APPENDIX 5

Intrinsic functions in Fortran 90

Introduction

1. Function which determines if a certain argument is in an actual argument list

2. Numerical functions

3. Mathematical functions

4. Character string functions

5. Character string function for request

6. Kind functions

7. Logical function

8. Numerical inquiry functions

9. Bit inquiry function

10. Bit manipulation functions

11. Transfer functions

12. Floating-point manipulation functions

13. Vector- and matrix-multiplication functions

14. Array functions

15. Array inquiry functions

16. Array construct functions

 17. ARRAY reshape function

18. ARRAY manipulation functions

19. Array location functions

20. Pointer inquiry functions

21. Intrinsic subroutines

o Time routines

o Bit copy routine

o Random number routines

Introduction
There is a large a number of intrinsic functions and five intrinsic subroutinesin Fortran
90. I treat the numeric and mathematical routines very shortly, since they are not
changed from Fortran 77 and therefore should be well-known.

This section is based on section 13 of the ISO standard (1991), which contains a more
formal treatment. We follow the arrangement of the different functions and
subroutines in the standard, but explain directly in the list. For a more detailed
treatment we refer to Metcalf and Reid (1990, 1993).

When a parameter below is optional it is given in lower case characters. When an

argument list contains several arguments the function can be called either by position
related arguments or by a keyword. Keyword must be used if some previous argument
is not included. Keywords are normally the names that are given below.

We have not always given all the natural limitations to the variables, for example that
the rank is not permitted to be negative.

1. Function which determines if a certain argument is in an

actual argument list:
The function PRESENT(A) returns .TRUE. if the argument A is in the calling
list, .FALSE. in the other case. The use is illustrated in the example program in chapter
8 of the main text.
2. Numerical functions:
The following are available from Fortran 77: ABS, AIMAG, AINT, ANINT, CMPLX,
CONJG, DBLE, DIM, DPROD, INT, MAX, MIN, MOD, NINT, REAL and SIGN.

In addition, CEILING, FLOOR and MODULO have been added to Fortran 90. Only the last
one is difficult to explain, which is most easily done with the examples from ISO
(1991)
MOD (8,5) gives 3 MODULO (8,5) gives 3
MOD (-8,5) gives -3 MODULO (-8,5) gives 2
MOD (8,-5) gives 3 MODULO (8,-5) gives -2
MOD (-8,-5) gives -3 MODULO (-8,-5) gives -3
The following functions from Fortran 77 can use a kind-parameter like in AINT(A,
kind), namely AINT, ANINT, CMPLX, INT, NINT and REAL.

A historic fact is that the numerical functions in Fortran 66 had to have specific
(different) names in different precisions, and these explicit names are still the only
ones which can be used when a function name is passed as an argument.

A complete table of all the numerical functions follow. Those names that are indicated
with a star * are not permitted to be used as arguments. Some functions,
like INT and IFIX have two specific names, either can be used. On the other hand,
some functions do not have any specific name. Below I use C for complex floating
point values, D for floating point values in double precision, I for integers, and R for
floating point values in single precision.

Function Generic Specific Data type

name name Arg Res

Conversion INT - I I
to integer * INT R I
* IFIX R I
* IDINT D I
(of the real part) - C I

Conversion REAL * REAL I R

to real * FLOAT I R
- R R
* SNGL D R
(real part) - C R

Conversion DBLE - I D
to double - R D
precision - D D
(real part) - C D
Conversion CMPLX - I (2I) C
to complex - R (2R) C
- D (2D) C
- C C

Truncation AINT AINT R R

DINT D D

Rounding ANINT ANINT R R

DNINT D D
NINT NINT R I
IDNINT D I

Absolute ABS IABS I I

value ABS R R
DABS D D
CABS C R

Remainder MOD MOD 2I I

AMOD 2R R
DMOD 2D D
MODULO - 2I I
- 2R R
- 2D D

Floor FLOOR - I I
- R R
- D D

Ceiling CEILING - I I
- R R
- D D

Transfer SIGN ISIGN 2I I

of sign SIGN 2R R
DSIGN 2D D

Positive DIM IDIM 2I I

difference DIM 2R R
DDIM 2D D

Inner product - DPROD R D

Maximum MAX * MAX0 I I

* AMAX1 R R
* DMAX1 D D
- * AMAX0 I R
- * MAX1 R I

Minimum MIN * MIN0 I I

* AMIN1 R R
* DMIN1 D D
- * AMIN0 I R
- * MIN1 R I

Imaginary part - AIMAG C R

Conjugate - CONJG C C
Truncation is towards zero, INT(-3.7) becomes -3, but rounding is correct, NINT(-
3.7) becomes -4. The new functions FLOOR and CEILING truncate towards minus and
plus infinity, respectively.

