ПРИМЕРЫ

Примеры компиляции и запуска программ с использованием популярных библиотек

BLAS

$ # Example - 1. Using BLAS Level 1 Function "sdot"

$ # GNU GCC Fortran and Netlib BLAS

$ BLAS_TEST_PATH=/usr/local/hpc/testfiles/mathlib/blas/

$ BLAS_LIB_PATH=/usr/local/mathlib/blas/3.8.0/lib/

$ cat $BLAS_TEST_PATH/test_blas_1.f90

program dot_main

real x(10), y(10), sdot, res

integer n, incx, incy, i

external sdot

n = 5

incx = 2

incy = 1

do i = 1, 10

x(i) = 2.0e0

y(i) = 1.0e0

end do

res = sdot (n, x, incx, y, incy)

print*, 'SDOT = ', res

end

$ gfortran $BLAS_TEST_PATH/test_blas_1.f90 -L$BLAS_LIB_PATH -lblas

$ ./a.out

SDOT = 10.000000

$ # Intel Fortran and Intel BLAS (MKL)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ ifort $BLAS_TEST_PATH/test_blas_1.f90 -mkl

$ ./a.out

SDOT = 10.00000

$ # ------------------------------------------------------------------------------------------------------------------

$ # Example - 2. Using BLAS Level 1 Routine "scopy"

$ # GNU GCC Fortran and Netlib BLAS

$ cat $BLAS_TEST_PATH/test_blas_2.f90

program copy_main

real x(10), y(10)

integer n, incx, incy, i

n = 3

incx = 3

incy = 1

do i = 1, 10

x(i) = i

end do

call scopy (n, x, incx, y, incy)

print*, 'Y = ', (y(i), i = 1, n)

end

$ gfortran $BLAS_TEST_PATH/test_blas_2.f90 -L$BLAS_LIB_PATH -lblas

$ ./a.out

Y = 1.0000000 4.0000000 7.0000000

$ # ------------------------------------------------------------------------------------------------------------------

$ # Example - 3. Using BLAS Level 2 Routine "sger"

$ # GNU GCC Fortran and Netlib BLAS

$ cat $BLAS_TEST_PATH/test_blas_3.f90

program ger_main

real a(5,3), x(10), y(10), alpha

integer m, n, incx, incy, i, j, lda

m = 2

n = 3

lda = 5

incx = 2

incy = 1

alpha = 0.5

do i = 1, 10

x(i) = 1.0

y(i) = 1.0

end do

do i = 1, m

do j = 1, n

a(i,j) = j

end do

call sger (m, n, alpha, x, incx, y, incy, a, lda)

print*, 'Matrix A: '

do i = 1, m

print*, (a(i,j), j = 1, n)

end do

end

$ gfortran $BLAS_TEST_PATH/test_blas_3.f90 -L$BLAS_LIB_PATH -lblas

$ ./a.out

Matrix A:

1.5000000 2.5000000 3.5000000

$ # ------------------------------------------------------------------------------------------------------------------

$ # Example - 4. Using BLAS Level 3 Routine "ssymm"

$ # GNU GCC Fortran and Netlib BLAS

$ cat $BLAS_TEST_PATH/test_blas_4.f90

program symm_main

real a(3,3), b(3,2), c(3,3), alpha, beta

integer m, n, lda, ldb, ldc, i, j

character uplo, side

uplo = 'u'

side = 'l'

m = 3

n = 2

lda = 3

ldb = 3

ldc = 3

alpha = 0.5

beta = 2.0

do i = 1, m

do j = 1, m

a(i,j) = 1.0

end do

do i = 1, m

do j = 1, n

c(i,j) = 1.0

b(i,j) = 2.0

end do

call ssymm (side, uplo, m, n, alpha, a, lda, b, ldb, beta, c, ldc)

print*, 'Matrix C: '

do i = 1, m

print*, (c(i,j), j = 1, n)

end do

end

$ gfortran $BLAS_TEST_PATH/test_blas_4.f90 -L$BLAS_LIB_PATH -lblas

$ ./a.out

Matrix C:

