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
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 -3The 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 CTruncation 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 DThe 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).
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_ARRAYThe 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 -23.0000000PACK(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 -23.0000000SPREAD(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_MATRIXThe 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 14.0A 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 13.0TRANSPOSE (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
boein@nsc.liu.se
No comments:
Post a Comment