Are 64-bit integer instructions natively supported by GPU?

The .cu source code do support 64-bit integer, and the ptx specification includes instruction like add.u64 to do 64-bit calculation.

Are those 64-bit integer instructions supported by hardware natively? And how about clock cycles used compared to 32-bit instructions…

The current hardware does not support 64-bit native integer instructions (it doesn’t even support full 32-bit multiplies with a single instruction). Fermi improves the 64-bit support by adding full 32-bit multiples, but does not have full 64-bit integer math w/ single instructions either.