The application should pick a certain OpenCL kernel implementation based on the current OpenCL platform/device. For now there will be three implementations:
for NVIDIA GPUs
for AMD GPUs
for any other platform/device. For a start, this one could be identical with the current kernel. We will fix it later if it does not fully comply with the OpenCL specification.
Assume the chosen platform/device does not change during one mdrun execution.
The application should pick a certain OpenCL kernel implementation based on the current OpenCL platform/device. For now there will be three implementations:
Assume the chosen platform/device does not change during one mdrun execution.