root / synthbench / euroben-dm / mod2a / libver.f @ 0:839f52ef7657
History | View | Annotate | Download (3.4 kB)
| 1 |
Subroutine libver( trans, m, n, a, lda, x, y, nrep, extime, |
|---|---|
| 2 |
& perfor ) |
| 3 |
! ---------------------------------------------------------------------- |
| 4 |
! --- This routine performs the timing and calculates the performance |
| 5 |
! of the Special library version of Matrix-Vector multiplication |
| 6 |
! routine DGEMV. |
| 7 |
*** If DGEMV is not available, replace it with the appropriate *** |
| 8 |
*** equivalent. *** |
| 9 |
! |
| 10 |
! Parameters |
| 11 |
! ========== |
| 12 |
! |
| 13 |
! TRANS - Input - Determines the operation te be performed by MVSR |
| 14 |
! M, N - Input - Dimension of the problem |
| 15 |
! A(M,N) - Input - Contains the matrix A |
| 16 |
! LDA - Input - Leading Dimension of A |
| 17 |
! X(*) - Input - Contains the vector x |
| 18 |
! Y(*) - Output - Contains result vector y |
| 19 |
! NREP - Input - The initial repetition factor to increase |
| 20 |
! accuracy of the timing |
| 21 |
! EXTIME - Output - Contains the execution time for the operation |
| 22 |
! PERFOR - Output - Contains the performance in Mflop/s |
| 23 |
! ---------------------------------------------------------------------- |
| 24 |
Use numerics |
| 25 |
Include 'mpif.h' |
| 26 |
|
| 27 |
Character*1 :: trans |
| 28 |
Integer :: m, n, lda, nrep |
| 29 |
Real(l_) :: a(lda,*), x(*), y(*) |
| 30 |
Real(l_) :: extime, perfor |
| 31 |
! ---------------------------------------------------------------------- |
| 32 |
! --- Local Constants |
| 33 |
Real(l_), Parameter :: zero = 0.0_l_, p9 = 0.9_l_, one = 0.0_l_ |
| 34 |
! |
| 35 |
! --- Local variables |
| 36 |
Integer :: nop, k, krep, limrep |
| 37 |
Real(l_) :: t1, t2 |
| 38 |
Real(l_) :: alpha1, alphak, beta1, betak |
| 39 |
! ---------------------------------------------------------------------- |
| 40 |
! --- Operation count. |
| 41 |
|
| 42 |
nop = 2*m*n + m + n |
| 43 |
|
| 44 |
! --- Set the ALPHA and BETA parameters |
| 45 |
! |
| 46 |
alpha1 = one |
| 47 |
beta1 = zero |
| 48 |
alphak = p9 |
| 49 |
betak = p9 |
| 50 |
|
| 51 |
! --- Set the initial and maximum repetition factor. |
| 52 |
! |
| 53 |
krep = nrep |
| 54 |
limrep = 100 * nrep |
| 55 |
! |
| 56 |
10 Continue |
| 57 |
! |
| 58 |
If( trans == 'N' ) Then |
| 59 |
! |
| 60 |
****** Please insert the appropriate call for DGEMV, i.e. the library |
| 61 |
****** routine that calculates y := alpha*A*x + beta*y |
| 62 |
! |
| 63 |
t1 = MPI_Wtime() |
| 64 |
|
| 65 |
! --- Activate the following statement, or the appropriate equivalent |
| 66 |
canceled CALL DGEMV( TRANS, M, N, ALPHA1, A, LDA, X, 1, BETA1, Y, 1 ) |
| 67 |
Do 20 k = 2,krep |
| 68 |
|
| 69 |
! --- Activate the following statement, or the appropriate equivalent |
| 70 |
canceled CALL DGEMV( TRANS, M, N, ALPHAK, A, LDA, X, 1, BETAK, Y, 1 ) |
| 71 |
20 Continue |
| 72 |
t2 = MPI_Wtime() - t1 |
| 73 |
Else |
| 74 |
! |
| 75 |
****** Please insert the appropriate call for DGEMV, i.e. the library |
| 76 |
****** routine that calculates y := alpha*A'*x + beta*y |
| 77 |
! |
| 78 |
t1 = MPI_Wtime() |
| 79 |
|
| 80 |
! --- Activate the following statement, or the appropriate equivalent |
| 81 |
canceled CALL DGEMV( TRANS, M, N, ALPHA1, A, LDA, X, 1, BETA1, Y, 1 ) |
| 82 |
Do 30 k = 2,krep |
| 83 |
|
| 84 |
! --- Activate the following statement, or the appropriate equivalent |
| 85 |
canceled CALL DGEMV( TRANS, M, N, ALPHAK, A, LDA, X, 1, BETAK, Y, 1 ) |
| 86 |
30 Continue |
| 87 |
t2 = MPI_Wtime() - t1 |
| 88 |
Endif |
| 89 |
|
| 90 |
! --- Prevent smart compilers optimizing too much ... |
| 91 |
! |
| 92 |
Call ddummy( y, 1 ) |
| 93 |
! |
| 94 |
If ( t2 <= 1.0E-9_l_ .AND. krep < limrep ) Then |
| 95 |
krep = 5 * krep |
| 96 |
Goto 10 |
| 97 |
Endif |
| 98 |
! |
| 99 |
extime = t2 / krep |
| 100 |
perfor = ( nop * 1.0E-06_l_ )/Max( extime, 1.0E-9_l_ ) |
| 101 |
! ---------------------------------------------------------------------- |
| 102 |
End Subroutine libver |