Switch default OpenMP schedule to 'runtime'

This also implements direct manipulation of OpenMP settings (schedule
and number of threads) via python.

src/graph/Makefile.am

@@ -24,6 +24,7 @@ libgraph_tool_core_la_SOURCES = \
     graph_copy.cc \
     graph_filtering.cc \
     graph_io.cc \
+    graph_openmp.cc \
     graph_properties.cc \
     graph_properties_copy.cc \
     graph_properties_group.cc \

src/graph/centrality/graph_betweenness.cc

@@ -45,7 +45,7 @@ void normalize_betweenness(const Graph& g,
 
     int i, N = num_vertices(g);
     #pragma omp parallel for default(shared) private(i)   \
-        schedule(static) if (N > 100)
+        schedule(runtime) if (N > 100)
     for (i = 0; i < N; ++i)
     {
         typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/centrality/graph_closeness.hh

@@ -68,7 +68,7 @@ struct get_closeness
         get_vertex_dists_t get_vertex_dists;
         size_t HN = HardNumVertices()(g);
         int i, N = num_vertices(g);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             vertex_t v = vertex(i, g);

src/graph/centrality/graph_eigentrust.hh

@@ -49,7 +49,7 @@ struct get_eigentrust
 
             int i, N = num_vertices(g);
             #pragma omp parallel for default(shared) private(i)  \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -73,7 +73,7 @@ struct get_eigentrust
             c_sum.reserve(num_vertices(g));
             int i, N = num_vertices(g);
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -91,7 +91,7 @@ struct get_eigentrust
         // init inferred trust t
         int i, N = num_vertices(g), V = HardNumVertices()(g);
         #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -106,7 +106,7 @@ struct get_eigentrust
         {
             delta = 0;
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100) reduction(+:delta)
+                schedule(runtime) if (N > 100) reduction(+:delta)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -138,7 +138,7 @@ struct get_eigentrust
         if (iter % 2 != 0)
         {
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/centrality/graph_eigenvector.hh

@@ -59,7 +59,7 @@ struct get_eigenvector
         {
             norm = 0;
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100) reduction(+:norm)
+                schedule(runtime) if (N > 100) reduction(+:norm)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -85,7 +85,7 @@ struct get_eigenvector
 
             delta = 0;
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100) reduction(+:delta)
+                schedule(runtime) if (N > 100) reduction(+:delta)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -105,7 +105,7 @@ struct get_eigenvector
         if (iter % 2 != 0)
         {
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/centrality/graph_hits.hh

@@ -55,7 +55,7 @@ struct get_hits
         // init centrality
         int i, N = num_vertices(g), V = HardNumVertices()(g);
         #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -73,7 +73,7 @@ struct get_hits
         {
             x_norm = 0;
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100) reduction(+:x_norm)
+                schedule(runtime) if (N > 100) reduction(+:x_norm)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -108,7 +108,7 @@ struct get_hits
 
             delta = 0;
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100) reduction(+:delta)
+                schedule(runtime) if (N > 100) reduction(+:delta)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -130,7 +130,7 @@ struct get_hits
         if (iter % 2 != 0)
         {
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/centrality/graph_katz.hh

@@ -57,7 +57,7 @@ struct get_katz
         while (delta >= epsilon)
         {
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -81,7 +81,7 @@ struct get_katz
 
             delta = 0;
             #pragma omp parallel for default(shared) private(i) \
-                schedule(static) if (N > 100) reduction(+:delta)
+                schedule(runtime) if (N > 100) reduction(+:delta)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -100,7 +100,7 @@ struct get_katz
         if (iter % 2 != 0)
         {
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/centrality/graph_pagerank.hh

@@ -43,7 +43,7 @@ struct get_pagerank
         // init degs
         int i, N = num_vertices(g);
         #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -62,7 +62,7 @@ struct get_pagerank
         {
             delta = 0;
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100) reduction(+:delta)
+                schedule(runtime) if (N > 100) reduction(+:delta)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =
@@ -96,7 +96,7 @@ struct get_pagerank
         if (iter % 2 != 0)
         {
             #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/centrality/graph_trust_transitivity.hh

@@ -106,7 +106,7 @@ struct get_trust_transitivity
             property_traits<InferredTrustMap>::value_type::value_type t_type;
 
         int i, N = num_vertices(g);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             vertex_t v = vertex(i, g);
@@ -116,7 +116,7 @@ struct get_trust_transitivity
         }
 
