#pragma inside macro

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi,

I want to define,

#define PRAGMA_ACC_KERNELS #pragma acc kernels

But that is not working. Is it supported in OpenACC?

2

Hi hyuzuguzel,

This isn’t a legal C macro definition. Instead, you need to use the C99 “_Pragma” operator to define a pragma as part of a macro.

Try:

#define PRAGMA_ACC_KERNELS _Pragma("acc kernels")

Hope this helps,
Mat

3

Hi,

I’ve already tried your suggestion but the project did not compile. It gives the following error:

error: identifier “_Pragma” is undefined
PRAGMA_ACC_KERNELS

I also tried __pragma(“acc kernels”). With this the project compiles but it doesn’t recognizes the pragma and thus doesn’t create the kernel.

Best regards,
H.

4

Can you please provide more details such as compiler version and OS, as well as a reproducing example?

Given the error message, it seems you’re using C++? PGI’s C++ compiler will recognize the C99 _Pragma operator.

% cat testPragma.cpp
#include <iostream>
#define PRAGMA_ACC_KERNELS _Pragma("acc kernels loop")
#define N 1024

int main() {
  int A[N];
  int i;
  PRAGMA_ACC_KERNELS
  for (i=0; i<N; ++i) {
     A[i]=i;
  }
  std::cout << "A[1]=" << A[1] << std::endl;
  return 0;
}

% pgc++ testPragma.cpp -acc -Minfo=accel
main:
      9, Loop is parallelizable
         Accelerator kernel generated
          9, #pragma acc loop gang, vector(128) /* blockIdx.x threadIdx.x */
      9, Generating copyout(A[:])
         Generating Tesla code
5

I am using Windows 7 and “pgcpp 14.10-0 64-bit target on x86-64 Windows -tp haswell The Portland Group - PGI Compilers and Tools” compiler. I also tried the testpragma.cpp but I again get the same error. I compile with →

pgcpp testPragma.cpp -acc -Minfo=accel

6

Ah, in that case you need to add the “–c++0x” flag to get the appropriate language support.

PGI$ pgcpp testpragma.cpp  -acc -Minfo=accel --c++0x
main:
     10, Loop is parallelizable
         Accelerator kernel generated
         10, #pragma acc loop gang, vector(128) /* blockIdx.x threadIdx.x */
         Generating present_or_copyout(A[:])
         Generating Tesla code

Alternatively you can use the MSVC++ “__pragma” operator.

#define PRAGMA_ACC_KERNELS __pragma(acc kernels loop)
7

Thanks, now it works.