It is possible to get different useful data from your callback by casting cbinfo->functionParams to a specific type.
From the callback_timestamp example:
traceData->memcpy_bytes = ((cudaMemcpy_v3020_params *)(cbInfo->functionParams))->count;
I noticed that there is an option for kernels and that I could get the kernel’s parameters (blockSize and gridSize) by doing this.
Is it possible to get the kernel’s arguments, i.e. the pointers that are used in the kernel?
for example, could I get A, B and C from the kernel below?
__global__ void
VecAdd(const int* A, const int* B, int* C, int N)
{
int i = blockDim.x * blockIdx.x + threadIdx.x;
if (i < N)
C[i] = A[i] + B[i];
}