The function CMPLX can have one or two arguments, if two arguments are present
these must be of the same type but not COMPLEX.

The function MOD(X,Y) calculates X - INT(X/Y)*Y.

The sign transfer function SIGN(X,Y) takes the sign of the second argument and puts it
on the first argument, ABS(X) if Y >= 0 and -ABS(X) if Y < 0.

Positive difference DIM is a function I have never used, but DIM(X,Y) gives X-Y if this
is positive and zero in the other case.

Inner product DPROD on the other hand is a very useful function which gives the
product of two numbers in single precision as a double precision number. It is both
fast and accurate.

The two functions MAX and MIN are unique in that they may have an arbitrary number
of arguments, but at least two. The arguments have to be of the same type, but are not
permitted to be of type COMPLEX.

3. Mathematical functions:
Same as in Fortran 77. All trigonometric functions work in radians. The following are
available: ACOS, ASIN, ATAN, ATAN2, COS, COSH, EXP, LOG, LOG10, SIN, SINH,
SQRT, TAN and TANH.

A historic fact is that the mathematical functions in Fortran 66 had to have specific
(different) names in different precisions, and these explicit names are still the only
ones which can be used when a function name is passed as an argument.

A complete table of all the mathematical functions follow. Below I use C for complex
floating point values, D for floating point values in double precision, I for integers,
and R for floating point values in single precision.

Function Generic Specific Data type

name name Arg Res

Square root SQRT SQRT R R

DSQRT D D
 CSQRT C C

Exponential EXP EXP R R

DEXP D D
CEXP C C

Natural LOG ALOG R R

logarithm DLOG D D
CLOG C C

Common LOG10 ALOG10 R R

logarithm DLOG10 D D

Sine SIN SIN R R

DSIN D D
CSIN C C

Cosine COS COS R R

DCOS D D
CCOS C C

Tangent TAN TAN R R

DTAN D D

Arcsine ASIN ASIN R R

DASIN D D

Arccosine ACOS ACOS R R

DACOS D D

Arctangent ATAN ATAN R R

DATAN D D
ATAN2 ATAN2 2R R
DATAN2 2D D

Hyperbolic SINH SINH R R

sine DSINH D D

Hyperbolic COSH COSH R R

cosine DCOSH D D

Hyperbolic TANH TANH R R

tangent DTANH D D
The purpose of most of these functions is obvious. Note that they are all only defined
for floating point numbers, and not for integers. You can therefore not calculate the
square root of 4 as SQRT(4), but instead you can use NINT(SQRT(REAL(4))). Please also
note that all complex functions return the principal value.

The square root gives a real result for a real argument in single or double precision,
and a complex result for a complex argument. So SQRT(-1.0) gives an error message
(usually already at compile time), while you can get the complex square root using the
following statements.
COMPLEX, PARAMETER :: MINUS_ONE = -1.0
COMPLEX :: Z
Z = SQRT(MINUS_ONE)
The argument for the usual logarithms has to be positive, while the argument
for CLOG must be different from zero.

The modulus for the argument to ASIN and ACOS has to be at most 1. The result will be
within [-pi/2, pi/2] and [0, pi], respectively.

The function ATAN will return a value in [-pi/2, pi/2].

The function ATAN2(Y,X) = arctan(y,x) will return a value in (-pi, pi]. If Y is positive
the result will be positive. If Y is zero the result will be zero if X is positive, and pi
if X is negative. If Y is negative the result will be negative. If X is zero the result will
be plus or minus pi/2. Both X and Y are not permitted to be zero simultaneously. The
purpose of the function is to avoid division by zero.

A natural limitation for the mathematical functions is the limited accuracy and range,
which means that for example EXP can cause underflow or overflow at rather common
values of the argument. The trigonometric functions will get very low accuracy for
large arguments. These limitations are implementation dependent, and should be given
in the vendor's manual.

4. Character string functions:

The functions below perform operations from and to character strings. Please note
that ACHAR works with the standard ASCII character set while CHAR works with the
representation in the computer you are using.
ACHAR(I) Returns the ASCII character which has number I
ADJUSTL(STRING) Adjusts to the left
ADJUSTR(STRING) Adjusts to the right
CHAR(I, kind) Returns the character that has the number I
IACHAR(C) Returns the ASCII number of the character C
ICHAR(C) Returns the number of character C

INDEX(STRING, SUBSTRING, back) Returns the starting position for a

substring within a string. If BACK is true then you get the
last starting position, in the other case, the first one.

