The problem is that the __CUDA_ARCH__ macro is defined only in device
code, so it can't be used for configuring the kernel launches.

Implementace Array Operations pro MIC, vedoucí k zpřehlednění kódu -> kód použitý pro srovnávací test MIC-krakow, MIC-sharedpoiters, MIC-ArrayOperations

Apparently there is an abs function defined in <cstdlib>, which works
only on integers. We want the std::abs overloads defined in <cmath>,
which are used by TNL::abs.

clang rocks!

http://stackoverflow.com/questions/29136607/c-what-is-wrong-with-abs#29136774

This prevented scaling of GMRES on small grids (up to 240x240)!