CUDA Profiler Error

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I get the following error when I try to use cuda profiler to monitor my project.
The problem is I don’t get this error when I use the profiler on one of the existing nvidia sample projects.

My projects executes completely with no errors. But, I have no clue what I’m doing wrong here. Any suggestions???

Below is the error I get:

=========================================================

=== Start profiling for session ‘Session2’ ===
Start program xterm -e /home/maryam/NVIDIA_CUDA_SDK/bin/linux/release/denoising, run #1 …
Program run #1 completed.
Error -97 in reading profiler output.
Unable to open file.

=========================================================

2

Are you seeing this with the Visual Profiler? If so, are you using the latest release:
[url=“http://forums.nvidia.com/index.php?showtopic=65489”]http://forums.nvidia.com/index.php?showtopic=65489[/url]

The best way to identify the root cause is to run the application with profiling enabled from the command line.

3

Sorry for barging in here, I believe I have a similar problem.

I’m using the driver API, I have compiled my kernels using -pg. I have the following environment variables set:

export CUDA_PROFILE='1'

export CUDA_PROFILE_CONFIG='temp_cuda_profiler.conf'

export CUDA_PROFILE_CSV='1'

export CUDA_PROFILE_LOG='temp_cuda_profiler_0.csv'

temp_cuda_profiler.conf contains

gld_incoherent

gld_coherent

gst_incoherent

gst_coherent

(I just tried to copy what the visual profiler is doing, hope I caught everything)

After I do all that and run my program, temp_cuda_profiler_0.csv gets created, but is empty.

For reference, if I unset everything but CUDA_PROFILE, I get a file cuda_profile.log with this here:

method,gputime,cputime,occupancy

method=[ fill ] gputime=[ 4.992 ] cputime=[ 12.000 ] occupancy=[ 0.333 ]

method=[ fill ] gputime=[ 3.008 ] cputime=[ 5.000 ] occupancy=[ 0.333 ]

method=[ fill ] gputime=[ 2.944 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.048 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.048 ] cputime=[ 7.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 5.984 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.048 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.112 ] cputime=[ 7.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 5.856 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.272 ] cputime=[ 7.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 6.016 ] cputime=[ 8.000 ] occupancy=[ 0.333 ]

method=[ axpbyz ] gputime=[ 5.888 ] cputime=[ 7.000 ] occupancy=[ 0.333 ]

etc etc

CUDA2beta, Linux x86-64, Driver 174.55, Ubuntu 8.04, Linux 2.6.24-17

What am I missing?

Thanks

Andreas

4

The solution is simple: Detailed profiling does not work on a Quadro FX570, it’s fine on other cards.

Andreas