computeProf Capabilities

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,
I have just started trying to use computeprof (ver 3.1.1) and while I
can see times for the kernel that is called it doesn’t seem to break this
down further. What I’m really interested in is how long it spends using the
other (non kernel) functions to try and see which of these is using the
most time.

Is this possible, or does it simply report aggregate time for the named
kernel that was called (including calls to other functions) ?

–
jason

2

Hi,
I have just started trying to use computeprof (ver 3.1.1) and while I
can see times for the kernel that is called it doesn’t seem to break this
down further. What I’m really interested in is how long it spends using the
other (non kernel) functions to try and see which of these is using the
most time.

Is this possible, or does it simply report aggregate time for the named
kernel that was called (including calls to other functions) ?

–
jason

3

I doubt there is a way how to find what you want. I think all functions are inlined, so as a result there is only one long kernel code.

4

I doubt there is a way how to find what you want. I think all functions are inlined, so as a result there is only one long kernel code.

5

Oh yes, that makes sense, thanks.

–

jason

6

Oh yes, that makes sense, thanks.

–

jason