LEN_TRIM(STRING) Returns the length of the string without the possibly

trailing blanks.

LGE(STRING_A, STRING_B)
LGT(STRING-A, STRING_B)
LLE(STRING_A, STRING_B)
LLT(STRING_A, STRING_B)
The above routines compare two strings using sorting according to ASCII. If a string
is shorter than the other, blanks are added at the end of the short string. If a string
contains a character outside the ASCII character set, the result is implementation-
dependent.
REPEAT(STRING, NCOPIES) Concatenates a character string NCOPIES
times with itself.
SCAN(STRING, SET, back) Returns the position of the first occurrence
of any character in the string SET in the string
STRING. If BACK is true, you will get
the rightmost such character.
TRIM(STRING) Returns the character string STRING without
trailing blanks.
VERIFY(STRING, SET, back) Returns the position of the first character
in STRING which is not in SET. If BACK
is TRUE, you get the last one!
The result is zero if all characters are
included!

5. Character string function for request:

LEN(STRING) returns the length of a character string. There does not have to be
assigned a value to the variable STRING.

6. Kind functions:
KIND(X)
SELECTED_INT_KIND(R)
SELECTED_REAL_KIND(p, r)
The first returns the kind of the actual argument, which can be of the type INTEGER,
REAL, COMPLEX, LOGICAL or CHARACTER. The argument X does not have to be assigned
any value. The second returns an integer kind with the requested number of digits, and
the third returns the kind for floating-point numbers with numerical precision at
least P digits and one decimal exponent range between -R and +R. The
parameters P and R must be scalar integers. At least one of P and R must be given.

The result of SELECTED_INT_KIND is an integer from zero and upward, if the desired
kind is not available you will get -1. If several implemented types satisfy the
condition, the one with the least decimal range is used. If there still are several types
or kinds that satisfy the condition, the one with the smallest kind number will be used.

The result of SELECTED_REAL_KIND is also an integer from zero and upward; if the
desired kind is not available, then -1 is returned if the precision is not available, -2 if
the exponent range is not available and -3 if no one of the requirements are available.
If several implemented types satisfy the condition, the one with the least decimal
precision is returned, and if there are several of them, the one with the least kind
number is returned.

Examples are given in chapter 2 of the main text. Examples of kinds in a few different
implementations (NAG and Cray) are given in Appendix 6.

7. Logical function:
LOGICAL(L, kind) converts between different kinds of logical variables. Logical
variables can be implemented in various ways, for example with a physical
representation occupying one bit (not recommended), one byte, one word or perhaps
even one double word. This difference is important if COMMON and EQUIVALENCE with
logical variables have been misused in a program in the traditional way of Fortran 66
programming.

8. Numerical inquiry functions:

These functions work with a certain model of integer and floating-point arithmetics,
see ISO (1991), section 13.7.1. The functions return properties of numbers of the same
kind as the variable X, which can be real and in some cases integer. Functions that
return properties of the actual argument X are available in section 12 below, floating-
point manipulation functions.
DIGITS(X) The number of significant digits
EPSILON(X) The least positive number that added
to 1 returns a number that is greater than 1
HUGE(X) The largest positive number
MAXEXPONENT(X) The largest exponent
MINEXPONENT The smallest exponent
PRECISION(X) The decimal precision
RADIX(X) The base in the model
RANGE(X) The decimal exponent
TINY(X) The smallest positive number

9. Bit inquiry function:

BIT_SIZE(I) returns the number of bits according to the model of bit representation in
the standard ISO (1991), section 13.5.7. Normally we get the number of bits in a
(whole) word.

10. Bit manipulation functions:

The model for bit representation in the standard ISO (1991), section 13.5.7 is used.
BTEST(I, POS) .TRUE. if the position number POS of I is 1
IAND(I, J) logical addition of the bit characters in
 variables I and J