         N = (target == -1) ? num_vertices(g) : target + 1;
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = (target == -1) ? 0 : target; i < N; ++i)
         {
             vertex_t tgt = vertex(i, g);
@@ -210,7 +210,7 @@ struct get_trust_transitivity
                          distance_zero(t_type(1)));
 
                     #pragma omp parallel for default(shared) private(j) \
-                        schedule(static) if (N > 100)
+                        schedule(runtime) if (N > 100)
                     for (j = 0; j < N2; ++j)
                     {
                         vertex_t src = vertex(j, g);
@@ -224,7 +224,7 @@ struct get_trust_transitivity
                 }
 
                 #pragma omp parallel for default(shared) private(j) \
-                    schedule(static) if (N > 100)
+                    schedule(runtime) if (N > 100)
                 for (j = 0; j < N2; ++j)
                 {
                     vertex_t src = vertex(j, g);

src/graph/clustering/graph_clustering.hh

@@ -107,7 +107,7 @@ struct get_global_clustering
         int i, N = num_vertices(g);
 
         #pragma omp parallel for default(shared) private(i,temp) \
-            schedule(static) if (N > 100) reduction(+:triangles, n)
+            schedule(runtime) if (N > 100) reduction(+:triangles, n)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -125,7 +125,7 @@ struct get_global_clustering
         double cerr = 0.0;
 
         #pragma omp parallel for default(shared) private(i,temp) \
-            schedule(static) if (N > 100) reduction(+:cerr)
+            schedule(runtime) if (N > 100) reduction(+:cerr)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -151,7 +151,7 @@ struct set_clustering_to_property
         typename get_undirected_graph<Graph>::type ug(g);
         int i, N = num_vertices(g);
 
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/clustering/graph_extended_clustering.hh

@@ -121,7 +121,7 @@ struct get_extended_clustering
 
         int i, N = num_vertices(g);
 
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             vertex_t v = vertex(i, g);

src/graph/clustering/graph_motifs.hh

@@ -357,7 +357,7 @@ struct get_all_motifs
 
         int i, N = (p < 1) ? V.size() : num_vertices(g);
         #pragma omp parallel for default(shared) private(i, sig) \
-            schedule(static) if (N > 100)
+            schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             std::vector<std::vector<typename graph_traits<Graph>::vertex_descriptor> >

src/graph/community/graph_community.hh

@@ -108,7 +108,7 @@ struct get_communities
             // sample a new spin for every vertex
             int NV = num_vertices(g),i;
             #pragma omp parallel for default(shared) private(i)\
-                reduction(+:E) schedule(static) if (NV > 100)
+                reduction(+:E) schedule(runtime) if (NV > 100)
             for (i = 0; i < NV; ++i)
             {
                 vertex_t v = vertex(i, g);
@@ -211,7 +211,7 @@ struct get_communities
         {
             int NV = num_vertices(g), i;
             #pragma omp parallel for default(shared) private(i)\
-                schedule(static) if (NV > 100)
+                schedule(runtime) if (NV > 100)
             for (i = 0; i < NV; ++i)
             {
                 vertex_t v = vertex(i, g);
@@ -380,7 +380,7 @@ public:
     void Update(size_t k, size_t old_s, size_t s)
     {
         int i, NK = _degs.size();
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (NK > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (NK > 100)
         for (i = 0; i < NK; ++i)
         {
             size_t k1 = _degs[i], k2 = k;
@@ -406,7 +406,7 @@ public:
             _Nks[k].erase(old_s);
         _Nks[k][s]++;
 
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (NK > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (NK > 100)
         for (i = 0; i < NK; ++i)
         {
             size_t k1 = _degs[i], k2 = k;

src/graph/correlations/graph_assortativity.hh

@@ -52,7 +52,7 @@ struct get_assortativity_coefficient
         SharedMap<map_t> sa(a), sb(b);
         int i, N = num_vertices(g);
         #pragma omp parallel for default(shared) private(i) firstprivate(sa,sb)\
-            schedule(static) if (N > 100) reduction(+:e_kk, n_edges)
+            schedule(runtime) if (N > 100) reduction(+:e_kk, n_edges)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -86,7 +86,7 @@ struct get_assortativity_coefficient
 
