I have code which uses the NVidia .cl files from the oclRadixSort example that i haven’t been able to get working with the AMD OpenCL implementation. Is it ok to provide AMD with the code so that they can dig into their binaries to determine what is causing the problem? Perhaps this isn’t the right forum… if so pls let me know.
On a more general note, do GPUs or their drivers have any notion of an application holding onto state/resources and, if so, the lifetime of that application? Resetting it to a known state before starting an application or when recovering from an error (ideally Visual Studio could do this automatically on termination of a debug session) would be nice.
Debugging OpenCL kernels would also be helpful. Nexus perhaps, or emudebug? Is the Nexus beta going to be generally available at some point? The “Getting the Right Answer” section of the OpenCL Best Practices Guide appears to be missing in my copy.