fortran training - large-scale simulations with fortran 95 an object-based approach

8/7/2019 FORTRAN TRAINING - Large-scale simulations with Fortran 95 An object-based approach

1/196


2/196


3/196


4/196


5/196


6/196


7/196


8/196


9/196


10/196


11/196


12/196


13/196


14/196


15/196


16/196


17/196


18/196


19/196


20/196


21/196


22/196


23/196


24/196


25/196


26/196


27/196


28/196


29/196


30/196


31/196


32/196


33/196


34/196


35/196


36/196


37/196


38/196


39/196


40/196


41/196


42/196


43/196


44/196


45/196


46/196


47/196


48/196


49/196


50/196


51/196


52/196


53/196


54/196


55/196


56/196


57/196


58/196


59/196


60/196


61/196


62/196


63/196


64/196


65/196


66/196


67/196


68/196


69/196


70/196


71/196


72/196


73/196


74/196


75/196


76/196


77/196


78/196


79/196


80/196


81/196


82/196


83/196


84/196


85/196


86/196


87/196


88/196


89/196


90/196


91/196


92/196


93/196


94/196


95/196


96/196


97/196

interface chargemodule procedure atom_charge

end interface charge


98/196

9 8 o f 1 0 6

end interface charge

interface set_mass

module procedure set_atom_massend interface set_mass

interface set_chargemodule procedure set_atom_charge

end interface set_charge

interface as_stringmodule procedure atom_as_string

end interface as_string

contains


99/196

9 9 o f 1 0 6

contains

subroutine atom_new (a, name)

...... the rest as it was before...

...end module atoms

M od ule m olecu les i s v ery s im ila r.

module molecules


100/196

1 0 0 o f

module moleculesuse precisionimplicit noneprivate

type moleculeprivate

real(adequate) massinteger chargeinteger number_of_elements, number_of_atoms

end type molecule

public molecule, mass, charge, set_mass, &set_charge, new, as_string


101/196

1 0 1 o f

interface newmodule procedure molecule_new

end interface new

interface massmodule procedure molecule_mass

end interface mass...

T here a r e p ro s and cons t o u s ing th i sp r iva t e com po nen t s ty le .


102/196

1 0 2 o f

Pro: better locality. For example,

i f you ha ve a tom (m , c ) a l l ove r the c od e , an d

y o u g e t a n e w c o m p o n e n t , th o s e r e f e re n c e sm us t be a l l cha ng ed . You ha ve m oref lex ib i l ity i f there a re o n e o r m o reco ns t ruc to r s t ha t you w r ite , pos s ib ly

us ing ke yw ord o r op t ion a l a rgu m en ts .

O O p rog ram m ers h av e lea rne d tha t if a c l ien tcan R E A D a co m po ne n t d i r ec t ly the y canW R IT E i t, po ss ib ly c rea t ing an ob jec t tha t


103/196

1 0 3 o f

W R IT E i t, po ss ib ly c rea t ing an ob jec t tha tn o lo n g e r h a s a v a lid v a lu e .

In f ac t , a .b o n the l e f t side o f an a s s ign m en top e ra to r i s a r ed f lag to an exp e r ience d O Op ro g ra m m e r.

C o n : ta k e s m o re w o r k .

Besides, writing a%b = c is nice and clear and I never make


104/196

1 0 4 o f

gmistakes. And I can do the generic functions whether or notthe components are private. Anyway, I dont trust thecompiler to do those accessors efficiently.

H o m e w o r k

Write a program with a user-defined type point that has x,h d f d


105/196

1 0 5 o f

y, z components. Write another user-defined typequadrilateral that has a component vertices that is anarray of 4 points. (For simplicity, make everything public).

Try creating and manipulating some quadrilaterals andarrays of quadrilaterals.

If q is an array of quadrilaterals,q(1)%vertices(1)%x

is the xth coordinate of the first vertex of the firstquadrilateral in q.

But can you do q(1)%vertices%x, q(1)%vertices(2:3)%x,q(2:4)%vertices%x?

Just how cool is this?


106/196

1 0 6 o f

Just how cool is this?

And what about

integer:: iv(3) = (/ 1, 2, 4 /)

real:: integer x(20)

What about:x(iv)?

q(1)%vertices(iv)?


