We are using the NVIDIA’s OpenCL conformant 1.0 SDK to develop an application. We are compiling the OpenCL code and getting the “.ptx” files as binaries ( different devices have different formats ). When we try to run the code we get an “CL_OUT_OF_RESOURCES (-5)” error. Therefore there is a problem with device capabilities. How can we learn which property had caused the error ? Is there a possibility to propagate device specific error to the application level ?
Also we have seen that the ptx files are version 1.5. I think this compiler is in the works right now. We would like to get info for usage of registers, memory etc from the ptx file. Will this compiler be released soon ?
How healty it would be to convert version to 1.4 and compile and see ? ( We have done this but not sure about the results ).
Currently OpenCL compiler has no optimization so we would like to do it using the nvcc, is it possible using it and run compiled cubins ?
Ahmet,
This probably isn’t the problem you have, but CL_OUT_OF_RESOURCES is also used (incorrectly) to indicate that an unknown kernel name was passed at clCreateKernel time. It doesn’t seem to give the error until later trying to enqueue the kernel.
It wasn’t a kernel name error. We have fixed the error by decreasing the number of threads per block. We still don’t know why it is happening but we suppose that resources per block is not enough. It is now working but we have another weird error on the way :)
Regarding to our last message we have found that this weird error is about Vista’s behaviour in restarting the graphics driver.
As stated in the
[url=“http://developer.download.nvidia.com/compute/cuda/2_3/toolkit/docs/cudatoolkit_release_notes_windows.txt”]http://developer.download.nvidia.com/compu...tes_windows.txt[/url]
vista has a behaviour which is stated in the
[url=“http://www.microsoft.com/whdc/device/display/wddm_timeout.mspx”]http://www.microsoft.com/whdc/device/displ...dm_timeout.mspx[/url].
We have realized that our renderer can run smooth on a smaller time period. But when it comes to longer renderings the device driver was restarting. When we have read about the issues in the CUDA 2.3 Toolkit we have realized that when gpu does not respond at least 2 seconds it was restarting. NVIDIA OpenCL is also effected by the case.
The case can be solved by using the reg below but windows may hang during a long kernel run.
Windows Registry Editor Version 5.00
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers]
“TdrLevel”=dword:00000000