5.0000000 5.0000000

CBLAS

$ # Example - 1. Using BLAS Level 2 Function "dgemv"

$ # GNU GCC C and Netlib BLAS

$ CBLAS_LIB_PATH=/usr/local/mathlib/cblas/lib/

$ CBLAS_INC_PATH=/usr/local/mathlib/cblas/include/

$ CBLAS_TEST_PATH=/usr/local/hpc/testfiles/mathlib/cblas/

$ BLAS_LIB_PATH=/usr/local/mathlib/blas/3.8.0/lib/

$ cat $CBLAS_TEST_PATH/test_cblas.c

#include <stdio.h>

#include <cblas.h>

double m[] = {

3, 1, 3,

1, 5, 9,

2, 6, 5

};

double x[] = {

-1, -1, 1

};

double y[] = {

0, 0, 0

};

int main()

{

int i, j;

for (i=0; i<3; ++i) {

for (j=0; j<3; ++j) printf("%5.1f", m[i*3+j]);

putchar('\n');

}

cblas_dgemv(CblasRowMajor, CblasNoTrans, 3, 3, 1.0, m, 3, x, 1, 0.0, y, 1);

for (i=0; i<3; ++i) printf("%5.1f\n", y[i]);

return 0;

}

$ gcc -c $CBLAS_TEST_PATH/test_cblas.c -I$CBLAS_INC_PATH

$ gfortran test_cblas.o -L$CBLAS_LIB_PATH -lcblas -L$BLAS_LIB_PATH -lblas

$ ./a.out

3.0 1.0 3.0

1.0 5.0 9.0

2.0 6.0 5.0

-1.0

3.0

-3.0

$ # Intel C and Intel CBLAS (MKL)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ cat $CBLAS_TEST_PATH/test_cblas_mkl.c

#include <stdio.h>

#include <mkl_cblas.h>

double m[] = {

3, 1, 3,

1, 5, 9,

2, 6, 5

};

double x[] = {

-1, -1, 1

};

double y[] = {

0, 0, 0

};

int main()

{

int i, j;

for (i=0; i<3; ++i) {

for (j=0; j<3; ++j) printf("%5.1f", m[i*3+j]);

putchar('\n');

}

cblas_dgemv(CblasRowMajor, CblasNoTrans, 3, 3, 1.0, m, 3, x, 1, 0.0, y, 1);

for (i=0; i<3; ++i) printf("%5.1f\n", y[i]);

return 0;

}

$ icc $CBLAS_TEST_PATH/test_cblas_mkl.c -mkl

$ ./a.out

3.0 1.0 3.0ls

1.0 5.0 9.0

2.0 6.0 5.0

-1.0

3.0

-3.0

LAPACK

$ Example - 1. Using LAPACK Function "DGESV"

$ GNU GCC Fortran and Netlib LAPACK & BLAS

$ BLAS_LIB_PATH=/usr/local/mathlib/blas/3.8.0/lib/

$ LAPACK_LIB_PATH=/usr/local/mathlib/lapack/3.8.0/lib/

$ LAPACK_TEST_PATH=/usr/local/hpc/testfiles/mathlib/lapack/

$ cat $LAPACK_TEST_PATH/test_lapack.f90

program LinearEquations

! solving the matrix equation A*x=b using LAPACK

Implicit none

! declarations

double precision :: A(3,3), b(3)

integer :: i, pivot(3), ok

! matrix A

A(1,:)=(/3, 1, 3/)

A(2,:)=(/1, 5, 9/)

A(3,:)=(/2, 6, 5/)

! vector b

b(:)=(/-1, 3, -3/)

!b(:)=(/2, 2, 9/)

! find the solution using the LAPACK routine DGESV

call DGESV(3, 1, A, 3, pivot, b, 3, ok)

! print the solution x

do i=1, 3

write(*,9) i, b(i)

end do

9 format('x[', i1, ']= ', f5.2)

end program LinearEquations

$ gfortran $LAPACK_TEST_PATH/test_lapack.f90 -L$LAPACK_LIB_PATH -llapack -L$BLAS_LIB_PATH -lblas