107/196

1 0 7 o f


108/196

1 0 8 o f

Lesson 3: Types and objects

Paul F. Dubois

X-Division

[email protected]

In ou r f ir s t tw o l e ssons , we ha veconcen tr a t ed on a r r ay s , m odu le s , andder ived types .


109/196

1 0 9 o f

It is the combination of types and modules that allows us towork in a style I will call object-based.

The homework should have helped you get a feel forcreating and working with derived types.

M in i -le s son #3 : G ive ea ch c om p on en t in ade r ive d typ e a d efau lt in i t ia l va lue .

type particle


110/196

11 0 o f 1 5 1

type particletype(point):: location = point(0.0, 0.0, 0.0)type(point):: momentum = point(0.0, 0.0, 0.0)real(adequate):: mass = 1.0real(adequate):: charge = 0.0

end type particle

The default initial value can be over-ridden by an explicitone.You must use the entity-oriented form of declaration with

the two colons.

W e w ish to im plem ent an

ab strac t data type, inw hich no illeg al o bjects


111/196

111 o f 151

yp ,w hich no illeg al o bjectsof this type can ever be

made .Initializing each instance properly is an important part of this.

M od u les an d de r ived types ea ch ha ve som easp ec t o f c lasses .

Modules can contain both data and functions, and do


112/196

11 2 o f 1 5 1

Modules can contain both data and functions, and doinformation hiding. But there is only one instance of thedata in a module.

Derived types have a constructor that can be used as acookie cutter to produce instances, but they do not containmethods.

Derived types do have the possibility of hiding all or part of their data but you lose some conveniences if you do so.

=> The secret is to combine the two.

M in i -le s son #4 : U se a m odu le t o m ode ls ing le tons .

A singleton is a class which has exactly one instance.


113/196

11 3 o f 1 5 1

g y

So, if you have some data which should exist exactly once

in your application, package it into a module, together withthe functions whose exclusive purpose is to manipulate thatdata.

Then think out what you want the public interface to look like.

T h e g en er ic in te r face ca pa b ili ty is m u chm ore p ow er fu l than w e have s een so f a r.

We used it to define a single name that could be applied to


114/196

11 4 o f 1 5 1

g ppobjects of varied types. You can also use it to:

d e f in e o p e r a to r s s u c h a s + a n d * ; ad d ne w op e ra to r s su ch as .do t ., .i n teg ra te . ,

o r . inv e r se . , o f y ou r ow n ch oo s ing ;

t ake ov e r the a s s ign m en t ope ra to r ;

ev en ex ten d the m ean ing o f in t r ins i c n am ess u c h a s s q rt

H e re is a m o d u le im p le m e n t in g a 3 - D p o in tob jec t .

module points


115/196

11 5 o f 1 5 1

puse precisionprivatepublic dot, sqrt, normpublic operator(+), operator(-), operator(*), operator(/)

type,public:: pointreal(adequate):: x = 0.0real(adequate):: y = 0.0real(adequate):: z = 0.0

end type point

Weve defined a new type named point. Weve givenevery component a default value so there wont be anyuninitialized points.


116/196

11 6 o f 1 5 1

Note the alternate syntax used for giving type(point) the

attribute public.

N ow w e de fine t he ba s ic a r ithm e ticope ra to r s to im p lem en t vec to r a r it hm e ticon po in t s .


117/196

11 7 o f 1 5 1

We supply three versions of each so that we can dopoint+point, point+real, real+point, for convenient

operations such as translation and scaling.

interface operator(+)module procedure point_add, point_add_s, point_s_add

end interface

interface operator(-)module procedure point_subtract, point_subtract_s, &

point_s_subtract, point_negateend interface


118/196

11 8 o f 1 5 1

end interface

interface operator(*)module procedure point_multiply, point_multiply_s, &point_s_multiply

end interface

interface operator(/)module procedure point_divide, point_divide_s, &

point_s_divideend interface

M od u le p rocedu re s d o the r ea l w ork .

function point_add (a, b) result(r)type(point), intent(in):: a, b


119/196

11 9 o f 1 5 1

type(point):: rr%x = a%x + b%xr%y = a%y + b%yr%z = a%z + b%z

end function point_add

function point_add_s (a, b) result(r)type(point), intent(in):: areal(adequate), intent(in):: b

type(point):: r

r%x = a%x + br%y = a%y + br%z = a%z + b

end function point add s


120/196

1 2 0 o f

end function point_add_s

function point_s_add (a, b) result(r)type(point), intent(in):: breal(adequate), intent(in):: atype(point):: r

r%x = a + b%xr%y = a + b%yr%z = a + b%z

end function point_s_addThese are all in the procedure part of the module points.

