Open mw66 opened 6 years ago
CL_DEVICE_TYPE_*
constants are defined in CL/cl.h
, which may be unavailable for non-opencl-based backends. How about providing using dummy vex::Filter::GPU
, vex::Filter::CPU
, vex::Filter::Accelerator
instead?
EDIT: those are already defined for the CUDA backend: https://github.com/ddemidov/vexcl/blob/master/vexcl/backend/cuda/filter.hpp#L53-L56
ctx = new vex::Context( vex::Filter::Type(CL_DEVICE_TYPE_GPU) && vex::Filter::DoublePrecision && vex::Filter::Count(1) );
I think vex should define a dummy func in this case, so the client code no matter which backend it choose to use.