$ ./a.out

x[1]= -1.00

x[2]= -1.00

x[3]= 1.00

$ # Intel Fortran and Intel LAPACK (MKL)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ ifort $LAPACK_TEST_PATH/test_lapack.f90 -mkl

$ ./a.out

x[1]= -1.00

x[2]= -1.00

x[3]= 1.00

LAPACKE

$ # Example - 1. Using LAPACK Function "dgels"

$ # GNU GCC C and Netlib LAPACKE

$ BLAS_LIB_PATH=/usr/local/mathlib/blas/3.8.0/lib/

$ LAPACK_LIB_PATH=/usr/local/mathlib/lapack/3.8.0/lib/

$ LAPACKE_LIB_PATH=/usr/local/mathlib/lapacke/3.8.0/lib/

$ LAPACKE_INC_PATH=/usr/local/mathlib/lapacke/3.8.0/include/

$ LAPACKE_TEST_PATH=/usr/local/hpc/testfiles/mathlib/lapacke/

$ cat $LAPACKE_TEST_PATH/test_lapacke.c

/* Calling DGELS using row-major order */

#include <stdio.h>

#include <lapacke.h>

int main (int argc, const char * argv[])

{

double a[5][3] = {1,1,1,2,3,4,3,5,2,4,2,5,5,4,3};

double b[5][2] = {-10,-3,12,14,14,12,16,16,18,16};

lapack_int info,m,n,lda,ldb,nrhs;

int i,j;

m = 5;

n = 3;

nrhs = 2;

lda = 3;

ldb = 2;

info = LAPACKE_dgels(LAPACK_ROW_MAJOR,'N',m,n,nrhs,*a,lda,*b,ldb);

for(i=0;i<n;i++)

{

for(j=0;j<nrhs;j++)

{

printf("%lf ",b[i][j]);

}

printf("\n");

}

return(info);

}

$ gcc -c $LAPACKE_TEST_PATH/test_lapacke.c -I$LAPACKE_INC_PATH

$ gfortran test_lapacke.o -L$LAPACKE_LIB_PATH -llapacke -L$LAPACK_LIB_PATH -llapack -L$BLAS_LIB_PATH -lblas

$ ./a.out

2.000000 1.000000

1.000000 1.000000

1.000000 2.000000

$ # Intel C and Intel LAPACKE (MKL)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ icc $LAPACKE_TEST_PATH/test_lapacke.c -I$LAPACKE_INC_PATH -mkl

$ ./a.out

2.000000 1.000000

1.000000 1.000000

1.000000 2.000000

CLAPACK

$ # Example - 1. Using CLAPACK Function "dgesvd"

$ # GNU GCC C and Netlib CLAPACK and CBLAS and F2C

$ CLAPACK_LIB_PATH=/usr/local/mathlib/clapack/3.2.1/lib/

$ CLAPACK_INC_PATH=/usr/local/mathlib/clapack/3.2.1/include/

$ CLAPACK_TEST_PATH=/usr/local/hpc/testfiles/mathlib/clapack/

$ cat $CLAPACK_TEST_PATH/test_clapack.c

#include "f2c.h"

#include "stdio.h"

#include "clapack.h"

#define SIZE 4

void MAIN_(){}

void MAIN__(){}

void _MAIN_(){}

main( )

{

char JOBU;

char JOBVT;

int i;

integer M = SIZE;

integer N = SIZE;

integer LDA = M;

integer LDU = M;

integer LDVT = N;

integer LWORK;

integer INFO;

integer mn = min( M, N );

integer MN = max( M, N );

double a[SIZE*SIZE] = { 16.0, 5.0, 9.0 , 4.0, 2.0, 11.0, 7.0 , 14.0, 3.0, 10.0, 6.0, 15.0, 13.0, 8.0, 12.0, 1.0};

double s[SIZE];

double wk[201];

double uu[SIZE*SIZE];

double vt[SIZE*SIZE];

JOBU = 'A';

JOBVT = 'A';

LWORK = 201;

