Chapter 3 FORTRAN 77 and VMS Intrinsic Functions

This chapter lists the set of FORTRAN 77 intrinsic functions accepted by f95 and is provided to aid migration of programs from legacy FORTRAN 77 to Fortran 95.

f95 recognizes as intrinsic functions all the FORTRAN 77 and VMS functions listed in this chapter, along with all the Fortran 95 functions listed in the previous chapter. As an aid to migrating legacy FORTRAN 77 programs to f95, compiling with -f77=intrinsics causes the compiler to recognize only FORTRAN 77 and VMS functions as intrinsics, and not the Fortran 95 intrinsics.

Intrinsic functions that are Sun extensions of the ANSI FORTRAN 77 standard are marked with ¤. Programs using non-standard intrinsics and library functions may not be portable to other platforms.

Intrinsic functions have generic and specific names when they accept arguments of more than one data type. In general, the generic name returns a value with the same data type as its argument. However, there are exceptions such as the type conversion functions (Table 3–2) and the inquiry functions (Table 3–7). The function may also be called by one of its specific names to handle a specific argument data type.

With functions that work on more than one data item (for example, sign(a1,a2) ), all the data arguments must be the same type.

In the following tables, the FORTRAN 77 intrinsic functions are listed by:

• Intrinsic Function– description of what the function does

• Definition– a mathematical definition

• No. of Args.– number of arguments the function accepts

• Generic Name– the function’s generic name

• Specific Names– the function’s specific names

• Argument Type– data type associated with each specific name

• Function Type– data type returned for specific argument data type

  ---

  Note –

  Compiler option -xtypemap changes the default sizes of variables and has an affect on intrinsic references. See 3.4 Remarks , and the discussion of default sizes and alignment in the Fortran User’s Guide.

  ---

3.1 Arithmetic and Mathematical Functions

This section details arithmetic, type conversion, trigonometric, and other functions. “a” stands for a function’s single argument, “a1” and “a2” for the first and second arguments of a two argument function, and “ar” and “ai” for the real and imaginary parts of a function’s complex argument.

3.1.1 Arithmetic Functions

3.1.2 Type Conversion Functions

  See Note (1).

  See Note (2).

   See Note (3).

REAL*16
   See Note (3’).

COMPLEX
  See Notes (4)   and (8).

CMPLX

DOUBLE COMPLEX
   See Note (8).

DCMPLX@

COMPLEX*32
   See Note (8).

QCMPLX@

INTEGER
   See Note (5).

-
-

CHARACTER
   See Note (5).

-
-

On an ASCII platform, including Sun systems:

• ACHAR is a nonstandard synonym for CHAR

• IACHAR is a nonstandard synonym for ICHAR

On a non-ASCII platform, ACHAR and IACHAR were intended to provide a way to deal directly with ASCII.

3.1.3 Trigonometric Functions

3.1.4 Other Mathematical Functions

• The error function: 2/sqrt(pi) x integral from 0 to a of exp(-t*t) dt

3.2 Character Functions

On an ASCII platform (including Sun systems):

• ACHAR is a nonstandard synonym for CHAR

• IACHAR is a nonstandard synonym for ICHAR

On a non-ASCII platform, ACHAR and IACHAR were intended to provide a way to deal directly with ASCII.

3.3 Miscellaneous Functions

Other miscellaneous functions include bitwise functions, environmental inquiry functions, and memory allocation and deallocation functions.

3.3.1 Bit Manipulation ¤

None of these functions are part of the FORTRAN 77 Standard.

The above functions are available as intrinsic or extrinsic functions. See also the discussion of the library bit manipulation routines in the Fortran Library Reference manual.

3.3.2 Environmental Inquiry Functions ¤

None of these functions are part of the FORTRAN 77 Standard.

3.3.3 Memory ¤

None of these functions are part of the FORTRAN 77 Standard.

3.4 Remarks

The following remarks apply to all of the intrinsic function tables in this chapter.

