Question about "total" coumn for an event

Accelerated Computing CUDA CUDA Programming and Performance
#1

What does the Total column mean for an event in nvprof?

Invocations                                Event Name         Min         Max         Avg       Total
Kernel: void findRadixOffsets<unsigned int=0, bool=0, bool=0>(uint2*, unsigned int*, unsigned int*, unsigned int, unsigned  int)
   1900     thread_inst_executed     2352688     2875340     2550612  4846163556

The value of Total is far greater than AVG*32 (as a guess). Any thought on that?

#2

The kernel was invoked (launched) 1900 times.

Min - Minimum event value reported in 1900 invocations of findRadixOffsets<…>
Max- Maximum event value reported in 1900 invocations of findRadixOffsets<…>
Avg - Average event value reported in 1900 invocations of findRadixOffsets<…>
Total - Sum of event value across all 1900 invocations of findRadixOffsets<…>

1 Like
#3