IBCLR(I, POS) puts a zero in the bit in position POS

IBITS(I, POS, LEN) uses LEN bits of the word I with
beginning in position POS, the additional bits
are set to zero. It requires that
POS + LEN <= BIT_SIZE(I)
IBSET(I, POS) puts the bit in position POS to 1
IEOR(I, J) performs logical exclusive OR
IOR(I, J) performs logical OR
ISHFT(I, SHIFT) performs logical shift (to the right if the number
of steps SHIFT < 0 and to the left if SHIFT > 0).
Positions that are vacated are set to zero.
ISHFTC(I, SHIFT, size) performs logical shift a number of steps
circularly to the right if SHIFT < 0,
circularly to the left if SHIFT > 0. If SIZE
is given, it is required that 0 < SIZE <=
BIT_SIZE(I). Shift is only done for the bits
that are in the SIZE rightmost positions, but
for all positions if SIZE is not given.
NOT(I) returns a logical complement

11. Transfer functions:

TRANSFER(SOURCE, MOULD, size) specifies that the physical representation of the first
argument SOURCE shall be treated as if it had type and parameters as the second
argument MOULD, but without converting it. The purpose is to give a possibility to move
a quantity of a certain type via a routine that does not have exactly that data type.

12. Floating-point manipulation functions:

These functions work in a certain model of integer and floating-point arithmetic, see
the standard ISO(1991), section 13.7.1. The functions return numbers related to the
actual variable X of the type REAL. Functions that return properties for the numbers of
the same kind as the variable X are under section 8 (Numerical inquiry functions).
EXPONENT(X) exponent of the number
FRACTION(X) the fractional part of the number
NEAREST(X, S) returns the next representable number in
the direction of the sign of S
RRSPACING(X) returns the inverted value of the distance
between the two nearest possible numbers
SCALE(X, I) multiplies X by the base to the power I
SET_EXPONENT(X, I) returns the number that has the fractional
part of X and the exponent I
SPACING(X) the distance between the two nearest
possible numbers

13. Vector- and matrix-multiplication functions:

DOT_PRODUCT(VECTOR_A, VECTOR_B) makes a scalar product of two vectors, which must
have the same length (same number of elements).
Please note that if VECTOR_A is of type COMPLEX the result
is SUM(CONJG(VECTOR_A)*VECTOR_B).

MATMUL(MATRIX_A, MATRIX_B) makes the matrix product of two matrices, which must
be consistent, i.e. have the dimensions like (M, K) and (K, N). Used in chapter 11 of
the main text.

14. Array functions:

ALL(MASK, dim) returns a logical value that indicates whether all relations
in MASK are .TRUE., along only the desired dimension if the second argument is given.

ANY(MASK, dim) returns a logical value that indicates whether any relation
in MASK is .TRUE., along only the desired dimension if the second argument is given.

COUNT(MASK, dim) returns a numerical value that is the number of relations

in MASK who are .TRUE., along only the desired dimension if the second argument is
given.

MAXVAL(ARRAY, dim, mask) returns the largest value in the array ARRAY, of those that
obey the relation in the third argument MASK if that one is given, along only the desired
dimension if the second argumentDIM is given.

MINVAL(ARRAY, dim, mask) returns the smallest value in the array ARRAY, of those that
obey the relation in the third argument MASK if that one is given, along only the desired
dimension if the second argumentDIM is given.

PRODUCT(ARRAY, dim, mask) returns the product of all the elements in the array ARRAY,
of those that obey the relation in the third argument MASK if that one is given, along
only the desired dimension if the second argument DIM is given.

SUM (ARRAY, dim, mask) returns the sum of all the elements in the array ARRAY, of
those that obey the relation in the third argument MASK if that one is given, along only
the desired dimension if the second argument DIM is given. An example is given
in Appendix 3, section 10.

15. Array inquiry functions:

See also Appendix 3, section 10.
ALLOCATED(ARRAY) is a logical function which indicates if the array is allocated.

LBOUND(ARRAY, dim) is a function which returns the lower dimension limit for
the ARRAY. If DIM (the dimension) is not given as an argument, you get an integer
vector, if DIM is included, you get the integer value with exactly that lower dimension
limit, for which you asked.

SHAPE(SOURCE) is a function which returns the shape of an array SOURCE as an integer

vector.

SIZE(ARRAY, dim) is a function which returns the number of elements in an

array ARRAY, if DIM is not given, and the number of elements in the relevant dimension
if DIM is included.

UBOUND(ARRAY, dim) is a function similar to LBOUND which returns the upper

dimensional limits.

16. Array construct functions:

MERGE(TSOURCE, FSOURCE, MASK) is a function which joins two arrays. It gives the
elements in TSOURCE if the condition in MASK is .TRUE. and FSOURCE if the condition
in MASK is .FALSE. The two fields TSOURCEand FSOURCE have to be of the same type and
the same shape. The result is also of this type and this shape. Also MASK must have the
same shape.

