Dear CUDA programmers,
I am programming an application with CUDA.
Generally, some applications include many memory accesses, for example FFT.
In those applications, we can utilize SHARED MEMORY and do that actually.
This time, I want to know “# of SHARE MEMORY ACEESS” to analyze performance from the point of view of MEMORY ACCESS.
Off course I know we can get such info as # of GLOBAL MEMORY and # of BANK CONFLICT in SHARED MEMORY, using CUDA PROFILER.
However, as far as I know, this PROFILER don’t seem to acquire “# of SHARED MEMORY ACCESS”.
Please more information or technique…