Fix ParallelSGD test step sizes and enable OpenMP for tests when found

CMakeLists.txt

@@ -52,6 +52,7 @@ endif()
 
 # Find OpenMP and link it.
 if(USE_OPENMP)
+  find_package(OpenMP)
   if(NOT TARGET OpenMP::OpenMP_CXX)
     find_package(Threads REQUIRED)
     add_library(OpenMP::OpenMP_CXX IMPORTED INTERFACE)
@@ -61,6 +62,8 @@ if(USE_OPENMP)
     set_property(TARGET OpenMP::OpenMP_CXX
         PROPERTY INTERFACE_LINK_LIBRARIES ${OpenMP_CXX_FLAGS} Threads::Threads)
   endif()
+
+  target_compile_definitions(ensmallen INTERFACE -DENS_USE_OPENMP)
   target_link_libraries(ensmallen INTERFACE OpenMP::OpenMP_CXX)
 endif()
 

include/ensmallen_bits/config.hpp

@@ -48,6 +48,7 @@
   #define ENS_PRAGMA_OMP_ATOMIC   _Pragma("omp atomic")
   #define ENS_PRAGMA_OMP_CRITICAL _Pragma("omp critical")
   #define ENS_PRAGMA_OMP_CRITICAL_NAMED _Pragma("omp critical(section)")
+  #include <omp.h>
 #else
   #define ENS_PRAGMA_OMP_PARALLEL
   #define ENS_PRAGMA_OMP_ATOMIC

tests/CMakeLists.txt

@@ -65,6 +65,9 @@ set(ENSMALLEN_TESTS_SOURCES
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
 add_executable(ensmallen_tests EXCLUDE_FROM_ALL ${ENSMALLEN_TESTS_SOURCES})
 target_link_libraries(ensmallen_tests PRIVATE ensmallen)
+if (USE_OPENMP)
+  target_link_libraries(ensmallen_tests PRIVATE OpenMP::OpenMP_CXX)
+endif ()
 
 # Copy test data into place.
 add_custom_command(TARGET ensmallen_tests

tests/moead_test.cpp

@@ -560,8 +560,6 @@ TEMPLATE_TEST_CASE("DefaultMOEAD_ZDT1Function", "[MOEAD]", ENS_TEST_TYPES)
  */
 TEMPLATE_TEST_CASE("DirichletMOEAD_ZDT3Function", "[MOEAD]", ENS_TEST_TYPES)
 {
-  typedef typename TestType::elem_type ElemType;
-
   //! Parameters taken from original ZDT Paper.
   ZDT3<TestType> zdt3(300);
   const double lowerBound = 0;
@@ -591,4 +589,4 @@ TEMPLATE_TEST_CASE("DirichletMOEAD_ZDT3Function", "[MOEAD]", ENS_TEST_TYPES)
   opt.Optimize(objectives, coords, finalPopulation, finalFront);
 
   REQUIRE(VariableBoundsCheck(finalPopulation));
-}
+}

tests/parallel_sgd_test.cpp

@@ -45,8 +45,8 @@ TEMPLATE_TEST_CASE("ParallelSGDTest_SparseFunction", "[ParallelSGD]",
 
     size_t batchSize = std::ceil((float) f.NumFunctions() / i);
 
-    ConstantStep decayPolicy(0.4 * batchSize);
-    ParallelSGD<ConstantStep> s(10000, batchSize, 1e-5, true, decayPolicy);
+    ConstantStep decayPolicy(0.02 * batchSize);
+    ParallelSGD<ConstantStep> s(10000, batchSize, 1e-7, true, decayPolicy);
     FunctionTest<SparseTestFunction>(s,
         Tolerances<TestType>::LargeObj,
         Tolerances<TestType>::LargeCoord);
@@ -62,22 +62,22 @@ TEMPLATE_TEST_CASE("ParallelSGD_GeneralizedRosenbrockFunction",
     // Create the generalized Rosenbrock function.
     GeneralizedRosenbrockFunction f(i);
 
-    ConstantStep decayPolicy(0.001 * f.NumFunctions());
+    ConstantStep decayPolicy(0.00005 * f.NumFunctions());
 
     ParallelSGD<ConstantStep> s(
-        100000, f.NumFunctions(), Tolerances<TestType>::Obj / 100, true,
+        1000000, f.NumFunctions(), Tolerances<TestType>::Obj / 100, true,
         decayPolicy);
 
     TestType coordinates = f.GetInitialPoint<TestType>();
 
     omp_set_num_threads(1);
     double result = s.Optimize(f, coordinates);
 
-    REQUIRE(result == Approx(0.0).margin(Tolerances<TestType>::Obj));
+    REQUIRE(result == Approx(0.0).margin(100 * Tolerances<TestType>::Obj));
     for (size_t j = 0; j < i; ++j)
     {
       REQUIRE(coordinates(j) ==
-          Approx(1.0).epsilon(Tolerances<TestType>::Coord));
+          Approx(1.0).epsilon(100 * Tolerances<TestType>::Coord));
     }
   }
 }