Call to cuModuleLoadData returned error 209 :No binary for GPU

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

I reported this error when I started OpenAcc under PVF14.10, but it works fine when OpenAcc is not enabled. There is no problem in compiling. May I ask what went wrong? I installed cuda11.2 and PGI Visual fortran14.10 under windows.I opened OpenACC in Property-Language

program main
use openacc
implicit none
real,allocatable::v1(:)
integer length,idx
call acc_set_device_num(2,acc_device_nvidia)
length=102410241024
allocate(v1(length))
v1(1:length)=0.0
!$acc data copy(v1(1:length))
call add1(v1,length)
!$acc update host(v1(1:5))
print*,‘v1(1)=’,v1(1:5)
do idx=1,length
v1(idx)=v1(idx)+10
enddo
print*,‘v1(1)=’,v1(1:5)
!$acc update host(v1(1:5))
print*,‘v1(1)=’,v1(1:5)
!$acc end data
deallocate(v1)
end program
subroutine add1(vec,length)
use openacc
implicit none
integer,intent(in)::length
real,intent(inout)::vec(1:length)
integer idx
!$acc parallel loop present(vec(1:length))
do idx=1,length
vec(idx)=vec(idx)+idx
enddo
end subroutine

2

Hi 996648624,

There’s nothing wrong with your program. Most likely the problem is that given PVF 14.10 is almost 9 years old, it’s not able to support the GPU you have installed.

Try compiling with “-ta=gpu:nordc” which will generate PTX code that can be JIT compiled by the newer GPU. Given the age, I can’t say it will work, but worth a try.

We did stop supporting native Windows about three years ago. While we hope to bring it back at some point, it was severely delay due to the pandemic and then due to other high priority projects. For now, if you need to update your compiler, you’ll need to run under WSL2, which is Linux running under Windows.

-Mat

3

Thanks for your reply, how to try it under PVF compiling with “-ta=gpu:nordc”，Which option does it set in the PVF compiler.