C------------------------------------------------------------------------------ C Example parallel program to show the use of the "pwgsmp" routine. C------------------------------------------------------------------------------ C.. This program can be obtained from: C C http://www.cs.umn.edu/~agupta/wsmp C C (C) IBM Corporation, 2004. C c Acceptance and use of this program constitutes the user's understanding c that he/she will have no recourse to IBM for any actual or consequential c damages, including, but not limited to, lost profits or savings, arising c out of the use or inability to use this program. C------------------------------------------------------------------------------ program pwgsmp_ex1 implicit none include 'mpif.h' integer mperr, myid call mpi_init_(mperr) if (mperr .ne. 0) then print *,'MPI initialization error' stop end if call mpi_comm_rank_(MPI_COMM_WORLD,myid,mperr) C.. This is a small matrix, so use only 1 thread per process. c NOTE: For real problems, use wsetmaxthrds only as needed. call wsetmaxthrds(1) C.. This program must be called on *at least* 3 nodes. if (myid .eq. 0) then call master0 else if (myid .eq. 1) then call master1 else if (myid .eq. 2) then call master2 else call slave (myid) end if call mpi_finalize_(mperr) stop end C------------------------------------------------------------------------------ subroutine master0 implicit none integer n, ldb, nrhs integer iparm(64) integer ia(4) integer ja(11) double precision dparm(64) double precision avals(11) double precision b(3) double precision rmisc ! just a placeholder in this program integer i double precision waltime, wsmprtc C.. Fill all arrays containing matrix data. data n /3/, ldb /3/, nrhs /1/ data ia /1,5,9,12/ data ja 1 /1, 3, 7, 8, 2 2, 3, 5, 9, 3 1, 3, 7 / data avals 1 /14.d0, -5.d0, -1.d0,-6.d0, 2 14.d0,-1.d0, -3.d0, -1.d0, 3 -1.d0, 16.d0, -2.d0 / C.. Executable part of the program starts here. waltime = wsmprtc() C.. Fill 'iparm' and 'dparm' arrays with default values. C.. As an alternative to this step, the values in 'iparm' and 'dparm' can be c filled with values suitable for the application either manually or by c using a "data" statement according to the description in the User's guide. iparm(1) = 0 iparm(2) = 0 iparm(3) = 0 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) print *,'Initialization complete in time - ',wsmprtc()-waltime if (iparm(64) .ne. 0) then print *,'The following ERROR was detected: ',iparm(64) return end if C.. This is a tiny problem, so use small granularity. c NOTE: Do not reset dparm(25) to a small value for real problems. dparm(25) = 1.d0 C.. Analysis. waltime = wsmprtc() iparm(2) = 1 iparm(3) = 1 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) print *,'Analysis complete in time - ',wsmprtc()-waltime if (iparm(64) .ne. 0) then print *,'The following ERROR was detected: ',iparm(64) return end if print *,'Number of expected nonzeros in factors = ',iparm(24) print *,'Number of expected FLOPS in factorization = ',dparm(24) C.. LU Factorization. waltime = wsmprtc() iparm(2) = 2 iparm(3) = 2 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) waltime = wsmprtc() - waltime print *,'LU complete in time - ',waltime print *,'Actual number of nonzeros in factors = ',iparm(23) print *,'Actual number of FLOPS in factorization = ',dparm(23) print *,'Factorization MegaFlops = ', + (dparm(23) * 1.d-6) / waltime if (iparm(64) .ne. 0) then print *,'The following ERROR was detected: ',iparm(64) return end if C.. Back substitution. do i = 1, n b(i) = 1.d0 end do waltime = wsmprtc() iparm(2) = 3 iparm(3) = 3 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) print *,'Back substitution done in time - ',wsmprtc()-waltime if (iparm(64) .ne. 0) then print *,'The following ERROR was detected: ',iparm(64) return end if C.. Itertative refinement. waltime = wsmprtc() iparm(2) = 4 iparm(3) = 4 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) print *,'Itertative ref. done in time - ',wsmprtc()-waltime print *,'Norm of residual - ',dparm(7) if (iparm(64) .ne. 0) then print *,'The following ERROR was detected: ',iparm(64) return end if print *,'The solution of the system is as follows:' do i = 1, n print *,i,' : ',b(i) end do return end C------------------------------------------------------------------------------ subroutine master1 implicit none integer n, ldb, nrhs integer iparm(64) integer ia(4) integer ja(10) double precision dparm(64) double precision avals(10) double precision b(3) double precision rmisc ! just a placeholder in this program integer i double precision waltime, wsmprtc C.. Fill all arrays containing matrix data. data n /3/, ldb /3/, nrhs /1/ data ia /1,3,6,11/ data ja 4 / 4, 8, 5 5, 6, 9, 6 2, 4, 6, 7, 8 / data avals 4 / 14.d0, -3.d0, 5 14.d0,-1.d0, -1.d0, 6 -2.d0, -1.d0, 16.d0,-2.d0,-4.d0 / C.. Executable part of the program starts here. C.. Fill 'iparm' and 'dparm' arrays with default values. C.. As an alternative to this step, the values in 'iparm' and 'dparm' can be c filled with values suitable for the application either manually or by c using a "data" statement according to the description in the User's guide. iparm(1) = 0 iparm(2) = 0 iparm(3) = 0 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. This is a tiny problem, so use small granularity. c NOTE: Do not reset dparm(25) to a small value for real problems. dparm(25) = 1.d0 C.. Analysis. iparm(2) = 1 iparm(3) = 1 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. LU Factorization. iparm(2) = 2 iparm(3) = 2 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Back substitution. do i = 1, n b(i) = 1.d0 end do iparm(2) = 3 iparm(3) = 3 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Itertative refinement. iparm(2) = 4 iparm(3) = 4 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return do i = 1, n print *,i+3,' : ',b(i) end do return end C------------------------------------------------------------------------------ subroutine master2 implicit none integer n, ldb, nrhs integer iparm(64) integer ia(4) integer ja(11) double precision dparm(64) double precision avals(11) double precision b(3) double precision rmisc ! just a placeholder in this program integer i double precision waltime, wsmprtc C.. Fill all arrays containing matrix data. data n /3/, ldb /3/, nrhs /1/ data ia /1,3,8,12/ data ja 7 / 4, 7, 8 1, 3, 5, 7, 8, 9 2, 3, 8, 9/ data avals 7 / -1.d0, 16.d0, 8 -3.d0, -4.d0, -3.d0, -4.d0,71.d0, 9 -1.d0,-2.d0, -4.d0,16.d0/ C.. Executable part of the program starts here. C.. Fill 'iparm' and 'dparm' arrays with default values. C.. As an alternative to this step, the values in 'iparm' and 'dparm' can be c filled with values suitable for the application either manually or by c using a "data" statement according to the description in the User's guide. iparm(1) = 0 iparm(2) = 0 iparm(3) = 0 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. This is a tiny problem, so use small granularity. c NOTE: Do not reset dparm(25) to a small value for real problems. dparm(25) = 1.d0 C.. Analysis. iparm(2) = 1 iparm(3) = 1 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. LU Factorization. iparm(2) = 2 iparm(3) = 2 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Back substitution. do i = 1, n b(i) = 1.d0 end do iparm(2) = 3 iparm(3) = 3 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Itertative refinement. iparm(2) = 4 iparm(3) = 4 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return do i = 1, n print *,i+6,' : ',b(i) end do return end C------------------------------------------------------------------------------ subroutine slave (myid) implicit none integer myid integer n, ldb, nrhs integer iparm(64) integer ia, ja double precision dparm(64) double precision avals, b ! just placeholders in this routine double precision rmisc ! just a placeholder in this program data n /0/, ldb /1/, nrhs /1/ C.. Fill 'iparm' and 'dparm' arrays with default values. iparm(1) = 0 iparm(2) = 0 iparm(3) = 0 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. This is a tiny problem, so use small granularity. c NOTE: Do not reset dparm(25) to a small value for real problems. dparm(25) = 1.d0 C.. Analysis. iparm(2) = 1 iparm(3) = 1 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. LU Factorization. iparm(2) = 2 iparm(3) = 2 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Back substitution. iparm(2) = 3 iparm(3) = 3 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return C.. Itertative refinement. iparm(2) = 4 iparm(3) = 4 call pwgsmp(n, ia, ja, avals, b, ldb, nrhs, rmisc, iparm, dparm) if (iparm(64) .ne. 0) return return end