I here give a rather complete example of the use of MERGE which also
uses RESHAPE from the next section in order to build suitable test matrices.

Note that the two subroutines WRITE_ARRAY and WRITE_L_ARRAY are test routines to
write matrices which in the first case are of a REAL type, in the second case of
a LOGICAL type.
IMPLICIT NONE
INTERFACE
SUBROUTINE WRITE_ARRAY (A)
REAL :: A(:,:)
END SUBROUTINE WRITE_ARRAY
SUBROUTINE WRITE_L_ARRAY (A)
LOGICAL :: A(:,:)
END SUBROUTINE WRITE_L_ARRAY
END INTERFACE

REAL, DIMENSION(2,3) :: TSOURCE, FSOURCE, RESULT

LOGICAL, DIMENSION(2,3) :: MASK
TSOURCE = RESHAPE( (/ 11, 21, 12, 22, 13, 23 /), &
 (/ 2, 3 /) )
FSOURCE = RESHAPE( (/ -11, -21, -12, -22, -13, -23 /), &
(/ 2,3 /) )
MASK = RESHAPE( (/ .TRUE., .FALSE., .FALSE., .TRUE., &
.FALSE., .FALSE. /), (/ 2,3 /) )

RESULT = MERGE(TSOURCE, FSOURCE, MASK)

CALL WRITE_ARRAY(TSOURCE)
CALL WRITE_ARRAY(FSOURCE)
CALL WRITE_L_ARRAY(MASK)
CALL WRITE_ARRAY(RESULT)
END

SUBROUTINE WRITE_ARRAY (A)

REAL :: A(:,:)
DO I = LBOUND(A,1), UBOUND(A,1)
WRITE(*,*) (A(I, J), J = LBOUND(A,2), UBOUND(A,2) )
END DO
RETURN
END SUBROUTINE WRITE_ARRAY

SUBROUTINE WRITE_L_ARRAY (A)

LOGICAL :: A(:,:)
DO I = LBOUND(A,1), UBOUND(A,1)
WRITE(*,"(8L12)") (A(I, J), J= LBOUND(A,2), UBOUND(A,2))
END DO
RETURN
END SUBROUTINE WRITE_L_ARRAY
The following output is obtained
11.0000000 12.0000000 13.0000000
21.0000000 22.0000000 23.0000000

-11.0000000 -12.0000000 -13.0000000

-21.0000000 -22.0000000 -23.0000000

T F F
F T F

11.0000000 -12.0000000 -13.0000000

-21.0000000 22.0000000 -23.0000000
PACK(ARRAY, MASK, vector) packs an array to a vector with the control of MASK. The
shape of the logical array MASK has to agree with the one for ARRAY or MASK must be a
scalar. If VECTOR is included, it has to be an array of rank 1 (i.e. a vector) with at least
as many elements as those that are true in MASK and have the same type as ARRAY.
If MASK is a scalar with the value .TRUE. then VECTOR instead must have the same
number of elements as ARRAY.

The result is a vector with as many elements as those in ARRAY that obey the
conditions if VECTOR is not included (i.e. all elements if MASK is a scalar with
value .TRUE.). In the other case the number of elements of the result will be as many
as in VECTOR. The values will be the approved ones, i.e. the values which fulfill the
condition, and will be in the ordinary Fortran order. If VECTOR is included and the
number of its elements exceeds the number of approved values, the lacking values
required for the result are taken from the corresponding locations in VECTOR.

The following example is based on the modification of the one for MERGE , but I give
now only the results.
ARRAY
11.0000000 12.0000000 13.0000000
21.0000000 22.0000000 23.0000000

VECTOR
-11.0000000
-21.0000000
-12.0000000
-22.0000000
-13.0000000
-23.0000000

MASK
T F F
F T F

PACK(ARRAY, MASK)
11.0000000
22.0000000

PACK(ARRAY, MASK, VECTOR)

11.0000000
22.0000000
-12.0000000
-22.0000000
-13.0000000
-23.0000000
SPREAD(SOURCE, DIM, NCOPIES) returns an array of the same type as the
argument SOURCE with the rank increased by one. The parameters DIM and NCOPIES are
integer. If NCOPIES is negative the value zero is used instead. If SOURCE is a scalar,
then SPREAD becomes a vector with NCOPIES elements that all have the same value
as SOURCE. The parameter DIM indicates which index is to be extended. It has to be
within the range 1 and 1+(rank of SOURCE), if SOURCE is a scalar then DIM has to be
one. The parameter NCOPIES is the number of elements in the new dimensions.
Additional discussion is given in the solution to exercise (11.1).

