Batch write

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

In my application, every thread needs to read from global memory to shared memory, then writes back something to global memory. Is there a way that I could do batch write? I guess this may improve the perfomance.

Thanks.

Casy

2

You’re basically looking for coalesced memory accesses–read chapter 5 of the programming guide. There’s no memcpy-like function available from within a kernel that will do this automatically, though.