T h e com p i le r ch oo ses an ap prop r ia t efun c tion ba sed on s ign a tu re .

type(point) x, y, z


121/196

1 2 1 o f

z = x + y !actually does z = point_add(x, y)

z = x + 1.0 !actually does z = point_add_s(x, 1.0)

z = 1.0 + x !actually does z = point_s_add(1.0, x)

W e a lso de f ine a gen e r ic no rm func t ion ,a b ina ry o pe ra to r fo r do t p rod uc t , and ane lem en t -w ise squ a re roo t.

Users can write norm(v) v dot w and sqrt(v) And yes


122/196

1 2 2 o f

Users can write norm(v), v .dot. w, and sqrt(v). And yes,sqrt(2.0) still works!

interface normmodule procedure point_norm

end interface

interface operator(.dot.)module procedure point_dot

end interface

interface sqrtmodule procedure point_sqrt

end interface


123/196

1 2 3 o f

W e lis ted in the in t e r face fo r each o pera to rthe fu n c t ion s tha t th e co m p ile r co u ld u se to

d o t h e jo b .

The functions were all in the module itself so all we had to


124/196

1 2 4 o f

The functions were all in the module itself so all we had todo was list them as module procedures.

We could also use functions defined elsewhere by puttinginterface specifications for them into the interface blocks.

In the contains section of the module we put the actualworker routines. These are all private to the module.

I t is a l so h e lp fu l to a dd un ary m inu s to thesub t rac t in t e r face .

function point_negate (a) result(r)type(point) intent(in):: a


125/196

1 2 5 o f

type(point), intent(in):: atype(point):: r

r%x = -a%xr%y = -a%yr%z = -a%z

end function point_negate

N o w w e d o t h e n o r m , .d o t ., a n d s q r t.

function point_norm(self) result(r)type(point),intent(in):: self

l( d )


126/196

1 2 6 o f

real(adequate) rr = sqrt(self%x**2 + self%y**2 + self%z**2)

end function point_norm

function point_dot(a, b) result(r)type(point), intent(in):: a, breal(adequate):: rreal(precise):: s


127/196

1 2 7 o f

s = real(a%x, precise) * b%x + &real(a%y, precise) * b%y + &real(a%z, precise) * b%z

r = real(s, adequate)end function point_dot

function point_sqrt(a) result(r)type(point), intent(in):: atype(point):: r

r%x = sqrt(a%x)


128/196

1 2 8 o f

r%y = sqrt(a%y)r%z = sqrt(a%z)

end function point_sqrtend module points

L et s te s t -d r ive m od u le po in t s .

type(point), parameter:: origin = point(0.0, 0.0, 0.0)type(point), parameter:: e1 = point(1.0, 0.0, 0.0)type(point) parameter:: e2 = point(0 0 1 0 0 0)


129/196

1 2 9 o f

type(point), parameter:: e2 = point(0.0, 1.0, 0.0)type(point), parameter:: e3 = point(0.0, 0.0, 1.0)type(point) x1, x2

x1 = 3.0 * e1 + 4.0 * e2 + 5.0 * e3x2 = -4.0 * x1 - (2.0_adequate * e2) + e3write(11, *) "x1 ", x1write(11, *) "x2 ", x2write(11, *) "norm(x1)", norm(x1)write(11, *) "sqrt(x1)", sqrt(x1)

write(11, *) "dot product", x1 .dot. x2

N o te th a t k in d c o n v e r s io n w a s p e r fo r m e d ,bu t typ e conve r s ion w on t be .

x1 = 3.0 * e1 + 4.0 * e2 + 5.0 * e3 !ok x1 = 3.0 * e1 + 4.0 * e2 + 5 * e3 !doesnt compile


130/196

1 3 0 o f

x1 3.0 e1 4.0 e2 5 e3 !doesn t compile

Naturally, you could define two more handlers so thateach operation would work with integers.

Su pp ose w e w an t to r ep re sen t po in t s on t heun it c ir c le , an d to b e ab le to a sk fo r the ir x ,

y, an d th e ta coo rd ina tes .However, we always want 0.0


131/196

1 3 1 o f

We also want to be able to add and subtract angles fromsuch points, and to create the points from real numbersrepresenting the angle.

