QR decomposition : rounding error


I am intending to use the QR Decomposition method from Cuda functions. However, my concern is about single floating rounding errors during the entire process. Do you know if this issue will be handled automatically by the Cuda functions or I have to find my own way to solve it ?


Meaning what specifically? Pretty much any real-life floating-point computation incurs rounding errors. It is no different in CUDA than with any other programming environment.

Handling what specifically? How specifically?