I am using cudaprof under linux. However, the option for gld/gst coalesced/uncoalesced cannot be selected. Other options are fine.
What could be wrong?
I would guess your are using a compute 1.2 or 1.3 capability GPU, and those counters are no longer valid for the latest GPUs. There are alternative, more useful memory transaction counters you can use instead.
Thanks for your reply.
Yes I am using cc 1.2
But Are you talking about the counters gld/gst_32/64/128b? But from those counters, we will do some calculation to figure out whether the global memory access is coalesced or not.
Does gld/gst_32/64/128b only count for direct global memory asscess nor it also counts the indirect global memory access though texture?