Fortran 語言MPI 平行計(jì)算程式設(shè)計(jì)

Fortran

MPI

I

coGinco

MPI

IBMSP2MPI

PCClusterMPI

n

nCPU

CPU

structuredgrid:

unstructuredgrid:

ParallelProcessingof1-D

Arrays-withoutdatapartition

1=1...........50

■IIIPO

1=51....100

.IIIP1■

1=101...150

IIIP2

1=151...200

■_________充」P3

DO,200_____>DOl=ISTART,IEND

A(I)=B(I)+C(I)*D(I)A(I)=B(I)+C(I)*D(I)

ENDDOENDDO

ParallelProcessingof1-D

Arrays-withdatapartition

占1.........501.........501..........501..........50localindex

「1IIII1I

II1I「III

占1.........5051....100ioi...150151...200globalindex

DO1=1,200DOl=ISTART,IEND

A(I)=B(I)+C(I)*D(I)AA(I)=B(I)+C(I)*D(I)

ENDDOENDDO

Parallelon2ndindexof2-D

Arrays-withoutdatapartition

J=50

J=1

1=1...8

DOJ=1,200DOJ=JSTART,JEND

DO1=1,8DO1=1,8

A(I,J)=B(I,J)+C(I,J)*D(I5J)AA(I5J)=B(IJJ)+C(I5J)*D(I,J)

ENDDOENDDO

ENDDOENDDO

Parallelon2ndindexof2-D

Arrays-withdatapartition

J=501J=501J=501J=501

心II心II心II心II

1=1...81=1...81=1...81=1...8

P1P1P2P3

DOJ=1,200DOJ=JSTART,JEND

DO1=1,8DO1=1,8

A(I,J)=B(I,J)+C(I,J)*D(I5J)AA(I5J)=B(IJJ)+C(I5J)*D(I,J)

ENDDOENDDO

ENDDOENDDO

Partitionon3rdindexof

3-DArrays

1=1...81=1...81=1...81=1...8

POP1P2P3

x(8,24,50)x(8,24,50)x(8,24,50)x(8,24,50)

y(8,24,50)y(8,24,50)y(8,24,50)7(8,24,50)

z(8,24,50)z(8,24,50)z(8,24,50)z(8,24,50)

DataRecursive

Parallelizableloops

DOJ=1,N

DO占1,M

Y(l,J)=0.25*(X(l-1,J)+X(I+1,J)+X(l,J-1)+X(l,J+1))+H*F(l,J)

ENDDO

ENDDO

Datarecursive-cannotbeparallelized

DOJ=1,N

DO占1,M

X(l,J)=0.25*(X(l-1,J)+X(I+1,J)+X(l,J-1)+X(l,J+1))+H*F(l,J)

ENDDO

ENDDO

MPI

IBMSP2MPI

PCClusterMPI

MPI

■MPI(MessagePassingInterface)

■FortranCC++

■ArgonneNationalLab

MPICH

■/mpi/mpich

■

MPI

IBMSP2MPI

PCClusterMPI

■MPI

>IBMp6905IBMSP2SMP,SGI03800,

HPsuperdome

■MPICH:PCcluster

■:IBMp690,IBMSP2SMP,

HPsuperdome,PCcluster

■(jobscheduler);

