command line profiling

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, i want to run some benchmarks on remote systems. for that i would like to turn on profiling, like we used to do with cuda before the visual profiler days :) …

will CUDA_PROFILE=1 work on opencl as well ? or are there other variables needed for opencl ?

thanks

2

OpenCL is all about run-time configuration… Profiling can be enabled with an api call. Simply search “profiling” in the Khronos OpenCL specification and you’ll get what you want:

5.9 Profiling Operations on Memory Objects and Kernels

3

Thanks pobelzile, i know about that, but i was looking for a way of profiling applications that i didn’t write and don’t want to modify in any way … and since i know that the visual profiler activates such a mode it’s only a question of how :)

4

OPENCL_PROFILE=1 works on my Linux installation. It generates a file opencl_profile_N.log which IIRC can be imported into the OpenCL Visual Profiler.

5

Thanks ! just what i was hoping to hear :)

6

Hi,

I’m trying to run a prfile on my kernel but I didn’t know how until I fount this topic,

is OPENCL_PROFILE=1 needs to be added to the kernel, or to c code, can anyone give details please.

Thanks.

7

Its an environment variable

8

For me, this creates the file but it is empty. Any ideas why?

EDIT: Found it if anyone is interested and stumbles across this thread.

http://developer.download.nvidia.com/compu…rofiler_3.0.txt

9

For me, this creates the file but it is empty. Any ideas why?

EDIT: Found it if anyone is interested and stumbles across this thread.

http://developer.download.nvidia.com/compu…rofiler_3.0.txt

10

Ok, so actually this outputs the log file just fine and reads my config file just fine but doesn’t actually put anything IN the log file!!??

Any ideas?

11

Ok, so actually this outputs the log file just fine and reads my config file just fine but doesn’t actually put anything IN the log file!!??

Any ideas?