/* Subroutine int dgesvd_(char *jobu, char *jobvt, integer *m, integer *n,

* doublereal *a, integer *lda, doublereal *s, doublereal *u, integer *

* ldu, doublereal *vt, integer *ldvt, doublereal *work, integer *lwork,

* integer *info)

* */

dgesvd_( &JOBU, &JOBVT, &M, &N, a, &LDA, s, uu,

&LDU, vt, &LDVT, wk, &LWORK, &INFO);

printf("\n INFO=%d", INFO );

for ( i= 0; i< SIZE; i++ ) {

printf("\n s[ %d ] = %f", i, s[ i ] );

}

return 0;

}

$ gcc $CLAPACK_TEST_PATH/test_clapack.c -I$CLAPACK_INC_PATH -L$CLAPACK_LIB_PATH -lclapack -lcblas -lf2c -lm

$ ./a.out

INFO=0

s[ 0 ] = 34.000000

s[ 1 ] = 17.888544

s[ 2 ] = 4.472136

s[ 3 ] = 0.000000

$ # Intel C and Intel CLAPACK (MKL)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ icc $CLAPACK_TEST_PATH/test_clapack.c -I$CLAPACK_INC_PATH -mkl

$ ./a.out

INFO=0

s[ 0 ] = 34.000000

s[ 1 ] = 17.888544

s[ 2 ] = 4.472136

s[ 3 ] = 0.000000

ScaLAPACK

$ # Example - 1. Using ScaLAPACK "PDGESV"

$ # GNU Fortran and Intel MPI and Netlib ScaLAPACK

$ BLAS_LIB_PATH=/usr/local/mathlib/blas/3.8.0/lib/

$ LAPACK_LIB_PATH=/usr/local/mathlib/lapack/3.8.0/lib/

$ SCALAPACK_LIB_PATH=/usr/local/mathlib/scalapack/2.0.2/lib/

$ SCALAPACK_TEST_PATH=/usr/local/hpc/testfiles/mathlib/scalapack/

$ cat $SCALAPACK_TEST_PATH/test_scalapack.f

PROGRAM EXAMPLE1

* Example Program solving Ax=b via ScaLAPACK routine PDGESV

* .. Parameters ..

INTEGER DLEN_, IA, JA, IB, JB, M, N, MB, NB, RSRC,

$ CSRC, MXLLDA, MXLLDB, NRHS, NBRHS, NOUT,

$ MXLOCR, MXLOCC, MXRHSC

PARAMETER ( DLEN_ = 9, IA = 1, JA = 1, IB = 1, JB = 1,

$ M = 9, N = 9, MB = 2, NB = 2, RSRC = 0,

$ CSRC = 0, MXLLDA = 5, MXLLDB = 5, NRHS = 1,

$ NBRHS = 1, NOUT = 6, MXLOCR = 5, MXLOCC = 4,

$ MXRHSC = 1 )

DOUBLE PRECISION ONE

PARAMETER ( ONE = 1.0D+0 )

* ..

* .. Local Scalars ..

INTEGER ICTXT, INFO, MYCOL, MYROW, NPCOL, NPROW

DOUBLE PRECISION ANORM, BNORM, EPS, RESID, XNORM

* ..

* .. Local Arrays ..

INTEGER DESCA( DLEN_ ), DESCB( DLEN_ ),

$ IPIV( MXLOCR+NB )

DOUBLE PRECISION A( MXLLDA, MXLOCC ), A0( MXLLDA, MXLOCC ),

$ B( MXLLDB, MXRHSC ), B0( MXLLDB, MXRHSC ),

$ WORK( MXLOCR )

* ..

* .. External Functions ..

DOUBLE PRECISION PDLAMCH, PDLANGE

EXTERNAL PDLAMCH, PDLANGE

* ..

* .. External Subroutines ..

EXTERNAL BLACS_EXIT, BLACS_GRIDEXIT, BLACS_GRIDINFO,

$ DESCINIT, MATINIT, PDGEMM, PDGESV, PDLACPY,

$ SL_INIT

* ..

* .. Intrinsic Functions ..

