when I build my project on GPU, I got the error"CL_BUILD_PROGRAM_FAILURE", "Linked failed". However, when the program is built on CPU, it passed. What's that problem?
Could you post the system configuration?
- HW: AthlonX2 (Brisbane -> with SSE3 support) / 4650
- Stream2-Beta4 (including the special graphics driver)
- with or without use of the Stream SDK
- with $ATISTREAMSDKROOT set to its directory and the 'bin' subdirectory as first directory in $Path
- with the Kernel from the HelloCL-Demo hardcoded in the binary.
- build for a single device (the only GPU device)
- build with no build options (clBuildProgram)
- build with no ntfy function (clBuildProgram)
Again: The code works fine for CPU-Targets. The only difference is the CL_DEVICE_TYPE_GPU instead of ...CPU while the creation of the context or the device list which is used to create the context.
Finally if you still dont trust my code: The HelloCL-Sample from the SDK is also affected if i change it to CL_DEVICE_TYPE_GPU.
I had a similar problem on running some opencl code developed on nvidia's impl on ati's impl: "Link Error". It appears "global" __constant definitions when converted to #defines cured the problem.
Retrieving data ...