Made number of threads command line configurable and set default to 1

src/chain/chain-generic-numerator.cc

@@ -241,7 +241,8 @@ bool GenericNumeratorComputation::ForwardBackward(
   derivs.Set(-std::numeric_limits<BaseFloat>::infinity());
 
   // Set total number of workers to the available hardware concurrency
-  unsigned int nthreads = opts_.multithreaded ? std::thread::hardware_concurrency() : 1;
+  unsigned int nthreads = opts_.num_threads > 0 ? opts_.num_threads :
+                              std::thread::hardware_concurrency();
   // Naive load balancing, each thread gets a chunk of the sequences to process
   unsigned int num_sequences_per_thread = 
     (num_sequences + nthreads - 1) / nthreads;

src/chain/chain-generic-numerator.h

@@ -103,12 +103,12 @@ namespace chain {
 
 
 struct GenericNumeratorComputationOptions {
-  bool multithreaded;
-  GenericNumeratorComputationOptions(): multithreaded(true) { }
+  unsigned int num_threads;
+  GenericNumeratorComputationOptions(): num_threads(1) { }
   void Register(OptionsItf *opts) {
-    opts->Register("multithreaded-numerator-graph", &multithreaded, "If true, "
-                   "use multiple threads to parallelize the chain numerator "
-                   "graph computation");
+    opts->Register("numerator-graph-threads", &num_threads, "Number of threads "
+                   "to use to parallelize the chain numerator graph computation. "
+                   "If 0, use available hardware concurrency.");
   }
 
 };