>IBM:LoadLeveler

>HPsuperdome:LSF

>PCcluster:DQSlike

MPI

IBMSP2MPI

PCClusterMPI

IBMSP2,p690MPI

■mpxlf-03-qarch=auto-qstrict-ofile.xfile.f

■llclass:CPUavailabilityforeachqueueclass

■llsubmitjob_command_file:submitabatchjob

■llq:jobqueuestatus

■llcanceljobid:deleteasubmittedjob

IBMSP2Jobcommandfile

■#!/bin/csh

■#@executable=/usr/bin/poe

■#@network.mpi=cssO,shared,us

■#@arguments=/working_dir/file.x

■#@output=outp4

■#@error=outp4

■#@job_type=parallel

■#@class=medium

■#@tasks_per_node=4

■#@node=1

■#@queue

IBMp690Jobcommandfile

■#!/bin/csh

■#@executable=/usr/bin/poe

■#@network,mpi=csss,shared,us

■#@arguments=/working_dir/file.x

■#@output=outp8

■#@error=outp8

■#@job_type=parallel

■#@class=8cpu

■#@tasks_per_node=8

■#@node=1

■#@queue

IBMSP2-llclass

■NameMaxJobCPUMaxProcCPUFreeMax

■d+hh:mm:ssd+hh:mm:ssSlotsSlots

■interactive-10+06:00:0088

■short-10+12:00:0046

■medium-11+00:00:0016128

■bigmem-12+00:00:001216

■mono-17+00:00:0011

IBMSP2-llq:jobqueuestatus

jobjduserjdsubmittedstatpriorclassrunningon

ivory2.4371.0u11kcj0010/219:25R50mediumivory33

ivory3.1285.0u32ltk0010/223:47R50bigmemivory41

ivory2.4431.0y19mkh0010/307:05R50bigmemivory41

ivory2.4348.0uOOjimOO10/206:49R50mediumivory13

ivory2.4437.0ullkcjOO10/310:08R50shortivory6

ivory3.1269.0u50lun0010/209:2850medium

ivory2.4368.0u07ish0010/219:1750medium

MPI

IBMSP2MPI

PCClusterMPI

PCClusterMPI

■mpif77-03-ofile.xfile.f

■qsub32job_command_file:submitabatchjob

■qstat32:jobqueuestatus

■qdel32jobid:deleteasubmittedjob

PCCluster:DQSjobqueuestatus

qstat32

userjdjobnameCPUsjobidjob_statussubmittedtime

cOOtchOOHUP4hpcsOI620:1rRUNNING02/26/9910:51:23

cOOtchOOHUP4hpcs02620:1rRUNNING02/26/9910:51:23

cOOtchOOHUP4hpcs03620:1rRUNNING02/26/9910:51:23

cOOtchOOHUP4hpcs04620:1rRUNNING02/26/9910:51:23

--PendingJobs

cOOtchOORAD5700:2QUEUED02/26/9919:24:32

PCClusterjobcommandfile

■#!/bin/csh

■#$-Iqty.eq.4,FastEthernet

■#$-NHUP4

■#$-Auserjd

■#$-cwd

■#$-jy

■cat$HOSTS_FILE>MPI_HOST

■mpirun-np4-machinefileMPI_HOSThubksp>&outp4

E

■MPI

MPIJNITMPI_FINALIZE

MPI_COMM_SIZEMPI_COMM_RANK

MPI_SENDMPI_RECV

■T2SEQ

■T2CP

■MPI_SCATTERMPI_GATHER

MPI_REDUCEMPI_ALLREDUCE

■T2DCP

mpif.hMPI(1)

INCLUDE'mpif.h'

REAL*8...

INTEGER...

CALLMPIJNIT(IERR)

■■■

CALLMPI_FINALIZE(IERR)

STOP

END

MPI(2)

■CALLMPI_COMM_SIZE

(MPI_COMM_WORLD,NPROC,IERR)

NPROCisthenumberofCPUsforthisprogram

■CALLMPI_COMM_RANK

(MPI_COMM_WORLD,MYID,IERR)

MYIDismyCPUid,countfromzero

MPI⑶

■PROGRAMT2CP

.PARAMETER(..)

.INCLUDE,mpif.h'

■REAL*8...

.INTEGERNPROC,MYID

■CALLMPIJNIT(IERR)

.CALLMPI_COMM_SIZE(MPI_COMM_WORLD,NPROC,IERR)

.CALLMPI_COMM_RANK(MPI_COMM_WORLD,MYID,IERR)

■...

.CALLMPI_FINALIZE(IERR)

.STOP

.END

MPI

CALLMPI_SEND(DATA,ICOUNT,DATATYPE,

IDEST,ITAG,MPI_COMM_WORLD,IERR)

CALLMPI_RECV(DATA,ICOUNT,DATA_TYPE,

ISRC,ITAG,MPI_COMM_WORLD,ISTAT,

IERR)

INTEGERISTAT(MPI_STATUS_SIZE)