UNPACK(VECTOR, MASK, ARRAY) scatters a vector to an array under control of MASK. The
shape of the logical array MASK has to agree with the one for ARRAY. The
array VECTOR has to have the rank 1 (i.e. it is a vector) with at least as many elements
as those that are true in MASK, and also has to have the same type as ARRAY. If ARRAY is
given as a scalar then it is considered to be an array with the same shape as MASKand
the same scalar elements everywhere.
The result will be an array with the same shape as MASK and the same type as VECTOR.
The values will be those from VECTOR that are accepted (i.e. those fulfilling the
condition in MASK), taken in the ordinary Fortran order, while in the remaining
positions in ARRAY the old values are kept.

17. ARRAY reshape function.

RESHAPE(SOURCE, SHAPE, pad, order) constructs an array with a specified
shape SHAPE starting from the elements in a given array SOURCE. If PAD is not included
then the size of SOURCE has to be at leastPRODUCT (SHAPE). If PAD is included it has to
have the same type as SOURCE. If ORDER is included, it has to be an INTEGER array with
the same shape as SHAPE and the values must be a permutation of (1,2,3,...,N), where
N is the number of elements in SHAPE , it has to be less than, or equal to 7.

The result has of course a shape SHAPE and the elements are those in SOURCE, possibly
complemented with PAD. The different dimensions have been permuted at the
assignment of the elements if ORDER was included, but without influencing the shape
of the result.

A few simple examples are given in the previous and the next section and also
in Appendix 3, section 9. A more complicated example, illustrating also the optional
arguments, follows.
! PROGRAM TO TEST THE OPTIONAL ARGUMENTS TO RESHAPE
INTERFACE
SUBROUTINE WRITE_MATRIX(A)
REAL, DIMENSION(:,:) :: A
END SUBROUTINE WRITE_MATRIX
END INTERFACE

REAL, DIMENSION (1:9) :: B = (/ 11, 12, 13, 14, 15, 16, 17, 18, 19 /)
REAL, DIMENSION (1:3, 1:3) :: C, D, E
REAL, DIMENSION (1:4, 1:4) :: F, G, H

INTEGER, DIMENSION (1:2) :: ORDER1 = (/ 1, 2 /)

INTEGER, DIMENSION (1:2) :: ORDER2 = (/ 2, 1 /)
REAL, DIMENSION (1:16) :: PAD1 = (/ -1, -2, -3, -4, -5, -6, -7, -8, &
& -9, -10, -11, -12, -13, -14, -15, -16 /)

C = RESHAPE( B, (/ 3, 3 /) )
CALL WRITE_MATRIX(C)

D = RESHAPE( B, (/ 3, 3 /), ORDER = ORDER1)

CALL WRITE_MATRIX(D)

E = RESHAPE( B, (/ 3, 3 /), ORDER = ORDER2)

CALL WRITE_MATRIX(E)
 F = RESHAPE( B, (/ 4, 4 /), PAD = PAD1)
CALL WRITE_MATRIX(F)

G = RESHAPE( B, (/ 4, 4 /), PAD = PAD1, ORDER = ORDER1)

CALL WRITE_MATRIX(G)

H = RESHAPE( B, (/ 4, 4 /), PAD = PAD1, ORDER = ORDER2)

CALL WRITE_MATRIX(H)

END

SUBROUTINE WRITE_MATRIX(A)
REAL, DIMENSION(:,:) :: A
WRITE(*,*)
DO I = LBOUND(A,1), UBOUND(A,1)
WRITE(*,*) (A(I,J), J = LBOUND(A,2), UBOUND(A,2))
END DO
END SUBROUTINE WRITE_MATRIX
The output from the above program is as follows.
11.0000000 14.0000000 17.0000000
12.0000000 15.0000000 18.0000000
13.0000000 16.0000000 19.0000000

11.0000000 14.0000000 17.0000000

12.0000000 15.0000000 18.0000000
13.0000000 16.0000000 19.0000000

11.0000000 12.0000000 13.0000000

14.0000000 15.0000000 16.0000000
17.0000000 18.0000000 19.0000000

11.0000000 15.0000000 19.0000000 -4.0000000

12.0000000 16.0000000 -1.0000000 -5.0000000
13.0000000 17.0000000 -2.0000000 -6.0000000
14.0000000 18.0000000 -3.0000000 -7.0000000

