|
In _Heterogeneous Computing with OpenCL_ by Gaster et. al. (p. 88) there is a recommendation that an OpenCL dispatch be an even multiple of the device's SIMD width and that this value can be obtained through the kernel object's property in the title of this post. I don't see its equivalent in the high level api of CLKernel for JOCL. Being somewhat new to OpenCL, would you please comment on whether this is used much in the implementation of algorithms?
|