MPI

MPIdatatypesFortrandatatype

MPI_CHARACTERCHARACTER

MPI_LOGICALLOGICAL

MPIJNTEGERINTEGER

MPI_REALREAL

MPI_REAL8REAL*8

MPI_COMPLEXCOMPLEX

MPI_COMPLEX16COMPLEX*16

MPI

MPIJNITMPI_FINALIZE

MPI_COMM_SIZEMPI_COMM_RANK

MPI_SENDMPI_RECV

T2SEQ

■T2CP

■MPLSCATTERMPI_GATHER

MPI_REDUCEMPLALLREDUCE

T2DCP

T2SEQ

PROGRAMT2SEQ

PARAMETER(NTOTAL=200)

REAL*8A(NTOTAL),B(NTOTAL),C(NTOTAL),D(NTOTAL),SUMA

OPEN(7,FILE='input.dat',STATUS=,OLD',FORM='UNFORMATTED')

READ(7)B

READ(7)C

READ(7)D

SUMA=0.0

DOl=1,NTOTAL

A(I)=B(I)+C(I)*D(I)

SUMA=SUMA+A(I)

ENDDO

WRITE1,101)(A(I),I=1,NTOTAL,5)

WRITE1,102)SUMA

STOP

END

T2SEQ

10.0003.0562.5622.3832.2902.2342.1962.1682.1482.131

2.1182.1082.0992.0912.0852.0792.0742.0702.0662.063

2.0602.0572.0542.0522.0502.0482.0462.0442.0432.041

2.0402.0392.0372.0362.0352.0342.0332.0322.0312.031

SUMofarrayA=.43855E+03

MPI

MPIJNITMPI_FINALIZE

MPI_COMM_SIZEMPI_COMM_RANK

MPI_SENDMPI_RECV

T2SEQ

■T2CP

■MPLSCATTERMPI_GATHER

MPI_REDUCEMPLALLREDUCE

T2DCP

istartiendntotal

cpuO

istartiendntotal

cpu1

istartiendntotal

cpu2

istartiend

cpu3

arrayelementinsideterritoryarrayelementoutsideterritory

T2CP(1)

PROGRAMT2CP

INCLUDE'mpif.h'

PARAMETER(NTOTAL=200)

REAL*8A(NTOTAL),B(NTOTAL),C(NTOTAL),D(NTOTAL),SUMA

INTEGERNPROC,MYID,ISTAT(MPI_STATUS_SIZE)5ISTART,IEND,

1COMM,GCOUNT(0:7),GSTART(0:7),GEND(0:7)

CALLMPIJNIT(IERR)

CALLMPI_COMM_SIZE(MPI_COMM_WORLD,NPROC,IERR)

CALLMPI_COMM_RANK(MPI_COMM_WORLD,MYID,IERR)

CALLSTARTEND(NPROCJ,NTOTAL,GSTART,GEND,GCOUNT)

ISTART=GSTART(MYID)

IEND=GEND(MYID)

PRINT*；NPROC,MYID,ISTART,IEND」，NPROC,MYID,ISTART,IEND

COMM=MPI_COMM_WORLD

T2CP(2)

IF(MYID.EQ.O)THEN

OPEN(7,NLE='input.dat',STATUS='OLD',FORM='UNFORMATTED')

READ(7)B

READ(7)C

READ(7)D

DOIDEST=1,NPROC-1

IST1=GSTART(IDEST)

KNT1=GCOUNT(IDEST)

CALLMPI_SEND(B(IST1),KNT1,MPI_REAL8,IDEST,10,COMM5IERR)

CALLMPI_SEND(CQST1),KNT1,MPI_REAL8,IDEST,20,COMM,IERR)

CALLMPI_SEND(D(IST1),KNT1,MPI_REAL8,IDEST,30.COMM.IERR)

ENDDO

ELSE

KNT=GCOUNT(MYID)

CALLMPI_RECV(B(ISTART),KNT,MPI_REAL8,0,10,COMM,ISTAT,IERR)

CALLMPI_RECV(C(ISTART),KNT,MPI_REAL8,0,20,COMM,ISTAT,IERR)

CALLMPI_RECV(D(ISTART),KNT,MPI_REAL8,0,30,COMM,ISTAT,IERR)

ENDIF

T2CP(3)

CCDOl=1,NTOTAL

DOl=ISTART,IEND

A(I)=B(I)+C(I)*D(I)