11.0000000 15.0000000 19.0000000 -4.0000000

12.0000000 16.0000000 -1.0000000 -5.0000000
13.0000000 17.0000000 -2.0000000 -6.0000000
14.0000000 18.0000000 -3.0000000 -7.0000000

11.0000000 12.0000000 13.0000000 14.0000000

15.0000000 16.0000000 17.0000000 18.0000000
19.0000000 -1.0000000 -2.0000000 -3.0000000
-4.0000000 -5.0000000 -6.0000000 -7.0000000

18. ARRAY manipulation functions.

The shift functions return the shape of an array unchanged, but move the elements.
They are rather difficult to explain so I recommend to study also the standard ISO
(1991).
CSHIFT(ARRAY, SHIFT, dim) performs circular shift by SHIFT positions to the left
if SHIFT is positive and to the right if it is negative. If ARRAY is a vector the shift is
being done in a natural way, if it is an array of a higher rank then the shift is in all
sections along the dimension DIM. If DIM is missing it is considered to be 1, in other
cases it has to be a scalar integer number between 1 and n (where n equals the rank
ofARRAY ). The argument SHIFT is a scalar integer or an integer array of rank n-1 and
the same shape as the ARRAY, except along the dimension DIM (which is removed
because of the lower rank). Different sections can therefore be shifted in various
directions and with various numbers of positions.

EOSHIFT(ARRAY, SHIFT, boundary, dim) performs shift to the left if SHIFT is positive
and to the right if it is negative. Instead of the elements shifted out new elements are
taken from BOUNDARY. If ARRAY is a vector the shift is being done in a natural way, if it
is an array of a higher rank, the shift on all sections is along the dimension DIM.
If DIM is missing, it is considered to be 1, in other cases it has to have a scalar integer
value between 1 and n (where n equals the rank of ARRAY). The argument SHIFT is a
scalar integer if ARRAY has rank 1, in the other case it can be a scalar integer or an
integer array of rank n-1 and with the same shape as the array ARRAY except along the
dimension DIM (which is removed because of the lower rank).

The corresponding applies to BOUNDARY which has to have the same type as the ARRAY.
If the parameter BOUNDARY is missing you have the choice of values zero, .FALSE. or
blank being used, depending on the data type. Different sections can thus be shifted in
various directions and with various numbers of positions. A simple example of the
above two functions for the vector case follows, both the program and the output.
REAL, DIMENSION(1:6) :: A = (/ 11.0, 12.0, 13.0, 14.0, &
15.0, 16.0 /)
REAL, DIMENSION(1:6) :: X, Y
WRITE(*,10) A
X = CSHIFT ( A, SHIFT = 2)
WRITE(*,10) X
Y = CSHIFT (A, SHIFT = -2)
WRITE(*,10) Y
X = EOSHIFT ( A, SHIFT = 2)
WRITE(*,10) X
Y = EOSHIFT ( A, SHIFT = -2)
WRITE(*,10) Y
10 FORMAT(1X,6F6.1)
END

11.0 12.0 13.0 14.0 15.0 16.0

13.0 14.0 15.0 16.0 11.0 12.0
15.0 16.0 11.0 12.0 13.0 14.0
13.0 14.0 15.0 16.0 0.0 0.0
0.0 0.0 11.0 12.0 13.0 14.0
A simple example of the above two functions in the matrix case follows. I have here
used RESHAPE in order to create a suitable matrix to start work with. The program is
not reproduced here, only the main statements.
B = (/ 11.0, 12.0, 13.0, 14.0, 15.0, 16.0 /)

11.0 12.0 13.0 Z = RESHAPE( B, (/3,3/) )

14.0 15.0 16.0
17.0 18.0 19.0

17.0 18.0 19.0 X = CSHIFT (Z, SHIFT = 2)

11.0 12.0 13.0
14.0 15.0 16.0

13.0 11.0 12.0 X = CSHIFT ( Z, SHIFT = 2, DIM = 2)

16.0 14.0 15.0
19.0 17.0 18.0

14.0 15.0 16.0 X = CSHIFT (Z, SHIFT = -2)

17.0 18.0 19.0
11.0 12.0 13.0

17.0 18.0 19.0 X = EOSHIFT ( Z, SHIFT = 2)

0.0 0.0 0.0
0.0 0.0 0.0

13.0 0.0 0.0 X = EOSHIFT ( Z, SHIFT = 2, DIM = 2)

16.0 0.0 0.0
19.0 0.0 0.0

0.0 0.0 0.0 X = EOSHIFT ( Z, SHIFT = -2)

0.0 0.0 0.0
11.0 12.0 13.0
TRANSPOSE (MATRIX) transposes a matrix, which is an array of rank 2. It replaces the
rows and columns in the matrix.

