float2, float3, float4, and the like are built-in vector types in CUDA C which have operators defined for the usual math operations. Operations are per component. This is common use in all modern shading or compute languages.
If you have no experience with CUDA C which is used to build the OptiX programs, you should also read through the CUDA C Programming Guide you find inside the CUDA Toolkit installation’s doc/pdf folder.
CUDA 5.0 CUDA_C_Programming_Guide.pdf Appendix B. CUDA C Language Extensions explains this in B.3 Built-in Vector Types.