ENDDO

ITAG=110

IF(MYID.NE.O)THEN

KNT=GCOUNT(MYID)

CALLMPI_SEND(A(ISTART),KNT,MPI_REAL8,0,ITAG,COMM,IERR)

ELSE

DOISRC=1,NPROC-1

IST1=GSTART(ISRC)

KNT1=GCOUNT(ISRC)

CALLMPI_RECV(A(IST1),KNT1,MPI_REAL8,ISRC,ITAG,COMM,

ISTAT,IERR)

ENDDO

ENDIF

T2CP(4)

IF(MYID.EQ.O)THEN

WRITE(*,101)(A(I),I=1,NTOTAL,5)

SUMA=0.0

DOl=1,NTOTAL

SUMA=SUMA+A(I)

ENDDO

WRITE(*,102)SUMA

ENDIF

101FORMAT(10F8.3)

102FORMAT('SUMofarrayA=',E15.5)

CALLMPLFINALIZE(IERR)

STOP

END

T2CP(5)

SUBROUTINESTARTEND(NPROC,IS1,IS2,ISTART)IEND,ICOUNT)

INTEGERNPROC,IS1,IS2,ISTART(0:31),IEND(0:31),ICOUNT(0:31)

ILENGTH=IS2-IS1+1

IBLOCK=ILENGTH/NPROC

IR=ILENGTH-IBLOCK*NPROC

DOID=0,NPROC-1

IF(ID.LT.IR)THEN

ISTART(ID)=IS1+ID*(IBLOCK+1)

IEND(ID)=ISTART(ID)+IBLOCK

ELSE

ISTART(ID)=IS1+ID*IBLOCK+IR

IEND(ID)=ISTART(ID)+IBLOCK-1

ENDIF

IF(ILENGTH.LT.1)THEN

ISTART(ID)=1

IEND(ID)=0

ENDIF

ICOUNT(ID)=IEND(ID)-ISTART(ID)+1

ENDDO

END

-IL.1

T2CP

ATTENTION:0031-4084nodesallocatedbyLoadLeveler,continuing...

NPROC,MYID,ISTART)IEND=42101150

NPROC,MYID,ISTART)IEND=4151100

NPROC,MYID,ISTART5IEND=43151200

NPROC,MYID,ISTARTJIEND=40150

10.0003.0562.5622.3832.2902.2342.1962.1682.1482.131

2.1182.1082.0992.0912.0852.0792.0742.0702.0662.063

2.0602.0572.0542.0522.0502.0482.0462.0442.0432.041

2.0402.0392.0372.0362.0352.0342.0332.0322.0312.031

SUMofarrayA=.43855E+03

SPMD(SingleProgramMultipleData)

1.UseMYIDtocontroltheblockedIFstatement

2.UseISTART,IENDtocomputetheassignedrangeofarrays

3.Executeallstatementsotherwise

CALLMPIJNIT(IERR)

CALLMPI_COMM_SIZE(MPI_COMM_WORLD5NPROC,IERR)

CALLMPI_COMM_RANK(MPI_COMM_WORLD,MYID,IERR)

CALLSTARTENDiNPROC,1,NTOTAL,GSTART,GEND,GCOUNT)

ISTART=GSTART(MYID)

IEND=GEND(MYID)

PRINTNPROC,MYID,ISTARTJEND」，

NPROC,MYID,ISTART,IEND

MPI

MPIJNITMPI_FINALIZE

MPI_COMM_SIZEMPI_COMM_RANK

MPI_SENDMPI_RECV

T2SEQ

■T2CP

■MPI_SCATTERMPI_GATHER

MPI_REDUCEMPI_ALLREDUCE

■T2DCP

MPLSCATFER

CPIIOTT1T2T3T4CPUOB回

CPU1---------------------->CPU1B叵

SCATTER

CPU2CPU2B回

CPU3CPU3B"

IROOT=0

CALLMPLSCATTER(T,N,MPI_REAL8,B,N,MPI_REAL8,

&IROOT,MPI_COMM_WORLD5IERR)

MPIGATHER

CPIIOTIT1IT21T31T4CPUOB回

CPU1CPU1B叵

GATHER

CPU2<----------------CPU2B時(shí)

CPU3CPU3B"

