Compiling the CULA example

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hello

I am having trouble compiling the example posted here: http://www.pgroup.com/lit/articles/insider/v2n3a5.htm

I do have an old CULA Dense installation (R13) where the libraries are in their old format… It seems that I am typing the commands wrong ( I even pasted all the libraries in to the folder where the cuda example directory is and these errors keep appearing)

Please inform me if my commands in PGI Cmd are wrong or lacking some flags…

It would be most helpful if you could tell me what to type exactly, I am really lost.

Thank you in advance

Ahmed

2

I would really appreciate guidance. I cannot move on further without your help.

Thank you

Ahmed

3

Please reply to me with more information you might need from me.

I have been in this tight spot for a couple of days. I do not understand why the compiler is not accepting the check_status_ reference in particular… are there any libraries I am not linking right?

I hope you could answer. I have reached a dead end.

I have also done the following to reduce the warnings and I got an exe file. But an error occurs when I run the cula.exe…please have a look:

        module cula_test
      
            use cudafor
            contains
            
                       
            ! cula test (device interface)
            subroutine do_cula_device_test(n,nrhs,ain,bin)
            
                ! input
                real,dimension(:,:) :: ain,bin
                
                ! allocations (all on host)
                real,dimension(:,:),allocatable :: a,b,ans
                integer n,nrhs,status
                integer c1,c2,cr,cm
                real norm
                
                ! gpu memory
                real,device,dimension(:,:),allocatable :: a_dev,b_dev
                integer,device,dimension(:),allocatable :: ipiv_dev                

                ! back up input for reconstruction test
                allocate( a(n,n), b(n,nrhs), ans(n,nrhs) )
                a(1:n,1:n) = ain
                b(1:n,1:nrhs) = bin                
      
                open (6, file='output.txt',         status='unknown')
          
                ! allocate gpu memory
                allocate( a_dev(n,n), b_dev(n,nrhs), ipiv_dev(n) )
                
                ! start test
                call system_clock( c1,cr,cm )
                print *, 'starting cula (device interface) test...'
                
                ! copy memory to gpu
                a_dev = a
                b_dev = b
                
                ! call cula solver (device interface)
                status = cula_device_sgesv(n,nrhs,a_dev,n,ipiv_dev,b_dev,n)
                
                ! copy answer to host
                b = b_dev
                
                ! stop test
                call system_clock( count=c2 )
                print *, '  runtime:', 1.e3*real(c2-c1) / real(cr), 'ms'
                print *, '  gflops:', (0.66*n**3.)/(real(c2-c1)/real(cr))/(1.e9)

                write(6,*) 'runtime:', 1.e3*real(c2-c1) / real(cr), 'ms'
                write(6,*) 'gflops:', (0.66*n**3.)/(real(c2-c1)/real(cr))/(1.e9)
                pause
                ! cleanup
                deallocate(a,b,ans)
                deallocate(a_dev,b_dev,ipiv_dev)
                
            end subroutine do_cula_device_test
            
        end module cula_test
        
        ! main program

        program cula

            use cula_test
                    
            real error,eps
            
            ! Host memory
            real,dimension(:,:),allocatable :: a, b
            integer n, info, i, j, status

            n = 10000
            nrhs = 1

            print *,'cula + pgfortran test (matrix solve)'
            print *,'  array size: ', n, ' by ', n
            print *,'  right hand sides: ', nrhs
            print *,''
            allocate( a(n,n), b(n,nrhs) )
            
            ! intialize a and b
            call random_number(a)
            call random_number(b)                      
            
            ! Make sure a() isn't singular
            do i=1,n
                a(i,i) = 10. * a(i,i) + 10.
            enddo
            pause           
            ! initialize cula
            status = cula_initialize()

            pause
            ! do gpu test (device interface)
            call do_cula_device_test(n,nrhs,a,b)
            pause
 
        end program cula

compiled with: pgfortran cula_.cuf -lcula_pgfortran

and when I run the Cula.exe file I get this:

0: ALLOCATE: 400000000 bytes requested: status = 49(unspecified driver error)

I am really counting on you to give me a push.

Ahmed

4

FYI I tried the host interface part of this PGI example for CULA and it worked really well however it seems that the device interface gives me an allocation error I have never seen before.

Ahmed

5

I solved my problems…

It would be useful to do an article in the Insider on the new way to call CULA though. I am a new user and I was lost for a couple of days until I found the solution through trial and error.

It would be essential for other people yet to understand it.

Ahmed

6

The article you’re quoting is several years old and EM Photonics has since updated their interfaces. There’s a new article at: Account Login | PGI

For questions regarding CULA, it’s best to ask EM Photonics directly.

  • Mat