__ballot(..) and warpSize warpSize will never surpass 32 threads?

[Jimmy_Pettersson]

Hi,

I was reading a little bit about the __ballot(…) function in programming guide 3.0:

unsigned int __ballot(int predicate);

evaluates predicate for all threads of the warp and returns an integer whose Nth bit is set if and only if predicate evaluates to non-zero for the Nth thread of the warp. This function is only supported by devices of compute capability 2.0.

Since there are only 32-bit integers being returned, that should mean that you can only have a maximum of 32 threads / warp. Now if Nvidia decides to change the warpSize in the future that should cause a problem for this function shouldn’t it? Unless they start using 64-bit integers…

[Cygnus_X1]

If they are going to increase the warp size, it will be a device with a different Compute Capability and different set of intristic functions. It might however lead to some CUDA code portability problems…

[_Big_Mac]

Nobody says sizeof(int)=32 bits…

Portability issues may arise if one assumes so. That would be wrong even on CPUs.

[Jimmy_Pettersson]

@cygnus

Yes, it was portability issues i was thinking of.