Is MCUDA adapive to CPU? or I have to compile to CPU?

waiting for your comments on line

By MCUDA I am assuming you are referring to John Stratton’s implementation here .

Your title is not very descriptive, so I’ll just give an overview.

MCUDA is a source to source translator that converts CUDA augmented C into ANSI C with OpenMP pragmas. This allows the resulting code to be compiled by a C compiler for a multicore CPU. I think that NVIDIA announced at NVISION over the summer that they would support something like this in future releases of the nvcc, but I haven’t heard anything new since then.

I recently attented a CUDA tutorial by NVIDIA and they did mention that the latest release of nvcc (I think 2.1 beta or something) has this provision of compiling CUDA kernels into openMP for multicores.

The current beta does not include this option, although, yes, it was mentioned as being targeted for 2.1. (The feature may have slipped, but no one from NVIDIA has commented on this.)