• The abbreviation DOUBLE stands for DOUBLE PRECISION.

• An intrinsic that takes INTEGER arguments accepts INTEGER*2, INTEGER*4, or INTEGER*8.

• INTEGER intrinsics that take INTEGER arguments return values of INTEGER type determined as follows. Note that the -xtypemap option may alter the default sizes of actual arguments:

  • mod sign dim max min and iand or ior xor ieor— size of the value returned is the largest of the sizes of the arguments.

  • abs ishft lshift rshift lrshft ibset ibclr ishftc ibits— size of the value returned is the size of the first argument.

  • int epbase epprec— size of the value returned is the size of default INTEGER.

  • ephuge— size of the value returned is the size of the default INTEGER, or the size of the argument, whichever is largest.

  Options that change the default data sizes also change the way some intrinsics are used. For example, with -dbl in effect, a call to ZCOS with a DOUBLE COMPLEX argument will automatically become a call to CQCOS because the argument has been promoted to COMPLEX*32. The following functions have this capability:

  aimag alog amod cabs ccbrt ccos cdabs cdcbrt cdcos cdexp cdlog cdsin cdsqrt cexp clog csin csqrt dabs dacos dacosd dasin dasind datan datand dcbrt dconjg dcos dcosd dcosh ddim derf derfc dexp dimag dint dlog dmod dnint dprod dsign dsin dsind dsinh dsqrt dtan dtand dtanh idnint iidnnt jidnnt zabs zcbrt zcos zexp zlog zsin zsqrt

• The following functions permit arguments of an integer or logical type of any size:

  and iand ieor iiand iieor iior inot ior jiand jieor jior jnot lrshft lshift not or rshift xor

• An intrinsic that is shown to return a default REAL, DOUBLE PRECISION, COMPLEX, or DOUBLE COMPLEX value will return the prevailing type depending on certain compilation options. For example, if compiled with- xtypemap=real:64,double:64:

  • A call to a REAL function returns REAL*8

  • A call to a DOUBLE PRECISION function returns REAL*8

  • A call to a COMPLEX function returns COMPLEX*16

  • A call to a DOUBLE COMPLEX function returns COMPLEX*16

    Other options that alter the data sizes of default data types are– r8 and– dbl, which also promote DOUBLE to QUAD. The– xtypemap= option provides more flexibility than these older compiler options and is preferred.

  A function with a generic name returns a value with the same type as the argument—except for type conversion functions, the nearest integer function, the absolute value of a complex argument, and others. If there is more than one argument, they must all be of the same type.

• If a function name is used as an actual argument, then it must be a specific name.

• If a function name is used as a dummy argument, then it does not identify an intrinsic function in the subprogram, and it has a data type according to the same rules as for variables and arrays.

3.4.1 Notes on Functions

Tables and notes 1 through 12 are based on the “Table of Intrinsic Functions,” from ANSI X3.9-1978 Programming Language FORTRAN, with the Fortran extensions added.

(1) INT

If A is type integer, then INT(A) is A.

If A is type real or double precision, then:

if |A| < 1, then INT(A) is 0if |A| ≥ 1, then INT(A) is the greatest integer that does not exceed the magnitude of A, and whose sign is the same as the sign of A. (Such a mathematical integer value may be too large to fit in the computer integer type.)

If A is type complex or double complex, then apply the above rule to the real part of A.

If A is type real, then IFIX(A) is the same as INT(A).

(2) REAL

If A is type real, then REAL(A) is A.

If A is type integer or double precision, then REAL(A) is as much precision of the significant part of A as a real datum can contain.

If A is type complex, then REAL(A) is the real part of A.

If A is type double complex, then REAL(A) is as much precision of the significant part of the real part of A as a real datum can contain.

(3) DBLE

If A is type double precision, then DBLE(A) is A.

If A is type integer or real, then DBLE(A) is as much precision of the significant part of A as a double precision datum can contain.

