Any idea why If I have only a project in my VS2008SP1's solution, it's marked as startup project, I'm administrator and I can write perfectly my project's file do I get this message?
"The program failed to profile. Try to compile the active project manually and make sure you have a write access permission to the project's directory"
I use Windows 7 x64, VS2008 SP1, Cat.10.5, a Radeon 5750 and StreamSDK 2.1.
thx
What is in the Linker -> Output File field of your project property? If you can send the project file (just the vsproj file) to gputools.support@amd.com, that would be helpful for us to quickly diagnose the problem.
Originally posted by: bpurnomo What is in the Linker -> Output File field of your project property? If you can send the project file (just the vsproj file) to gputools.support@amd.com, that would be helpful for us to quickly diagnose the problem.
$(OutDir)\myOCLPlugin.dll
and OutDir is
..\..\deploy\MyPlugin\x86
( yes, I need relative paths, it's a fixed company's deployment policy )
The profiler needs to be run with a project that generates an executable (.exe). So you should set that project to be the startup project. Basically, you need to be able to run the project with Ctrl F5 successfully before running it with the profiler.
Originally posted by: bpurnomo The profiler needs to be run with a project that generates an executable (.exe).
And if my project is a a Photoshop filter ( .8bf ) or a 3dsmax plugin (.dlX)?
If you are able to run 3dsmax or photoshop with a command line argument to load your plugin and perform the GPU work, then you can run it with the command line interface of the profiler.
Otherwise, you can use GPUPerfAPI, a library to access GPU perfomance counters, to instrument your own code.
Originally posted by: bpurnomo If you are able to run 3dsmax or photoshop with a command line argument to load your plugin
Those programs don't work as that.
They just call LoadLibrary loading each DLL in a folder. The only command line I can execute is 3dsmax.exe of Photoshop.exe.
You really should take that into consideration, as the NVIDIA ocl profiler does.
Otherwise, you can use GPUPerfAPI, a library to access GPU perfomance counters, to instrument your own code.
Yep, but I really prefer to use a simple profiler like NVIDIA's one, but running on ATI.