H ere is how w e a r e go ing to u se en t it ie s o f type ucp o in t.

type(point) atype(ucpoint) r1, r2


132/196

1 3 2 o f

yp ( p ) ,

r1 = pi / 4.0_adequater2 = -r1 + pi / 12.0_adequatewrite (11, *) "r1 ", x (r1), y (r1), theta (r1)write (11, *) "r2 ", x (r2), y (r2), theta (r2)write (11, *) "r3 ", theta (as_ucpoint (-0.3 * pi))a = point(x(r1), y(r1), 0.0)write (11, *) "a ", a

module ucpoints !points on the unit circleuse precisionprivate


133/196

1 3 3 o f

privatepublic ucpoint, as_ucpointpublic operator(+), operator(-), assignment(=)public x, y, theta

type ucpointprivatereal(adequate):: theta = 0.0

end type ucpoint

interface assignment(=)module procedure ucpoint_set

end interface

i t f t (+)


134/196

1 3 4 o f

interface operator(+)module procedure ucpoint_add_s, ucpoint_s_add

end interface

interface operator(-)

module procedure ucpoint_subtract_s, &ucpoint_s_subtract, ucpoint_negateend interface

N o w in t h e p r o ce d u r e p a r t of th e m o d u lew e de f ine t he se op e ra to r s.

containsfunction as_ucpoint(r) result(p)


135/196

1 3 5 o f

real(adequate), intent(in):: r

type(ucpoint) pp = r

end function as_ucpoint

function x(p)type(ucpoint), intent(in):: preal(adequate) x

x = cos(p%theta)end function x

function y(p)type(ucpoint), intent(in):: preal(adequate) y

y = sin(p%theta)


136/196

1 3 6 o f

y = sin(p%theta)end function y

function theta(p)type(ucpoint), intent(in):: p

real(adequate) rtheta = p%thetaend function theta

To de f ine the a s s ign m en t o pe ra to r w esupp ly a su b rou t ine .

You can have multiple assignment handlers with differentsignatures for the second argument.


137/196

1 3 7 o f

subroutine ucpoint_set (p, r)

type(ucpoint), intent(out):: preal(adequate), intent(in):: rp%theta = modulo(r, 2.0_adequate * pi)

end subroutine ucpoint_set

Note that the intent of p must be out or inout. The intent of rmust be in.

The statement p = r will be handled by a call ucpoint_set(p,r).

B y u s ing ou r ow n ass ign m en t op e ra to r, w ecan insu re tha t the ta i s a lw ays in the

des ir ed r ang e .function ucpoint_add_s (a, b) result(p)

( i ) i (i )


138/196

1 3 8 o f

type(ucpoint), intent(in):: a

real(adequate), intent(in):: btype(ucpoint):: p

p = a%theta + b

end function ucpoint_add_s

Note that the assignment is really ucpoint_set (p, a%theta +b).

Thus, ucpo int_set m akes

sure on ly legal ucpointscan be created.


139/196

1 3 9 o f

If each point is born correct, and each (public) operationkeeps it correct, it is always correct.

F or t ran 95 s d e r ived types com pa reun favo rab ly to C ++ s .

Com parison o f derived types in F95and C++


140/196

1 4 0 o f

N etEffec tfo r F 9 0v s .C + +

A rea o f conce rn

M in u s T h e re a re n o p ro v is io n sfor dest ructors .

M in u s T h e re is o n ly th e o n eofficial con struc tor.


N etEffec tfo r F 9 0 A rea o f conce rn


141/196

1 4 1 o f

S lig h t P l u s T h e re is n o a u to m a tic s ig -na tu re m a tch ing o rim pl ic i t conve rs ion s .

H u g eM i n u s N o inher i tanc e in F or tran9 5

H u g eM i n u s

N o pa ram e te r ized typesin F9 5

H u ge P lu s Typ e sa fe ty

o 9 0v s .C + +

ea o co ce


N etEffec tfo r F 9 0 A rea o f conce rn


142/196

1 4 2 o f

H u g e P l us In te ra c tio n o f a rra y fa c ilityw ith der ived types .