         // "jackknife" variance
         double err = 0.0;
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)\
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)\
             reduction(+:err)
         for (i = 0; i < N; ++i)
         {
@@ -131,7 +131,7 @@ struct get_scalar_assortativity_coefficient
         double a = 0.0, b = 0.0, da = 0.0, db = 0.0;
         int i, N = num_vertices(g);
         #pragma omp parallel for default(shared) private(i) \
-            schedule(static) if (N > 100) reduction(+:e_xy,n_edges,a,b,da,db)
+            schedule(runtime) if (N > 100) reduction(+:e_xy,n_edges,a,b,da,db)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);
@@ -167,7 +167,7 @@ struct get_scalar_assortativity_coefficient
         r_err = 0.0;
 
         double err = 0.0;
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)\
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)\
             reduction(+:err)
         for (i = 0; i < N; ++i)
         {

src/graph/correlations/graph_avg_correlations.hh

@@ -67,7 +67,7 @@ struct get_avg_correlation
 
         int i, N = num_vertices(g);
         #pragma omp parallel for default(shared) private(i) \
-            firstprivate(s_sum, s_sum2, s_count) schedule(static) if (N > 100)
+            firstprivate(s_sum, s_sum2, s_count) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/correlations/graph_corr_hist.hh

@@ -60,7 +60,7 @@ struct get_correlation_histogram
 
             int i, N = num_vertices(g);
             #pragma omp parallel for default(shared) private(i) \
-                firstprivate(s_hist) schedule(static) if (N > 100)
+                firstprivate(s_hist) schedule(runtime) if (N > 100)
             for (i = 0; i < N; ++i)
             {
                 typename graph_traits<Graph>::vertex_descriptor v =

src/graph/generation/graph_geometric.hh

@@ -135,7 +135,7 @@ struct get_geometric
 
         int i;
         #pragma omp parallel for default(shared) private(i, box) \
-            schedule(static) if (N > 100)
+            schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/generation/graph_lattice.hh

@@ -71,7 +71,7 @@ struct get_lattice
 
         vector<int> pos(shape.size());
         //#pragma omp parallel for default(shared) private(i) 
-        //    firstprivate(pos) schedule(static) if (N > 100)
+        //    firstprivate(pos) schedule(runtime) if (N > 100)
         for (int i = 0; i < N; ++i)
         {
             get_pos(i, shape, pos);

src/graph/graph_adjacency.hh

@@ -628,7 +628,7 @@ inline void remove_vertex(Vertex v, adj_list<Vertex>& g)
 
     int i, N = g._out_edges.size();
     #pragma omp parallel for default(shared) private(i) \
-        schedule(static) if (N > 100)
+        schedule(runtime) if (N > 100)
     for (i = 0; i < N; ++i)
     {
         auto& oes = g._out_edges[i];

src/graph/graph_bind.cc

@@ -60,6 +60,7 @@ struct LibInfo
     }
 };
 
+
 template <class ValueType>
 struct vector_from_list
 {
@@ -373,15 +374,6 @@ string get_graph_type(GraphInterface& g)
     return name;
 }
 
-bool openmp_enabled()
-{
-#ifdef USING_OPENMP
-    return true;
-#else
-    return false;
-#endif
-}
-
 // numpy array interface weirdness
 void* do_import_array()
 {
@@ -401,6 +393,8 @@ void mark_edges(GraphInterface& gi, boost::any prop);
 
 void export_python_interface();
 
+void export_openmp();
+
 BOOST_PYTHON_MODULE(libgraph_tool_core)
 {
     using namespace boost::python;
@@ -426,7 +420,7 @@ BOOST_PYTHON_MODULE(libgraph_tool_core)
         .def("empty", &boost::any::empty);
 
     def("graph_filtering_enabled", &graph_filtering_enabled);
-    def("openmp_enabled", &openmp_enabled);
+    export_openmp();
 
     boost::mpl::for_each<boost::mpl::push_back<scalar_types,string>::type>(export_vector_types());
 

src/graph/graph_copy.cc

@@ -123,7 +123,7 @@ struct copy_vertex_property_dispatch
             ptgt->get_unchecked(num_vertices(tgt));
 
         int i, N = num_vertices(src);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<GraphSrc>::vertex_descriptor v = vertex(i, src);
@@ -197,7 +197,7 @@ struct copy_edge_property_dispatch
             ptgt->get_unchecked(num_edges(tgt));
 
