Hi
Does anyone know of a profiling tool for CUDA that will show you the GPU execution time at the source code level, ie so you can see time spent in each function within a complex kernel?
Many thanks
On Linux: Nvidia Visual Profiler (nvvp) in combination with the -lineinfo compile switch
And on Windows?
Nsight Visual Studio Edition for Windows has built in profiling features.
It seems the NVidia Visual Profiler is in fact a cross platform tool and also available on Windows.
hi sir,
i am not able to see the hotspot and source code along with its path in the kernel profile in the unguided analysis in nvvp
the steps done were
- nvcc wrap_number_increased.cu -lineinfo -o wrap_number_increased
- nvprof -o wrap_number_increased.nvvp ./wrap_number_increased
- nvvp wrap_number_increased.nvvp
The code wrap_number_increased.cu is shown below
#include <cuda.h>
#include <stdio.h>
__global__ void test()
{
int i,j;
for(i=0;i<10;i++)
for(j=0;j<100;j++)
{
// printf("Hello gpu\n");
}
}
int main()
{
printf("Hello from the CPU\n");
test<<<2,32>>>();
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("CUDA Error: %s\n", cudaGetErrorString(err));
}
printf("next is cudaDeviceSynchronize()\n");
cudaDeviceSynchronize();
return 0;
}
In nvvp GUI at the timeline i have selected the function “test” then-> unguided analysis-> then select the some part of function test timeline → then selected the kernel profile- pc sampling
but it is saying
Unable to create source/assembly view for kernel Profile-PC sampling analysis.
As shown below
Sir did you face this kind of issue?? can you kindly guide me in solving this issue