H u g e P lu s E a s e o f le a rn in g to u se

de r ived typ e s correct ly.

v s .C + +

T h e job o f a des t ruc to r is to re lease an yasse t s acq u i red b y an ob jec t d u r ing i ts

l i fe t ime.module curves !points on the unit circle

use precision


143/196

1 4 3 o f

use precision

privatepublic curvepublic length, segment, create, destroypublic get_x, get_y

type curveprivatereal(adequate), pointer:: x(:) => Null()real(adequate), pointer:: y(:) => Null()

end type curve

interface createmodule procedure curve_create

end interface create


144/196

1 4 4 o f

interface destroymodule procedure curve_destroy

end interface destroy

interface lengthmodule procedure curve_length

end interface length

D efine a s s ign m en t o f one cu rv e to an o the rso tha t it m ak es a cop y.

interface assignment(=)module procedure curve_set_curve

end interface


145/196

1 4 5 o f

end interface

T he con s truc to r an d de s t ruc to r andas s ign m en t ope ra to r w ork t oge the r to

av o id m em ory leak s .subroutine curve_create(c, x, y)

real(adequate), intent(in):: x(:), y(:)


146/196

1 4 6 o f

real(adequate), intent(in):: x(:), y(:)

type(curve), intent(inout):: cif(size(x) /= size(y)) thenstop "curve assignment: incorrect sizes"

endif

call curve_destroy(c)allocate(c%x(size(x)))allocate(c%y(size(y)))c%x = xc%y = y

end subroutine curve_create

subroutine curve_set_curve(p, q)type(curve), intent(inout):: ptype(curve), intent(in):: q


147/196

1 4 7 o f

call curve_create (p, q%x, q%y)end subroutine curve_set_curve


148/196

M em ory in t eg r it y t hen dep end s on ap r o g r a m m e r c o n v e n t io n to d e stro y e a c h

it em tha t i s c rea ted .subroutine try_curves

use precision


149/196

1 4 9 o f

p

use curvesimplicit nonetype(curve) c1, c2, c3real(adequate) xx(4), yy(4)

xx = (/ 1., 2., 3., 4. /)yy = xx**2 + 1call create(c1, xx, yy)call create(c2, xx, yy / 2.)write(11, *) "y(c2) ", y(c2)

... some coding which may or may not use c3....call destroy(c1)

call destroy(c2)call destroy(c3)

end subroutine try_curves


150/196

1 5 0 o f

H o m e w o r k

A tic-tac-toe board consists of a 3 by 3 matrix. Each entrycan either be empty, an X, or an O. Create a tic-tac-toeboard object which is sufficiently rich to support writing atic-tac-toe game. In the game, users can undo moves right


151/196

1 5 1 o f

g g , g

back to the start. The board class should refuse to makeillegal moves and return an error flag if so requested.

There is no right answer.

How robust you make the input procedure is up to you.


152/196

1 5 2 o f

Lesson 4: New features


153/196

1 5 3 o f

Lesson 4: New features

Paul F. Dubois

X-Division

[email protected]

To com plete ou r u pg rad e t o F95 , th i sw eek w e ll cove r som e o f t he new f ea tu re s

o f the lang uag e .Scaling considerations

More about working with arrays


154/196

1 5 4 o f

More about working with arrays

Looping and control statements

More intrinsic functions

Whats still missing?

Caution: todays example is strongly F95, not just F90.Youll have to comment out quite a few things to get it to

compile F90.

T h e en ve lope , p l ease ...

(The contest was won by George Zimmerman).


155/196

1 5 5 o f

T he m od u le is the k ey to ha pp ines s , bu tthe re a re p rob lem s fo r l a rge p rogr am s .

The module is what remembers information between files.So clearly the compiler must remember some informationabout each module it encounters.


156/196

1 5 6 o f

Typically, this is done by creating a module_name.mod file.(Not universal, not standard).

When the compiler sees a use statement, it looks for the

.mod file.

M od u le use m us t fo rm an acyc l ic d ir ectedg r a p h .

Depending on the situation, you may decide to separate datafrom functions.


157/196

1 5 7 o f

module a_data...end module a_datamodule b_data...end module b_datamodule a

use a_data

use b_datacontains

functions that manipulate a_databut sometimes need b_data

end module a_data


158/196

1 5 8 o f

Similarly for b.

M in i -le s son #5 : O ne m od u le , on e f ile

Often the module contains one type and implements anobject.

