Feature is not yet implemented error What causes it?

Accelerated Computing CUDA CUDA Programming and Performance
1

In WinXP 32 I can run the already-compiled SDK assemblies fine.

However when I compile the assemblies (no editing at all) from the source code, I get “Feature is not yet implemented” error on certain lines. For instance, in DeviceQuery line:

CUDA_SAFE_CALL(cudaGetDeviceCount(&deviceCount));

What causes this error?

Thanks

2

Can you give us a test case that causes this as well as the nvcc command line you’re using?

3

Hi

Thanks for looking into this. I create the .exe by building devicequery.cu in MS Visual Studio 2005. Then I try to run the resulting devicequery.exe. This is the result:

C:\Program Files\NVIDIA Corporation\NVIDIA CUDA SDK\bin\win32\Debug>devicequery.

exe

Cuda error in file ‘deviceQuery.cu’ in line 53 : feature is not yet implemented.

I notice however that my .exe file requires cutil32D.dll whereas the one you provide with the SDK (which runs fine) requires cutil32.dll

4

Usually, this is caused by a mismatch between the toolkit and driver versions you have installed. Perhaps you have the pre-compiled CUDA 1.1 SDK and the corresponding driver, but are recompiling it with the CUDA 2.0 beta toolkit?

5

Well, it seems that you are right, although I’m not sure if is a version erro.

What’s happening is that the pre-compiled .exe files, which run fine, call for cudart.dll. However, when I compile the files myself the .exe files look for cudartD.dll.

Why would that happen? How do I make my files call for cudart.dll?

Thanks for your help!

6

Sorry, the files are cutil32.dll and cutil32D.dl, not cart…

Thanks