19. Array location functions:

MAXLOC(ARRAY, mask) returns the position of the greatest element in the array ARRAY,
if MASK is included only for those which fulfill the conditions in MASK. The result is an
integer vector! It is used in the solution of exercise (11.1).

MINLOC(ARRAY, mask) returns the position of the smallest element in the array ARRAY ,
if MASK is included only for those which fulfill the conditions in MASK. The result is an
integer vector!

20. Pointer inquiry functions:

ASSOCIATED(POINTER, target) is logical function that indicates if the
pointer POINTER is associated with some target, and if a specific TARGET is included it
indicates if it is associated with exactly that target. If both POINTER and TARGET are
pointers, the result is .TRUE. only if both are associated with the same target. I refer
the reader to chapter 12 of the main text, Pointers.

21. Intrinsic subroutines:

Time routines:

DATE_AND_TIME(date, time, zone, values)

A subroutine which returns the date, the time and the time zone. At least one
argument has to be given.

DATE must be a scalar character string variable with at least 8 characters and it
is assigned the value CCYYMMDD for century, year, month and day. All are given
numerically, with blanks if the system does not include the date.

TIME must also be a scalar character string variable with at least 10 characters
and it is assigned a value hhmmss.sss for time in hours, minutes, seconds and
milliseconds. All are given numerically with blanks if the system does not
include a clock.

ZONE must be a scalar character string variable with at least 5 characters and it
is assigned the value +hhmm for sign, time in hours and minutes for the local
time difference with UTC (which was previously called Greenwich Mean
Time). All are given numerically, with blanks if the system does not include a
clock. In Sweden we therefore get +0100 in winter and +0200 in summer, in
Novosibirsk we get +0700 .

The variable VALUES is instead an integer vector with at least 8 elements, it

gives the easiest way of using the results from DATE_AND_TIME at the
calculations in a program. If the system does not include the date or the time
you get the value -HUGE(0), that is the smallest integer number in the model, as
output. The vector will include the following elements: year, month, day, time
difference in minutes, hours, minutes, seconds and milliseconds.
SYSTEM_CLOCK(COUNT, COUNT_RATE, COUNT_MAX)

Subroutine which returns the system time. At least one argument has to be
given. COUNT is a scalar integer which is increased by one for each cycle up
to COUNT_MAX , where it starts once again. If there is no system clock then -
HUGE(0) is returned.
 COUNT_RATE is a scalar integer that gives the number of cycles per second. If
there is no system clock the value zero is returned.

COUNT_MAX is a scalar integer which gives the maximum value that COUNT can
reach. If there is no system clock, zero is returned instead.

Bit copy routine:

MVBITS(FROM, FROMPOS, LEN, TO, TOPOS)

A subroutine which copies the sequence of bits in position FROMPOS and has the
length LEN to target TO starting in position TOPOS. The remaining bits are not
changed. All quantities have to be integers and all except TO have to
have INTENT(IN) while TO is supposed to have INTENT(INOUT) and be of the
same kind type as FROM. The same variable can be both FROM and TO. Some
natural restrictions apply to the values of LEN, FROMPOS and TOPOS and you also
have to consider the value of BIT_SIZE.

Random number routines:

A sequence of pseudo random numbers can be generated from a starting value

which is stored as an integer vector. The subroutines offer a portable interface
towards an implementation dependent random number sequence.
RANDOM_NUMBER(HARVEST)

This subroutine returns in the floating-point number variable HARVEST one (or
several if HARVEST is an array) random numbers between zero and 1.
RANDOM_SEED(size, put, get)

This subroutine resets, or gives information about, the random number

generator. No arguments have to be provided. The output variable SIZE must be
a scalar integer and gives the number of integers (N) the processor uses for the
starting value. The input variable PUT is an integer vector which puts the
starting numbers provided by the user into the random number generator. The
output variable GET(also an integer vector)reads the present starting value.
Example:
CALL RANDOM_SEED ! Initializing
CALL RANDOM SEED (SIZE=K) ! Sets K = N
CALL RANDOM_SEED (PUT = SEED (1:K)) ! Uses the starting value
! given by the user
CALL RANDOM_SEED (GET = OLD(1:K)) ! Returns the present
 ! starting value

A simple example on the use of these functions is now available.

Last modified: August 10, 2009

[email protected]