IDEST=O

CALLMPI_GATHER(A,N,MPI_REAL8,T,N,MPI_REAL8,

&IDEST,MPI_COMM_WORLD,IERR)

MPIALLGATHER

CPUOA回

CPU1A叵

ALLGATHER

CPU2A同

CPU3A叵

CALLMPI_ALLGATHER(A,N,MPI_REAL8,T,N,MPI_REAL8,

&MPI_COMM_WORLD,IERR)

MPI_REDUCE

CPUOSUMA返CPUOGSUM1-7

CPU1SUMA巫CPU1

REDUCE

CPU2SUMA叵---------------------?CPU2

CPU3SUMA國(guó)CPU3

IROOT=0

KOUNT=1

CALLMPI_REDUCE(SUMA,GSUM,KOUNT,MPI_REAL8,

&MPI_SUM1ROOT,MPI_COMM_WORLDIERR)

-------------------------------5---------------=--------=------------5------7-

MPI_ALLREDUCE

CPUOSUMA返CPUOGSUM1-7

CPU1SUMA巫CPU1GSUMLQ7

ALLREDUCE

CPU2SUMA叵CPU2GSUM

CPU3SUMA國(guó)CPU3GSUM1-7

KOUNT=1

CALLMPI_ALLREDUCE(SUMA,GSUM,KOUNT,MPI_REAL8,

&MPI_SUM,MPI_COMM_WORLD,IERR)

MPIReductionFunction

MPIOperationDatatype

MPI_SUMsumMPIJNTEGER,MPI_REAL,MPI_REAL8,

MPI_PRODproductMPLCOMPLEX,MPi_COMPLEX16

MPI_MAXmaximumMPIJNTEGER,MPI_REAL,

MPI_MINminimumMPI_REAL8

MPI_MAXLOCmaxvalue&locationMPI_2INTEGER,MPI_2REAL,

MPI_MINLOCminvalue&locationMPI_2REAL8

MPI_LANDlogicalANDMPI_LOGICAL

MPI_LORlogicalOR

MPI_LXORlogicalexclusiveOR

MPI_BANDbinaryANDMPIJNTEGER,MPI_BYTE

MPI_BORbinaryOR

MPI_BXORbinaryexclusiveOR

T2DCP⑴

PROGRAMT2DCP

C

CData&ComputationalPartitionUsingMPI_SCATTER,MPI_GATHER

CNP=4mustbemodifiedwhenrunonotherthan4processors

C

PARAMETER(NTOTAL=200,NP=4,N=NTOTAL/NP)

INCLUDE'mpif.h'

REAL*8A(N),B(N),C(N),D(N),T(NTOTAL),SUMA,GSUM

INTEGERNPROC,MYID,ISTAT(MPI_STATUS_SIZE),ISTART,

IEND,COMM~~

CALLMPIJNIT(IERR)

CALLMPI_COMM_SIZE(MPI_COMM_WORLD,NPROC,IERR)

CALLMPI_COMM_RANK(MPi_COMM_WORLD,MYID,IERR)

PRINT*；"PROG^YIDWNPROC,MYID

T2DCP(2)

IF(MYID.EQ.O)THEN

OPEN(7,FILE='input.dat\STATUS='OLD',FORM='UNFORMATTED')

READ(7)T!readarrayB

ENDIF

IROOT=0

CALLMPI_SCATTER(T,N,MPI_REAL8,B,N,MPI_REAL8,

&IROOT,MPI_COMM_WORLD,IERR)

IF(MYID.EQ.O)THEN

READ⑺T!readarrayC

ENDIF

CALLMPLSCATTER(T,N,MPI_REAL8,C,N,MPI_REAL8,

&IROOT,MPI_COMM_WORLD,IERR)

IF(MYID.EQ.O)THEN

READ(7)T!readarrayD

ENDIF

CALLMPLSCATTER(T,N,MPI_REAL8,D,N,MPI_REAL8,

&IROOT,MPI_COMM_WORLD,IERR)

C

T2DCP(3)

SUMA=0.0

CDOl=1,NTOTAL

DO1=1,N

A(I)=B(I)+C(I)*D(I)

SUMA=SUMA+A(I)