If A is type complex, then DBLE(A) is as much precision of the significant part of the real part of A as a double precision datum can contain.

If A is type COMPLEX*16, then DBLE(A) is the real part of A.

(3’) QREAL

If A is type REAL*16, then QREAL(A) is A.

If A is type integer, real, or double precision, then QREAL(A) is as much precision of the significant part of A as a REAL*16 datum can contain.

If A is type complex or double complex, then QREAL(A) is as much precision of the significant part of the real part of A as a REAL*16 datum can contain.

If A is type COMPLEX*16 or COMPLEX*32, then QREAL(A) is the real part of A.

(4) CMPLX

If A is type complex, then CMPLX(A) is A.

If A is type integer, real, or double precision, then CMPLX(A) is REAL(A) + 0i.

If A1 and A2 are type integer, real, or double precision, then CMPLX(A1,A2) is REAL(A1) + REAL(A2)*i.

If A is type double complex, then CMPLX(A) is REAL( DBLE(A) ) + i*REAL( DIMAG(A) ).

If CMPLX has two arguments, then they must be of the same type, and they may be one of integer, real, or double precision.

If CMPLX has one argument, then it may be one of integer, real, double precision, complex, COMPLEX*16, or COMPLEX*32.

(4’) DCMPLX

If A is type COMPLEX*16, then DCMPLX(A) is A.

If A is type integer, real, or double precision, then DCMPLX(A) is DBLE(A) + 0i.

If A1 and A2 are type integer, real, or double precision, then DCMPLX(A1,A2) is DBLE(A1) + DBLE(A2)*i.

If DCMPLX has two arguments, then they must be of the same type, and they may be one of integer, real, or double precision.

If DCMPLX has one argument, then it may be one of integer, real, double precision, complex, COMPLEX*16, or COMPLEX*32.

(5) ICHAR

ICHAR(A) is the position of A in the collating sequence.

The first position is 0, the last is N-1, 0≤ICHAR(A)≤N-1, where N is the number of characters in the collating sequence, and A is of type character of length one.

CHAR and ICHAR are inverses in the following sense:

• ICHAR(CHAR(I)) = I, for 0≤I≤N-1

  • CHAR(ICHAR(C)) = C, for any character C capable of representation in the processor

    (6) COMPLEX

    A COMPLEX value is expressed as an ordered pair of reals, (ar, ai), where ar is the real part, and ai is the imaginary part.

    (7) Radians

    All angles are expressed in radians, unless the “Intrinsic Function” column includes the “(degrees)” remark.

    (8) COMPLEX Function

    The result of a function of type COMPLEX is the principal value.

    (8’) CBRT

    If a is of COMPLEX type, CBRT results in COMPLEX RT1=(A, B), where:A ≥ 0.0, and -60 degrees≤ arctan (B/A) < + 60 degrees.

    Other two possible results can be evaluated as follows:

  • RT2 = RT1 * (-0.5, square_root (0.75))

  • RT3 = RT1 * (-0.5, square_root (0.75))

    (9) Argument types

    All arguments in an intrinsic function reference must be of the same type.

    (10) INDEX

    INDEX(X,Y) is the place in X where Y starts. That is, it is the starting position within character string X of the first occurrence of character string Y.

    If Y does not occur in X, then INDEX(X,Y) is 0.

    If LEN(X) < LEN(Y), then INDEX(X,Y) is 0.

    INDEX returns default INTEGER*4 data. If compiling for a 64-bit environment, the compiler will issue a warning if the result overflows the INTEGER*4 data range. To use INDEX in a 64-bit environment with character strings larger than the INTEGER*4 limit (2 Gbytes), the INDEX function and the variables receiving the result must be declared INTEGER*8.

    (11) LEN

    LEN returns the declared length of the CHARACTER argument variable. The actual value of the argument is of no importance.

