saxpy.cuf runtime error on cuda-8-0

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi,

I’m just getting started with CUDA C and CUDA-Fortran. I’m starting by verifying that the simple CUDA-C and CUDA-Fortran examples hosted at nvidia would and execute correctly. However while the CUDA-C example generates the expected result for the example “Max Error = 0.000000”, the CUDA-Fortran code generates an incorrect result “Max Error = 2.00000”. The links to the progams are here, and my source code is below the links. I am running an intel i7-6800k, with 2 Geforce GTX 1060. Could there be something off with my cuda-8-0 distribution? I would like to note that the same source code file compiled and executed correctly on a different machine running cuda-7-5 (amd processor, 1 nvidia 560ti)

https://devblogs.nvidia.com/parallelforall/easy-introduction-cuda-c-and-c/ and https://devblogs.nvidia.com/parallelforall/easy-introduction-cuda-fortran/

module mathOps
contains
  attributes(global) subroutine saxpy(x, y, a)
    implicit none
    real :: x(:), y(:)
    real, value :: a
    integer :: i, n
    n = size(x)
    i = blockDim%x * (blockIdx%x - 1) + threadIdx%x
    if (i <= n) y(i) = y(i) + a*x(i)
  end subroutine saxpy
end module mathOps

program testSaxpy
  use mathOps
  use cudafor
  implicit none
  integer, parameter :: N = 40000
  real :: x(N), y(N), a
  real, device :: x_d(N), y_d(N)
  type(dim3) :: grid, tBlock

  tBlock = dim3(256,1,1)
  grid = dim3(ceiling(real(N)/tBlock%x),1,1)

  x = 1.0; y = 2.0; a = 2.0
  x_d = x
  y_d = y
  call saxpy<<<grid, tBlock>>>(x_d, y_d, a)
  y = y_d
  write(*,*) 'Max error: ', maxval(abs(y-4.0))
end program testSaxpy
2

Hi pmccarter17311,

The 16.10 CUDA Fortran compiler is using CUDA 7.0 by default. However, GTX 1060’s and other Pascal compute capability 6.0 devices require binaries to be built with CUDA 8.0. This mismatch is causing the kernel to fail. To fix, compile with “-Mcuda=8.0” or “-Mcuda=cc60”.

% pgfortran -V16.10 -fast -Mcuda saxpy.cuf
% a.out
 Max error:     2.000000
% pgfortran -V16.10 -fast -Mcuda=cc60 saxpy.cuf
% a.out
 Max error:     0.000000

Hope this helps,
Mat

3

Hi Mat,

Thank you for the reply. Adding he flag did fix the issue, I was able to generate the correct result. Thanks for the note on how to specify the compilation environment for Pascal GPUs.