ENDDO

IDEST=0

CALLMPI_GATHER(A,N,MPI_REAL8,T,N,MPI_REAL8,

&IDEST,MPI_COMM_WORLb,IERR)

CALLMPI_REDUCE(SUMA,GSUM,1,MPI_REAL8,MPI_SUM,

&IDEST,MPI_COMM_WORLD,IERR)

IF(MYID.EQ.O)THEN

WRITE(*,101)(T(I),I=15NTOTAL,5)

WRITE1,102)GSUM

ENDIF

101FORMAT(10F8.3)

102FORMAT('SUMofarrayA=',E15.5)

CALLMPI_FINALIZE(IERR)

STOP

END

T2DCP

ATTENTION:0031-4084nodesallocatedbyLoadLeveler,continuing...

NPROC,MYID=40

NPROC,MYID=41

NPROC,MYID=42

NPROC,MYID=43

10.0003.0562.5622.3832.2902.2342.1962.1682.1482.131

2.1182.1082.0992.0912.0852.0792.0742.0702.0662.063

2.0602.0572.0542.0522.0502.0482.0462.0442.0432.041

2.0402.0392.0372.0362.0352.0342.0332.0322.0312.031

SUMofarrayA=.43855E+03

I

■MPI_SENDRECVMPI_BCAST

T3SEQ

T3CP

T3DCPJ

T3DCP2

MPISENDRECV

ITAG=110

CALLMPI_SENDRECV

(B(IEND),ICOUNT,DATA_TYPEJDESTJTAG,

B(ISTARTM1),IC0UNT)DATA_TYPEJSRC,ITAG5

MPI_COMM_WORLD,ISTATUS,IERR)

MPIBCAST

CPIIOBIB1|B2|B3|B4

CPU1

MPI_BCAST

CPU2--------------------->

CPU3

IROOT=0

CALLMPI_BCAST(B,ICOUNT,DATA_TYPE,IROOT,

&MPI_COMM_WORLD,IERR)

■MPI_SENDRECVMPI_BCAST

T3SEQ

T3CP

T3DCP_1

T3DCP2

T3SEQ(1)

PROGRAMT3SEQ

PARAMETER(NTOTAL=200)

REAL*8A(NTOTAL),B(NTOTAL),C(NTOTAL),D(NTOTAL),AMAX

OPEN(7,FILE='input.dat',STATUS='OLD',FORM='UNFORMATTED')

READ(7)B

READ(7)C

READ(7)D

AMAX-1.D12

DOl=2,NTOTAL-1

A(I)=C(I)*D(I)+(B(I-1)+2.0*B(l)+B(l+1))*0.25

AMAX=MAX(AMAX,A(I))

ENDDO

T3SEQ(2)

WRITER,101)(A(I),I=1,NTOTAL,5)

WRITE(*,102)AMAX

101FORMAT(10F8.3)

102FORMAT('MAXIMUMVALUEOFAARRAYis',E15.5)

STOP

END

■MPLSENDRECVMPI_BCAST

T3SEQ

T3CP

T3DCPJ

T3DCP_2

eftH

尸mpi_proc_nulliend1

：iend

；Iiend+1ntotal

V1

cpuO

1,1八A

1istart2?iendl

istart?iendntotal

Vi|iend+1

cp■u1....II

:iend

Iistart

lstart-1istart2!ien°1iend+1

1

cpu2I

lstart-1istart2mpijproc_null

rightisowneddataisexchangeddata

loopindex(1)

AMAX=-1,D12

DOl=2,NTOTAL-1

A(l)=C(l)*D(l)+(B(l-1)+2.0*B(l)+B(l+1))*0.25

AMAX=MAX(AMAX,A(I))

ENDDO

■indexI2NTOTAL-1

CPUO2

CPUistart

istart2

loopindex(2)

ISTART2=ISTART

IF(MYID.EQ.O)ISTART2=2

■loopCPU

NTOTAL-1iend-1

CPUiend

iendl

IEND1=IEND

IF(MYID.EQ.NPROC-I)IEND1=IEND-1

ISTARTM1=ISTART-1

IENDP1=IEND+1

L_NBR=MYID-1

FTNBR=MYID+I

IFfMYID.EQ.O)L_NBR=MPI_PROC_NULL

IF(MYID.EQ.NPROC-1)R_NBR=MPI_PRO2NULL

ITAG=110

CALLMPI_SENDRECV(B(IEND),1,MPI_REAL8,R_NBR,ITAG,

1-B(ISTARTM1),1,MP匚REAL8,L[NBR,ITAG,

2MPI_COMM_WORLD；ISTATUS,正RR)

