Skip to content

Several fixes for single-precision+vector(+CUDA) mode

Andrei Gheata requested to merge single_prec_vector into master

This fixes the compilation errors in single precision+vector mode, but also fixes all failing ctest tests in Debug mode. Some tests are still failing in Release mode. It turns out the compilation options we use for nvcc do not produce numerically equivalent results, and this is not (only) due to usage of fast math.

Merge request reports