INTRINSIC DBLE

* ..

* .. Data statements ..

DATA NPROW / 2 / , NPCOL / 3 /

* ..

* .. Executable Statements ..

* INITIALIZE THE PROCESS GRID

CALL SL_INIT( ICTXT, NPROW, NPCOL )

CALL BLACS_GRIDINFO( ICTXT, NPROW, NPCOL, MYROW, MYCOL )

* If I'm not in the process grid, go to the end of the program

IF( MYROW.EQ.-1 )

$ GO TO 10

* DISTRIBUTE THE MATRIX ON THE PROCESS GRID

* Initialize the array descriptors for the matrices A and B

CALL DESCINIT( DESCA, M, N, MB, NB, RSRC, CSRC, ICTXT, MXLLDA,

$ INFO )

CALL DESCINIT( DESCB, N, NRHS, NB, NBRHS, RSRC, CSRC, ICTXT,

$ MXLLDB, INFO )

* Generate matrices A and B and distribute to the process grid

CALL MATINIT( A, DESCA, B, DESCB )

* Make a copy of A and B for checking purposes

CALL PDLACPY( 'All', N, N, A, 1, 1, DESCA, A0, 1, 1, DESCA )

CALL PDLACPY( 'All', N, NRHS, B, 1, 1, DESCB, B0, 1, 1, DESCB )

* CALL THE SCALAPACK ROUTINE

* Solve the linear system A * X = B

CALL PDGESV( N, NRHS, A, IA, JA, DESCA, IPIV, B, IB, JB, DESCB,

$ INFO )

IF( MYROW.EQ.0 .AND. MYCOL.EQ.0 ) THEN

WRITE( NOUT, FMT = 9999 )

WRITE( NOUT, FMT = 9998 )M, N, NB

WRITE( NOUT, FMT = 9997 )NPROW*NPCOL, NPROW, NPCOL

WRITE( NOUT, FMT = 9996 )INFO

END IF

* Compute residual ||A * X - B|| / ( ||X|| * ||A|| * eps * N )

EPS = PDLAMCH( ICTXT, 'Epsilon' )

ANORM = PDLANGE( 'I', N, N, A, 1, 1, DESCA, WORK )

BNORM = PDLANGE( 'I', N, NRHS, B, 1, 1, DESCB, WORK )

CALL PDGEMM( 'N', 'N', N, NRHS, N, ONE, A0, 1, 1, DESCA, B, 1, 1,

$ DESCB, -ONE, B0, 1, 1, DESCB )

XNORM = PDLANGE( 'I', N, NRHS, B0, 1, 1, DESCB, WORK )

RESID = XNORM / ( ANORM*BNORM*EPS*DBLE( N ) )

IF( MYROW.EQ.0 .AND. MYCOL.EQ.0 ) THEN

IF( RESID.LT.10.0D+0 ) THEN

WRITE( NOUT, FMT = 9995 )

WRITE( NOUT, FMT = 9993 )RESID

ELSE

WRITE( NOUT, FMT = 9994 )

WRITE( NOUT, FMT = 9993 )RESID

END IF

* RELEASE THE PROCESS GRID

* Free the BLACS context

CALL BLACS_GRIDEXIT( ICTXT )

10 CONTINUE

* Exit the BLACS

CALL BLACS_EXIT( 0 )

9999 FORMAT( / 'ScaLAPACK Example Program #1 -- May 1, 1997' )

9998 FORMAT( / 'Solving Ax=b where A is a ', I3, ' by ', I3,

$ ' matrix with a block size of ', I3 )

9997 FORMAT( 'Running on ', I3, ' processes, where the process grid',

$ ' is ', I3, ' by ', I3 )

9996 FORMAT( / 'INFO code returned by PDGESV = ', I3 )

9995 FORMAT( /

$ 'According to the normalized residual the solution is correct.'

$ )

9994 FORMAT( /

$ 'According to the normalized residual the solution is incorrect.'