ITAG=120__

CALLMPI_SENDRECV(B(ISTART),1,MPI_REAL8,L_NBR,ITAG,

1B(IENDP1),1,MPI_REAL8,R_NBR,ITAG,

2MPI_COMM_WOF^D,ISTATUS,IERR)

CALLMPI_ALLREDUCE(AMAX,GMAX,1,MPI_REAL8,MPI_MAX,

1MPI_COMM_WORLD,TERR)

T3CP(1)

PROGRAMT3CP

PARAMETER(NTOTAL=200)

INCLUDE'mpif.h'

REAL*8A(NTOTAL),B(NTOTAL),C(NTOTAL),D(NTOTAL),AMAX,GMAX

INTEGERNPROC,MYID,ISTAT(MPI_STATUS_SIZE),ISTART,IEND,

1L_NBR,R_NBR,COMM,GSTART(0:31)5GEND(0:31),GCOUNT(0:31)

CALLMPIJNIT(IERR)

CALLMPI_COMM_SIZE(MPI_COMM_WORLD,NPROC,IERR)

CALLMPI_COMM_RANK(MPi_COMM_WORLD,MYID,IERR)

CALLSTARTEND(NPROC,1,NTOTAlZGSTART,GEND,GCOUNT)

ISTART=GSTART(MYID)

IEND=GEND(MYID)

PRINTNPROC,MYID,ISTART,IEND=',NPROC,MYID,ISTART,IEND

T3CP(2)

COMM=MPI_COMM_WORLD

LASTP=NPR0C-1

ISTARTM1=ISTART-1

IENDP1=IEND+1

ISTART2=ISTART

IF(MYID.EQ.O)ISTART2=2

IEND1=IEND

IF(MYID.EQ.LASTP)IEND-=IEND-1

L_NBR=MYID-1

R_NBR=MYID+1

IF[MYID.EQ.O)L_NBR=MPI_PROC_NULL

IF(MYID.EQ.NPROC-1)R_NBR=MPLPROC_NULL

IF(MYID.EQ.O)THEN

,,,

OPEN(75FILE='input.dat,STATUS=OLD)FORM='UNFORMATTED')

T3CP(3)

READ(7)B

READ(7)C

READ(7)D

DOIDEST=LASTP

IST1=GSTART(IDEST)

KNT1=GCOUNT(IDEST)

CALLMPI_SEND(B(IST1),KNT1,MPI_REAL8,IDEST,10,COMM,IERR)

CALLMPI_SEND(C(IST1),KNT1,MPI_REAL8,IDEST,20,COMM,IERR)

CALLMPI_SEND(D(IST1),KNT1,MPI_REAL8,IDEST,30,COMM,IERR)

ENDDO

ELSE

KNT=GCOUNT(MYID)

CALLMPI_RECV(B(ISTART),KNT,MPI_REAL8,0,10,COMM,ISTAT,IERR)

CALLMPI_RECV(C(ISTART),KNT,MPI_REAL8,0,20,COMM,ISTAT,IERR)

CALLMPI_RECV(D(ISTART),KNT,MPI_REAL8,0,30,COMM,ISTAT,IERR)

ENDIF

T3cp⑷

CALLMPI_SENDRECV(B(IEND),1,MPI_REAL8,R_NBR,110,

1BQSTARTM1),1,MPI_REAL8,L_NBR,110,

2COMM,ISTAT,IERR)

CALLMPI_SENDRECV(B(ISTART),1,MPI_REAL8,L_NBR,120,

1B(IENDP1),1,MPI_REAL8,R_NBR,120,

2COMM,ISTAT,IERR)

AMAX-1,0D12

CDOl=2,NTOTAL-1

DOl=ISTART2,IEND1

A(l)=C(l)*D(l)+(B(l-1)+2.0*B(l)+B(l+1))*0.25

AMAX=MAX(AMAX,A(I))

