Can the computation part of C++ project be transfered to CUDA code witout changing other part code i

Accelerated Computing CUDA CUDA Programming and Performance
1

Suppose I have a big program coding in C++, I know the main cost of this program is a big for loop, I want to compute the for loop on GPU, but don’t change other source files. How can I do it?

I use visual 2005 with CUDA integrated in. Is there anyway I can realize my goal? Or If I have to develop the whole project in CUDA and rewrite all the source file?

2

The answer is most likely “yes”.

However, you should have asked the question “How can the computation part of C++ project be transfered to CUDA code witout changing other part code in original project?” if you had wanted to know more. ;)

Only small changes to your main program may be required, for example a call into an external function that provides the same functionality as your for() loop, but based on CUDA. One relevant SDK sample is the cppintegration sample

Christian