CUDA & the stl

Accelerated Computing CUDA CUDA Programming and Performance
1

Is it possible to use STL classes from host functions or to even #include an STL file in an nvcc-compiled source file?

The following cuda code generates all kinds of compiler errors:

#include <vector>

void test(void)

{

 ;

}

errors:

Compiling...

"C:\Program Files\Microsoft Visual Studio 8\VC\include\yvals.h", line 580: error: 

          identifier "__thiscall" is undefined

      explicit __thiscall _Lockit(); // set default lock

...

"C:\Program Files\Microsoft Visual Studio 8\VC\include\yvals.h", line 731: error: 

          member function with the same name as its class must be a constructor

      __thiscall _Mutex();

... etc
2

Encountered same problems here… does nvcc support generic programming? Also, when there was exception handling, nvcc thrown this error:

"C:\Program Files\Microsoft Visual Studio 8\VC\include\xlocale", line 369: error: 

          support for exception handling is disabled

    throw;

    ^
3

As the manual says, nvopencc is a C compiler. C++ stuff like templates, exceptions, etc will not work.

However, as you can see with nvcc -v there is quite a bit of preprocessing going on using cl or gcc. So you can use C++ constructs that can be resolved all the way down to pure C functions in the preprocessing. An example is the functions templates used in the CUDA includes and SDK examples.

Peter

4

Dear Peter,

Does that mean generic class and exceptions won’t work at all?

Did the manual say that preprocessing is done by cudafe, not cl? BTW I have tried to switch it by -ncfe, but I didn’t help at all.

Or is there exists some “preprocessor” that can resolve all C++ constructs to C constructs?

Thank you very much.

5

Correct.

The preprocessing uses both cl/gcc and cudafe. Do check whats going on with nvcc -v

Peter