ENDDO

T3CP(5)

ITAG=110

IF(MYID.NE.O)THEN

KNT=GCOUNT(MYID)

CALLMPI_SEND(A(ISTART),KNT,MPI_REAL8,0,ITAG,COMM,IERR)

ELSE

DOISRC=1,LASTP

IST1=GSTART(ISRC)

KNT1=GCOUNT(ISRC)

CALLMPI_RECV(A(IST1),KNT1,MPI_REAL8,ISRC,ITAG,

COMM,ISTAT,IERR)

ENDDO

ENDIF

T3CP(6)

CALLMPI_REDUCE(AMAX,GMAX,1,MPI_REAL8,MPI_MAX,

1COMM,IERR)

IF(MYID.EQ.O)THEN

WRITE(*,101)(A(I),I=1,NTOTALJ5)

WRITE(*,102)GMAX

ENDIF

101FORMAT(10F8.3)

102FORMAT('MAXIMUMVALUEofARRAYAis',E15.5)

CALLMPI_FINALIZE(IERR)

STOP

END

T3CP

ATTENTION:0031-4084nodesallocatedbyLoadLeveler,continuing...

NPROC,MYIDJISTART,IEND=43151200

NPROC,MYID,ISTART,IEND=42101150

NPROC,MYIDJISTART,IEND=4151100

NPROC5MYID,ISTARTJIEND=40150

.0003.0632.5632.3832.2902.2342.1962.1682.1482.131

2.1182.1082.0992.0912.0852.0792.0742.0702.0662.063

2.0602.0572.0542.0522.0502.0482.0462.0442.0432.041

2.0402.0392.0372.0362.0352.0342.0332.0322.0312.031

MAXIMUMVALUEOFARRAYAis.57500E+01

CPUO

MPI_BCAST

CPU

IF(MYID.EQ.O)THEN

,,,,,,

OPEN(7)FILE=input.dat,STATUS=OLD,FORM=UNFORMATTED)

READ(7)B

READ(7)C

READ(7)D

ENDIF

IROOT=0

ALLMPI_BCAST(B,NTOTAL,MPI_REAL8,IROOT,

1MPI_COMM_WORLD,IERR)

CALLMPI_BCAST(C,NTOTAL；MPLREAL8,IROOT,

1-MPLCOMM_WORLD,IERR)

CALLMPI_BCAST(D,NTOTAL,MPI_REAL8,IROOT,

1MPI_COMM_WdRLD,IERR)

■MPLSENDRECVMPI_BCAST

T3SEQ

T3CP

T3DCPJ

T3DCP_2

IP什

Ampi_proc_null

ianch

iend

Iinnd+1

cpuO?

A

lstart-1|istart2iendl

iend

istart

|iend+1

cpu1a

Iistartiend

iencllI

lstart-1istart2iend+1

cpu2

Iistart

lstart-1istart2mpij)roc_null

rightisowneddataisexchangeddata

T3DCP_1(1)

cPROGRAMT3DCP_1

c

cExchange1elementbothonrightandlefthandside

PARAMETER(NTOTAL=200,NP=4,N=NTOTAL/NP)

INCLUDE'mpif.h'

REAL*8A(0:N+1),B(0:N+1),C(0:N+1),D(0:N+1),T(NTOTAL),

1AMAX,GMAX

INTEGERNPROC,MYID,ISTAT(MPI_STATUS_SIZE),ISTART,IEND,

1L_NBR,R_NBR

CALLMPIJNIT(IERR)

CALLMPI_COMM_SIZE(MPI_COMM_WORLD5NPROC,IERR)

CALLMPI_COMM_RANK(MPI_COMM_WORLD,MYID,IERR)

T3DCP1(2)

ISTART=1

IEND=N

ISTARTM1=ISTART-1

IENDP1=IEND+1

ISTART2=1

IF(MYID.EQ.O)ISTART2=2

IEND1=N

IF(MYID.EQ.NPROC-I)IEND1=N-1

L_NBR=MYID-1

R_NBR=MYID+1

IF(MYID.EQ.O)L_NBR=MPI_PROC_NULL

IF(MYID.EQ.NPROC-1)R_NBR=MPLPROC_NULL

C