    LEN returns default INTEGER*4 data. If compiling for a 64-bit environment, the compiler will issue a warning if the result overflows the INTEGER*4 data range. To use LEN in a 64-bit environment with character variables larger than the INTEGER*4 limit (2 Gbytes), the LEN function and the variables receiving the result must be declared INTEGER*8.

    (12) Lexical Compare

    LGE( X, Y ) is true if X=Y, or if X follows Y in the collating sequence; otherwise, it is false.

    LGT( X, Y ) is true if X follows Y in the collating sequence; otherwise, it is false.

    LLE( X, Y ) is true if X=Y, or if X precedes Y in the collating sequence; otherwise, it is false.

    LLT( X, Y ) is true if X precedes Y in the collating sequence; otherwise, it is false.

    If the operands for LGE, LGT, LLE, and LLT are of unequal length, the shorter operand is considered as if it were extended on the right with blanks.

    (13) Bit Functions

    There are other bitwise operations in VMS Fortran, but these are not implemented.

    (14) Shift

    LSHIFT shifts a1 logically left by a2 bits (inline code).

    LRSHFT shifts a1 logically right by a2 bits (inline code).

    RSHIFT shifts a1 arithmetically right by a2 bits.

    ISHFT shifts a1 logically left if a2 > 0 and right if a2 < 0.

    The LSHIFT and RSHIFT functions are the Fortran analogs of the C << and >> operators. As in C, the semantics depend on the hardware.

    The behavior of the shift functions with an out of range shift count is hardware dependent and generally unpredictable. In this release, shift counts larger than 31 result in hardware dependent behavior.

    (15) Environmental inquiries

    Only the type of the argument is significant.

    (16) Epsilon

    Epsilon is the least e, such that 1.0 + e ≠ 1.0.

    (17) LOC, MALLOC, and FREE

    The LOC function returns the address of a variable or of an external procedure. The function call MALLOC( n ) allocates a block of at least n bytes, and returns the address of that block.

    LOC returns default INTEGER*4 in 32-bit environments, INTEGER*8 in 64-bit environments.

    MALLOC is a library function and not an intrinsic in FORTRAN 77. It too returns default INTEGER*4 in 32-bit environments, INTEGER*8 in 64-bit environments. However, MALLOC must be explicitly declared INTEGER*8 when compiling for 64-bit environments.

    The value returned by LOC or MALLOC should be stored in variables typed POINTER, INTEGER*4, or INTEGER*8 in 64-bit environments. The argument to FREE must be the value returned by a previous call to MALLOC and hence should have data type POINTER, INTEGER*4, or INTEGER*8.

    MALLOC64 always takes an INTEGER*8 argument (size of memory request in bytes) and always returns an INTEGER*8 value. Use this routine rather than MALLOC when compiling programs that must run in both 32-bit and 64-bit environments. The receiving variable must be declared either POINTER or INTEGER*8.

    (18) SIZEOF

    The SIZEOF intrinsic cannot be applied to arrays of an assumed size, characters of a length that is passed, or subroutine calls or names. SIZEOF returns default INTEGER*4 data. If compiling for a 64-bit environment, the compiler will issue a warning if the result overflows the INTEGER*4 data range. To use SIZEOF in a 64-bit environment with arrays larger than the INTEGER*4 limit (2 Gbytes), the SIZEOF function and the variables receiving the result must be declared INTEGER*8.

3.5 VMS Intrinsic Functions

This section lists VMS FORTRAN intrinsic routines recognized by f95. They are, of course, nonstandard. ¤

3.5.1 VMS Double-Precision Complex

3.5.2 VMS Degree-Based Trigonometric

3.5.3 VMS Bit-Manipulation

3.5.4 VMS Multiple Integer Types

The possibility of multiple integer types is not addressed by the Fortran Standard. The compiler copes with their existence by treating a specific INTEGER-to-INTEGER function name (IABS, and so forth) as a special sort of generic. The argument type is used to select the appropriate runtime routine name, which is not accessible to the programmer.

VMS Fortran takes a similar approach, but makes the specific names available.