You can put module statements in a file with other programunits but dont


159/196

1 5 9 o f

units, but don t.

F 95 sup p or t s go - to - le s s loop s .

x = (/ (i, i=1, n) /)y = sqrt(x**2/3.0)

The old way to find a maximum...t = x(1)


160/196

1 6 0 o f

t = x(1)k = 1do i = 2, n

if(x(i) > t) then

t = x(i)k = iendif

enddo

New way #1...t = x(1)

k = 1do i = 2, n

if(x(i)


161/196

1 6 1 o f

( )k = i

enddo

New way #2...

t = maxval(x)k = maxloc(x, 1)

T h e w o r d e x it n o w m e a n s to le a v e a lo o p ,no t the p rogra m !

t = 12.0k = 0do i = 1, n


162/196

1 6 2 o f

if(x(i) == t) thenk = iexit

endif

enddo

T h e do w hi le loop is o f ten used fo r it e ra tiveprocesses .

k = 0do while(k /= 0)

if(x(i) == t) k = i


163/196

1 6 3 o f

end do


164/196

write(11, *) "Sum of y elements within x range is ", t


165/196

1 6 5 o f

A labe l c an b e u sed t o con t ro l behav io r innes ted loo ps .

outer: do i=1, ninner: do j = 1, m

...


166/196

1 6 6 o f

if (...) exit outerenddoenddo

However, loops of any sort are not fun to write.

F95 to the rescue!

M ask ed a s s ign m ent e lim ina tes the ne ed fo rm a n y lo o p s .

where (w1


167/196

1 6 7 o f

z = 0.0end wherewrite(11, *) "Same thing, without a loop ", sum(z)

T he w he re con s truc t a lso ha s a on e -s ta tem en t fo rm .

logical cond(size(y))...z = 0.0


168/196

1 6 8 o f

cond = w1


169/196

1 6 9 o f

enddo

Now we face the same inequity with the array syntax:x = sqrt(y) !ok

x(1) = f(y(1))x = vector_version_of_f(y)

Generic interfaces would work so that you could just writef(y) but you would have to write a separate function for

each rank of array.


170/196

1 7 0 o f

I f f is dec la red e lem en ta l , f(x ) w ill w orkon an y a r r ay.

elemental function f(x)real(adequate), intent(in):: xreal(adequate):: f


171/196

1 7 1 o f

f = sqrt(x / 3.0)end function f

where(cond)

z = f(x)elsewhere

z = 0.0endwhere

write(11, *) "Same thing, using elemental function ",sum(z)

x = x - 4.0 ! make some values negativew1 = 0.0cond = w1


172/196

1 7 2 o f

where(cond)z = f(x)

elsewherez = 0.0

endwherewrite(11, *) "f(x) not evaluated at bad points ", sum(z)

L oo p s an d fun c t ion s d isgu i se in fo rm a t ionyou m ay ha ve abo u t p a r a l le lism .

Two new facilities are provided.

You can d ec l a re a f un c t ion p u re t o ind i ca tei t ha s no s ide e ffec ts .


173/196

1 7 3 o f

T he fo r a ll con s truc t exp re s se s p a r a lle l loo psco m p ac t ly an d c lear ly.

T h e fo ra ll s ta t em en t a lso com es in tw of lavors .

integer c(6)

forall(i=1:6) c(i) = 3 - i


174/196

1 7 4 o f

forall(i=1:6)c(i) = 3 - i

end forall

M u lt ip le d im en s ion s rep lace n es ted loo pcons t ruc t s .

integer, parameter:: n= 5, m = 6integer a(n, m), b(n, m)

forall(i=1:n, j=1:m) a(i,j) = 10 * i + j


175/196

1 7 5 o f

forall(i=1:n, j=1:m)a(i, j) = 10 * i + j

b(i, j) = x(i) - y(j)end forall


176/196

N am elis t is no w s tand a rd i zed .

You declare a namelist group to contain a group of variablesthat you wish to read or write as a group.

namelist /ab/ a, b


177/196

1 7 7 o f

T he ab is c a l led t he g roup nam e .

I t c an b e u se d fo r inp u t o r ou tpu t. I t ran read w h a t i t w ro te .

W h at i t w ro te m ig h t b e u g ly.

