WIP: Feature/openacc

This branch contains the OpenACC implementation of varopti for meshes with one type of element and polynomial order. OpenACC can be enabled or disabled with the compile-time variable USE_ACC, but a way to pass it from Cmake still needs to be devised. Compile OpenACC with pgi 19.4 and boost 1.57.0. Cmake CXX flags: -fPIC -noswitcherror -std=c++11 -Minfo=accel -ta=tesla or -ta=multicore. Can use flags -O3 -fast, but not -Mipa for inter-procedural analysis (Nektar's huge call tree will kill it). 3D algorithms are successfully tested against master, but not 2D, yet.

utilities/NekMesh/ProcessModules/ProcessVarOpti/ElUtil.cpp

@@ -76,6 +76,9 @@ ElUtil::ElUtil(ElementSharedPtr e, DerivUtilSharedPtr d, ResidualSharedPtr r,
         }
 
         m_idmap[ns[i]->m_id] = i;
+
+        // global ids
+        m_idglmap[i] = ns[i]->m_id;
     }
     MappingIdealToRef();
 }