Problems with type double


I just wanted to recompile the ScalarProduct Sample, but with double instead of float. I simply replaced every occurrence of float with double, but the Test failed. Even if I try to use [codebox]__dmul_ru()[/codebox] for double-double multiplication the compiler says that it is undefined. I have also written my own scalar product but with same result, the result vector C contains zeros only. What went wrong?

OK, problem solved.

One has to add “-arch=sm_13” to nvcc compiler options. Cuda by default does not support double.