You use t he N M L keyw ord in a r ead o rw r it e s t a tem en t in s t ead o f a fo rm at .

forall(i=1:n, j=1:m)a(i,j) = i + jb(i,j) = i - j

end foralli ( b)


178/196

1 7 8 o f

write(11, NML=ab)

Here is what it wrote.&ABA = 11, 21, 31, 41, 12, 22, 32, 42, 13,

23, 33, 43, 14, 24, 34, 44, 15, 25, 35,45,

B = 2*0, 2*1, 2*0, 2*1, 0, 3*1, 0, 3*1, 0, 3*1/

You can ca ll f unc t ion s u s ing keyw ords .

subroutine go(x, y, z)

call go(1, 2, 3)

call go(1, z=4, y=2)

call go(x=1 z=2 y=3)


179/196

1 7 9 o f

call go(x=1, z=2, y=3)

To do this, subroutine go must have an explicit interface.

T h ere a re som e in te res ting a r ray -o r i en tedin t r ins ic fu n c t ion s .

spread is used to replicate parts of arrays to make largerdimensional versions.

a = 10 * spread(iotan dim=2 ncopies=m) + &


180/196

1 8 0 o f

a = 10 spread(iotan, dim=2, ncopies=m) + &spread(iotam, dim=1, ncopies=n)

is the same as:

forall(i=1:m, j=1:n) a(i,j) = 10 * i + j

We could have done our outer product with spread.

H e re a r e so m e o t h e r a r r a y fu n c tio n s y o um igh t w an t to exp lo re .

csh i ft -- c i rcu la r sh i f t

eo sh i ft -- end -o ff sh i ft

m atm u l -- m a trix m u ltip ly


181/196

1 8 1 o f

m a tm u l -- m a trix m u ltip ly d o t _ p r o d u c t

co un t - - cou n t ing true e l em en t s in log ica le x p r e s s i o n s

m erge ( true_ sou rce , f a lse_ sou rce , m ask )

p ack (ar ray, m ask ) - - co m p ress

T h e resh ap e fun c tion is in c red ib lyp o w e r f u l .

Use it for reshaping and reordering subscripts.

reshape(source, shape, pad, order)


182/196

1 8 2 o f

reshape( iota(6), (/ 2, 3 /)) is the matrix1 3 52 4 6

reshape( iota(6), (/ 2, 4 /), (/ 0, 0 /), (/ 2, 1 /))1 2 3 45 6 0 0

R e sh a p e c a n b e u s ed t o m a k e a n a r r a y o n e -d i m e n s i o n a l

real(adequate) a(n,m), z(n*m)z = reshape(a, (/ size(a) /))

You can also do it with transfer(a,a). The transfer function isextremely powerful non portable and fun


183/196

1 8 3 o f

extremely powerful, non-portable, and fun.

I shouldnt tell you about this... -- Jeanne Martin

I deny ever telling you about it.

M os t o f the inqu iry o pe ra to r s on a r ray s a reab l e to t ake an ex t ra a rgum en t ind ica t ing ad i m e n s i o n .

shape(a) 4 5

Sum up columnssum(a,1) 14 18 22 26 30


184/196

1 8 4 o f

sum(a,1) 14 18 22 26 30

Sum up rows

sum(a,2) 20 25 30 35

Sum up arraysum(a) 110

A lso , m os t o f the sca la r in t r ins ic s a ree lem en ta l , so yo u can u se them on a r ray s .

sq r t , ex p , cos , . ..

d im (x , y ) - - x - y bu t f loo red a t ze ro

r and om n um be r (z )


185/196

1 8 5 o f

_ ( )

So w hy d idn t I wa n t to le a rn F 90 b e fo re?

Stup id i ty

N o c o m p i lers

N o p erfo rm an ce

but more seriously


186/196

1 8 6 o f

but more seriously

N o inh er i tan ce

N o p o lym o rph ism / d y n am ic b ind ing N o ga rbag e co l lec t ion

N o ex cep t ion han d l ing

N o ge ner i c type s

N o c o m m an d l ine a rgu m en ts

N o s igna l h and l ing

N o s t r ing s


187/196

1 8 7 o f

T h e s t r in g fac ilit ies in F 95 a re s t illm a s siv e ly a n n o y in g .

character(len=32):: s = hellocharacter(len=32):: t = worldcharacter(len=32):: greetinggreeting = s // ' ' // t !Not...sets greeting to hellogreeting = trim(s) // ' '


188/196

1 8 8 o f

g g ( )greeting(len(trim(s)) + 1: ) = t !Just shoot me...

