APPENDIX 5

Intrinsic functions in Fortran 90

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).

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 argument DIM 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 argument DIM 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 TSOURCE and 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 MASK and 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 least PRODUCT (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 of ARRAY ). 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:


Last modified: August 10, 2009
boein@nsc.liu.se