The webinar was interesting, and it is nice to finally have an open standard from Microsoft for a change.
It was said that Microsoft implements the functions through their VSC++ compiler via taking kernel codes inside the program, transcode it to HLSL, compile them into DX calls, store this byte data inside the x86 binary, and when the program is run, this bytecode is passed onto the display drivers, as if they were regular DX calls just spawned by the DX runtime, and let the driver to the binary translation to the specific HW as it is always done with DX.
It is only up to the compiler how it implements these functions, as I understood, there is no incompatibility between C++ AMP and OpenCL. All the AMP function calls can be implemented in C also. Infact that was my biggest disappointment. Kernels are only allowed to be passed and to be using basic datatypes (int, uint, float, double ONLY). I mean... come on. They boast about AMP being a C++ API, superior to C, but I cannot even pass structs to a kernel??? Even OpenCL is capable of that. This was my biggest disappointment, but please someone tell me if I understood it wrong. If this is true, it should really be called C-- AMP.
