Counting the number of bank conflicts.

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

Is there any information available as to how does the cudaProfiler calculates the number of bank conflicts and then populates the counter “warp serialized”?
The PTX I believe does not say anything on how the shared memory is accessed, then from where does the knowledge of bank conflicts come from?

Thanks.

2

My guess would be that there is a hardware performance counter that records this, but it is just speculation.