$ )

9993 FORMAT( / '||A*x - b|| / ( ||x||*||A||*eps*N ) = ', 1P, E16.8 )

STOP

END

SUBROUTINE MATINIT( AA, DESCA, B, DESCB )

* MATINIT generates and distributes matrices A and B (depicted in

* Figures 2.5 and 2.6) to a 2 x 3 process grid

* .. Array Arguments ..

INTEGER DESCA( * ), DESCB( * )

DOUBLE PRECISION AA( * ), B( * )

* ..

* .. Parameters ..

INTEGER CTXT_, LLD_

PARAMETER ( CTXT_ = 2, LLD_ = 9 )

* ..

* .. Local Scalars ..

INTEGER ICTXT, MXLLDA, MYCOL, MYROW, NPCOL, NPROW

DOUBLE PRECISION A, C, K, L, P, S

* ..

* .. External Subroutines ..

EXTERNAL BLACS_GRIDINFO

* ..

* .. Executable Statements ..

ICTXT = DESCA( CTXT_ )

CALL BLACS_GRIDINFO( ICTXT, NPROW, NPCOL, MYROW, MYCOL )

S = 19.0D0

C = 3.0D0

A = 1.0D0

L = 12.0D0

P = 16.0D0

K = 11.0D0

MXLLDA = DESCA( LLD_ )

