Hi all, I’d like to ask that for a particular kernel, can I see how many instructions are used from the output information of complier? or any other methods? Thanks!
Compile the device code to ptx (nvcc -ptx foo.cu) which gives you an assembler-type output in foo.ptx. Alternatively, look at the size of the cubin (nvcc -cubin foo.cu).
I’d like to ask another problem, that the “number” of device memory access output from Visual Profiler, is it just means the number of memory access instructions? Since I’d like to estimate the memory access time, could I utilize the output information from the profiler? Thanks!