Memory alignment when using cudamalloc?

[Ajes]

Hi there

When using the normal malloc() call on a linux system, newly allocated memory is always aligned at addresses that are a multiple of four. Does anyone know if this is also the case with cudamalloc()?

Im asking because I’m currently writing a program in which its important that the first two bits of a memory address are always zero, which is the case in newly allocated host memory on linux (at least when allocated with malloc())

I checked the CUDA programming guide, but couldnt find this specific information anywhere.

Thanks!

[Nico]

Section 5.3.2.1.1 of the programming guide.

Any address of a variable residing in global memory or returned by one of the

memory allocation routines from the driver or runtime API is always aligned to at

least 256 bytes.

N.

[Ajes]

Turns out I was checking the wrong section of the programming guide, sorry about that :)

Thanks for the reply!