IF( MYROW.EQ.0 .AND. MYCOL.EQ.0 ) THEN

AA( 1 ) = S

AA( 2 ) = -S

AA( 3 ) = -S

AA( 4 ) = -S

AA( 5 ) = -S

AA( 1+MXLLDA ) = C

AA( 2+MXLLDA ) = C

AA( 3+MXLLDA ) = -C

AA( 4+MXLLDA ) = -C

AA( 5+MXLLDA ) = -C

AA( 1+2*MXLLDA ) = A

AA( 2+2*MXLLDA ) = A

AA( 3+2*MXLLDA ) = A

AA( 4+2*MXLLDA ) = A

AA( 5+2*MXLLDA ) = -A

AA( 1+3*MXLLDA ) = C

AA( 2+3*MXLLDA ) = C

AA( 3+3*MXLLDA ) = C

AA( 4+3*MXLLDA ) = C

AA( 5+3*MXLLDA ) = -C

B( 1 ) = 0.0D0

B( 2 ) = 0.0D0

B( 3 ) = 0.0D0

B( 4 ) = 0.0D0

B( 5 ) = 0.0D0

ELSE IF( MYROW.EQ.0 .AND. MYCOL.EQ.1 ) THEN

AA( 1 ) = A

AA( 2 ) = A

AA( 3 ) = -A

AA( 4 ) = -A

AA( 5 ) = -A

AA( 1+MXLLDA ) = L

AA( 2+MXLLDA ) = L

AA( 3+MXLLDA ) = -L

AA( 4+MXLLDA ) = -L

AA( 5+MXLLDA ) = -L

AA( 1+2*MXLLDA ) = K

AA( 2+2*MXLLDA ) = K

AA( 3+2*MXLLDA ) = K

AA( 4+2*MXLLDA ) = K

AA( 5+2*MXLLDA ) = K

ELSE IF( MYROW.EQ.0 .AND. MYCOL.EQ.2 ) THEN

AA( 1 ) = A

AA( 2 ) = A

AA( 3 ) = A

AA( 4 ) = -A

AA( 5 ) = -A

AA( 1+MXLLDA ) = P

AA( 2+MXLLDA ) = P

AA( 3+MXLLDA ) = P

AA( 4+MXLLDA ) = P

AA( 5+MXLLDA ) = -P

ELSE IF( MYROW.EQ.1 .AND. MYCOL.EQ.0 ) THEN

AA( 1 ) = -S

AA( 2 ) = -S

AA( 3 ) = -S

AA( 4 ) = -S

AA( 1+MXLLDA ) = -C

AA( 2+MXLLDA ) = -C

AA( 3+MXLLDA ) = -C

AA( 4+MXLLDA ) = C

AA( 1+2*MXLLDA ) = A

AA( 2+2*MXLLDA ) = A

AA( 3+2*MXLLDA ) = A

AA( 4+2*MXLLDA ) = -A

AA( 1+3*MXLLDA ) = C

AA( 2+3*MXLLDA ) = C

AA( 3+3*MXLLDA ) = C

AA( 4+3*MXLLDA ) = C

B( 1 ) = 1.0D0

B( 2 ) = 0.0D0

B( 3 ) = 0.0D0

B( 4 ) = 0.0D0

ELSE IF( MYROW.EQ.1 .AND. MYCOL.EQ.1 ) THEN

AA( 1 ) = A

AA( 2 ) = -A

AA( 3 ) = -A

AA( 4 ) = -A

AA( 1+MXLLDA ) = L

AA( 2+MXLLDA ) = L

AA( 3+MXLLDA ) = -L

AA( 4+MXLLDA ) = -L

AA( 1+2*MXLLDA ) = K

AA( 2+2*MXLLDA ) = K

AA( 3+2*MXLLDA ) = K

AA( 4+2*MXLLDA ) = K

ELSE IF( MYROW.EQ.1 .AND. MYCOL.EQ.2 ) THEN

AA( 1 ) = A

AA( 2 ) = A

AA( 3 ) = -A

AA( 4 ) = -A

AA( 1+MXLLDA ) = P

AA( 2+MXLLDA ) = P

AA( 3+MXLLDA ) = -P

AA( 4+MXLLDA ) = -P

END IF

RETURN

END

SUBROUTINE SL_INIT( ICTXT, NPROW, NPCOL )

* .. Scalar Arguments ..

INTEGER ICTXT, NPCOL, NPROW

* ..

* Purpose

* =======

* SL_INIT initializes an NPROW x NPCOL process grid using a row-major

* ordering of the processes. This routine retrieves a default system

* context which will include all available processes. In addition it

* spawns the processes if needed.

* Arguments

* =========

* ICTXT (global output) INTEGER

* ICTXT specifies the BLACS context handle identifying the

* created process grid. The context itself is global.

* NPROW (global input) INTEGER

* NPROW specifies the number of process rows in the grid

* to be created.

* NPCOL (global input) INTEGER

* NPCOL specifies the number of process columns in the grid

* to be created.

* =====================================================================

* .. Local Scalars ..

INTEGER IAM, NPROCS

* ..

* .. External Subroutines ..

EXTERNAL BLACS_GET, BLACS_GRIDINIT, BLACS_PINFO,

$ BLACS_SETUP

* ..

* .. Executable Statements ..

* Get starting information

CALL BLACS_PINFO( IAM, NPROCS )

* If machine needs additional set up, do it now

IF( NPROCS.LT.1 ) THEN

IF( IAM.EQ.0 )

$ NPROCS = NPROW*NPCOL

CALL BLACS_SETUP( IAM, NPROCS )

END IF

* Define process grid

CALL BLACS_GET( -1, 0, ICTXT )

CALL BLACS_GRIDINIT( ICTXT, 'Row-major', NPROW, NPCOL )

RETURN

* End of SL_INIT

END

[amalevich@testvm-0 scalapack]$

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/impi/2018.3.222/bin64/mpivars.sh

$ mpif77 $SCALAPACK_TEST_PATH/test_scalapack.f -L$SCALAPACK_LIB_PATH -lscalapack -L$LAPACK_LIB_PATH -llapack -L$BLAS_LIB_PATH -lblas

$ mpirun -np 6 ./a.out

ScaLAPACK Example Program #1 -- May 1, 1997

Solving Ax=b where A is a 9 by 9 matrix with a block size of 2

Running on 6 processes, where the process grid is 2 by 3

INFO code returned by PDGESV = 0

According to the normalized residual the solution is correct.

||A*x - b|| / ( ||x||*||A||*eps*N ) = 0.00000000E+00

MKL (Intel BLAS, Intel LAPACK, Intel ScaLAPACK)

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64 # компиляторы: icc, ifort

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64 # Intel MKL

$ source /usr/local/ipsxece/ipsxece_2018_u3/impi/2018.3.222/bin64/mpivars.sh # Intel MPI