         int i, N = num_vertices(src);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<GraphSrc>::vertex_descriptor v = vertex(i, src);

src/graph/graph_openmp.cc

+// graph-tool -- a general graph modification and manipulation thingy
+//
+// Copyright (C) 2006-2014 Tiago de Paula Peixoto <tiago@skewed.de>
+//
+// This program is free software; you can redistribute it and/or
+// modify it under the terms of the GNU General Public License
+// as published by the Free Software Foundation; either version 3
+// of the License, or (at your option) any later version.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU General Public License
+// along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+#include "graph.hh"
+
+#include <boost/python.hpp>
+
+using namespace std;
+using namespace boost;
+using namespace graph_tool;
+
+// OpenMP control
+#ifdef USING_OPENMP
+#include <omp.h>
+#endif
+
+bool openmp_enabled()
+{
+#ifdef USING_OPENMP
+    return true;
+#else
+    return false;
+#endif
+}
+
+size_t openmp_get_num_threads()
+{
+#ifdef USING_OPENMP
+    return omp_get_max_threads();
+#else
+    throw GraphException("OpenMP was not enabled during compilation")
+#endif
+}
+
+void openmp_set_num_threads(int n)
+{
+#ifdef USING_OPENMP
+    omp_set_num_threads(n);
+#else
+    throw GraphException("OpenMP was not enabled during compilation")
+#endif
+}
+
+python::tuple openmp_get_schedule()
+{
+#ifdef USING_OPENMP
+    omp_sched_t kind;
+    int chunk;
+    omp_get_schedule(&kind, &chunk);
+    string skind;
+    switch (kind)
+    {
+        case omp_sched_static:
+            skind = "static";
+            break;
+        case omp_sched_dynamic:
+            skind = "dynamic";
+            break;
+        case omp_sched_guided:
+            skind = "guided";
+            break;
+        case omp_sched_auto:
+            skind = "auto";
+            break;
+        default:
+            throw GraphException("Unknown schedule type");
+    }
+
+    return python::make_tuple(skind, chunk);
+#else
+    throw GraphException("OpenMP was not enabled during compilation")
+#endif
+}
+
+void openmp_set_schedule(string skind, int chunk)
+{
+#ifdef USING_OPENMP
+    omp_sched_t kind;
+    if (skind == "static")
+        kind = omp_sched_static;
+    else if (skind == "dynamic")
+        kind = omp_sched_dynamic;
+    else if (skind == "guided")
+        kind = omp_sched_guided;
+    else if (skind == "auto")
+        kind = omp_sched_auto;
+    else
+        throw GraphException("Unknown schedule type: " + skind);
+    omp_set_schedule(kind, chunk);
+#else
+    throw GraphException("OpenMP was not enabled during compilation")
+#endif
+}
+
+
+void export_openmp()
+{
+    using namespace boost::python;
+
+    def("openmp_enabled", &openmp_enabled);
+    def("openmp_get_num_threads", &openmp_get_num_threads);
+    def("openmp_set_num_threads", &openmp_set_num_threads);
+    def("openmp_get_schedule", &openmp_get_schedule);
+    def("openmp_set_schedule", &openmp_set_schedule);
+};

src/graph/graph_properties.cc

@@ -264,7 +264,7 @@ struct do_mark_edges
     {
         int i, N = num_vertices(g);
         #pragma omp parallel for default(shared) private(i)     \
-                schedule(static) if (N > 100)
+                schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/graph_python_interface.cc

@@ -242,7 +242,7 @@ struct get_degree_map
         typename map_t::unchecked_t deg_map = cdeg_map.get_unchecked(num_vertices(g));
 
         int i, N = num_vertices(g);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v = vertex(i, g);

src/graph/layout/graph_sfdp.cc

@@ -255,7 +255,7 @@ struct do_sanitize_pos
     void operator()(Graph& g, PosMap pos) const
     {
         int i, N = num_vertices(g);
-        #pragma omp parallel for default(shared) private(i) schedule(static) if (N > 100)
+        #pragma omp parallel for default(shared) private(i) schedule(runtime) if (N > 100)
         for (i = 0; i < N; ++i)
         {
             typename graph_traits<Graph>::vertex_descriptor v =