You cant even write a function that returns a variably sizedstring such as a file name with no trailing blanks, I thought.

(I was wrong, as we shall see.)

H e y ! I d e a ! I c a n w r ite m y o w n s tr in g ty p e !

type stringcharacter(len=1), pointer:: storage(1)integer capacity

integer lengthend type string


189/196

1 8 9 o f

end type string

and go ahead and make a string object out of it.

Hey! Wait! If this is such a great idea, why didnt ISO do it?

A ctu a l ly, ISO d id d o i t.

They made a separate standard for it and never mention itout loud.

I had to ask Jeanne Martin for it, and she didnt even have acopy.


190/196

1 9 0 o f

MODULE ISO_VARYING_STRING

! Written by J.L.Schonfelder! Incorporating suggestions by C.Tanasescu, C.Weber, J.Wagener and W.Walter,! and corrections due to L.Moss, M.Cohen, P.Griffiths, B.T.Smith! and many other members of the committee ISO/IEC JTC1/SC22/WG5

! Version produced (??-Jul-94)

!-----------------------------------------------------------------------------!! This module defines the interface and one possible implementation for a !! dynamic length character string facility in Fortran 90. The Fortran 90 !! language is defined by the standard ISO/IEC 1539 : 1991. !! The publicly accessible interface defined by this module is conformant !

! with the auxilliary standard, ISO/IEC 1539-2 : 1994. !! The detailed implementation may be considered as an informal definition of !! the required semantics, and may also be used as a guide to the production !! of a portable implementation. !! N.B. Although every care has been taken to produce valid Fortran code in !

! construction of this module no guarantee is given or implied that this !! code will work correctly without error on any specific processor, nor !! is this implementation intended to be in any way optimal either in use !! of storage or CPU cycles. !!-----------------------------------------------------------------------------!

Gosh, how reassuring...


191/196

1 9 1 o f

Indeed , va ry ing s t ring s eem s to b es o m e th i n g lik e w h a t I im a g in e d .

TYPE VARYING_STRINGPRIVATECHARACTER,DIMENSION(:),POINTER :: chars

ENDTYPE VARYING_STRING...


192/196

1 9 2 o f

PUBLIC :: VARYING_STRING,VAR_STR,CHAR,LEN,GET,PUT,PUT_LINE,INSERT,REPLACE,SPLIT,REMOVE,REPEAT,EXTRACT,INDEX,SCAN,VERIFY,LLT,LLE,LGE,LGT,ASSIGNMENT(=),OPERATOR(//),OPERATOR(==),OPERATOR(/=),OPERATOR(),LEN_TRIM,TRIM,IACHAR,ICHAR,

ADJUSTL,ADJUSTR

L et s t ake it fo r a t es t d r ive .

subroutine try_string! try the ISO varying_string

use iso_varying_stringimplicit nonetype(varying_string) s, t, us = "hello"


193/196

1 9 3 o f

t = "world"u = s // var_str(" ") // t

write(11, *) char(u)end subroutine try_string

...and it nicely writes out hello world.

T h e s tr in g tha t is p r in ted tu rned ou t no t toh a v e a n y o b n o x i o u s p a d d in g o n t h e e n d .

Indeed, the source for the char() operator teaches us how:

FUNCTION s_to_c(string)

type(VARYING_STRING),INTENT(IN) :: stringCHARACTER(LEN=SIZE(string%chars)) :: s to c


194/196

1 9 4 o f

CHARACTER(LEN SIZE(string%chars)) :: s_to_c! returns the characters of string as an automatically! sized characterINTEGER :: lclc=SIZE(string%chars)DO i=1,lc

s_to_c(i:i) = string%chars(i)

ENDDOENDFUNCTION s_to_c

Remember how we couldnt do real(kind(x)) for anargument x?That doesnt apply here as it is simply a matter of how much

automatic storage is needed.


195/196

1 9 5 o f

H o w y o u g o n n a k ee p e m , d o w n o n th efa rm. . .

Now that you have looked over, and seen the promised land,Im sure youll get there.

I hope you have enjoyed your upgrade experience.

There is no support number.


196/196

1 9 6 o f

-- Paul

fortran training - large-scale simulations with fortran 95 an object-based approach

Documents