Примеры:

1) Автоматическая линковка с Intel MKL (OpenMP):

$ ifort myprog.f -mkl

2) Статическая линковка c Intel MKL (OpenMP):

$ MKLPATH=$MKLROOT/lib/ia64_lin

$ MKLINCLUDE=$MKLROOT/include

$ ifort myprog.f -L$MKLPATH -I$MKLINCLUDE \

-Wl,--start-group \

$MKLPATH/libmkl_intel_lp64.a \

$MKLPATH/libmkl_intel_thread.a \

$MKLPATH/libmkl_core.a \

-Wl,--end-group \

-liomp5 \

-lpthread \

-lm

3) Динамическая линковка c Intel MKL (OpenMP):

$ MKLPATH=$MKLROOT/lib/ia64_lin

$ MKLINCLUDE=$MKLROOT/include

$ ifort myprog.f -L$MKLPATH -I$MKLINCLUDE \

-lmkl_intel_lp64 \

-lmkl_intel_thread \

-lmkl_core \

-liomp5 \

-lpthread \

-lm

MPI

Пример 1 - сборка и запуск программы вычисления числа П на языке Си с использованием MPI.

$ # Инициализируем необходимые переменные окружения для Intel Parallel Studio 2018

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/mkl/bin/mklvars.sh intel64

$ source /usr/local/ipsxece/ipsxece_2018_u3/impi/2018.3.222/bin64/mpivars.sh

$ MPI_TEST_PATH=/usr/local/hpc/testfiles/mpi/

$ # Компилируем MPI-программу

$ mpiicc $MPI_TEST_PATH/pi_mpi.c

$ # Запускаем MPI-программу на 6 процессорах

$ mpirun -np 6 ./a.out 1000

pi is 3.1415927369231267, Error is 0.0000000833333336

wall clock time = 0.000465

OpenMP

Пример 1 - компиляция программы pi_omp.c при помощи Intel компилятора

$ source /usr/local/ipsxece/ipsxece_2018_u3/bin/compilervars.sh intel64

$ OMP_TEST_PATH=/usr/local/hpc/testfiles/openmp/

$ icc $OMP_TEST_PATH/pi_omp.c -qopenmp

$ ./a.out 1000

pi=3.141592736923126

eps=8.333333e-08

num threads = 1

$ export OMP_NUM_THREADS=10

$ ./a.out 1000

pi=3.141592736923126

eps=8.333333e-08

num threads = 10

$ export OMP_NUM_THREADS=14

$ ./a.out 1000

pi=3.141592736923126

eps=8.333333e-08

num threads = 14

$ unset OMP_NUM_THREADS

$ ./a.out 1000

pi=3.141592736923126

eps=8.333333e-08

num threads = 1

CUDA+NVCC

Пример 1 - сборка CUDA-программы при помощи компилятора nvcc и запуск на Nvidia Tesla k40 (кластер Huawei)

$ CUDA_PATH=/usr/local/cuda

$ CUDA_TEST_PATH=$CUDA_PATH/samples/1_Utilities/bandwidthTest/

$ export PATH=/usr/local/cuda/bin/:$PATH

$ CUDA_INC_PATH=/usr/local/cuda/samples/common/inc

$ nvcc $CUDA_TEST_PATH/bandwidthTest.cu -I$CUDA_INC_PATH

$ ./a.out

[CUDA Bandwidth Test] - Starting...

Running on...

Device 0: Tesla K40m

Quick Mode

Host to Device Bandwidth, 1 Device(s)

PINNED Memory Transfers

Transfer Size (Bytes) Bandwidth(MB/s)

33554432 10152.7

Device to Host Bandwidth, 1 Device(s)

PINNED Memory Transfers

Transfer Size (Bytes) Bandwidth(MB/s)

33554432 10269.2

Device to Device Bandwidth, 1 Device(s)

PINNED Memory Transfers

Transfer Size (Bytes) Bandwidth(MB/s)

33554432 183668.1

Result = PASS

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

OpenACC+PGI

Python

Форма поиска

ПРИМЕРЫ