesFast.next())
graphFast.isBranching(nodesFast.item());
end_t=chrono::system_clock::now();
cout << "time to do " << nodes.size() << " fast isBranching() query on all NodeFast (" << kmerSize << "-mers) using adjacency : " << (diff_wtime(start_t, end_t) / unit) - baseline_graphfast_time << " seconds" << endl;
/** We remove the graph. */
//graph.remove ();
//graphFast.remove (); // no actually, I want to keep the .h5 file
}