Module_Num_Constants¶

This file is part of statpack.

statpack is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

statpack is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You can find a copy of the GNU Lesser General Public License in the statpack/doc directory.

THIS MODULE PROVIDES SIMPLE NAMES AND ROUTINES FOR THE VARIOUS MACHINE DEPENDENT CONSTANTS. ALL ARE FOR PRECISION ‘stnd’.

LATEST REVISION : 28/09/2018

`function lamch ( cmach )`¶

Purpose¶

LAMCH determines machine parameters for precision STND.

Arguments¶

CMACH (INPUT) character*1

Specifies the value to be returned by LAMCH. If:

CMACH = ‘S’ or ‘s’, LAMCH := sfmin

CMACH = ‘T’ or ‘t’, LAMCH := t

CMACH = ‘R’ or ‘r’, LAMCH := rnd

CMACH = ‘G’ or ‘g’, LAMCH := grd

CMACH = ‘U’ or ‘u’, LAMCH := unitrnd

CMACH = ‘P’ or ‘p’, LAMCH := prec

where:

sfmin = safe minimum, such that 1/sfmin does not overflow.

t = number of (base) digits in the floating-point significand.

rnd = 0.0 when floating-point addition rounds upward, downward

or toward zero;

= 1.0 when floating-point addition rounds to nearest,

but not in the IEEE style;

= 2.0 when floating-point addition rounds in the IEEE style.

grd = 1. if floating-point arithmetic chops (rnd = 0.) and more

than t digits participate in the post-normalization shift of the floating-point significand in multiplication,

= 0.0 otherwise.

unitrnd = unit roundoff of the machine.

prec = unitrnd*machbase.

Further Details¶

The routine is based on the routine DLAMCH in LAPACK77 (version 3).

For any other characters, LAMCH returns the bit pattern corresponding to a quiet NaN.

`subroutine mach ( basedigits, irnd, iuflow, igrd, iexp, ifloat, expepspos, expepsneg, minexpbase, maxexpbase, epspos, epsneg, epsilpos, epsilneg, rndunit )`¶

Purpose¶

This subroutine is intended to determine the parameters of the floating-point arithmetic system specified below.

Arguments¶

BASEDIGITS (OUTPUT, OPTIONAL) integer(i4b)

The number of base digits in the floating-point significand.

IRND (OUTPUT, OPTIONAL) integer(i4b)

A parameter indicating whether proper rounding or chopping (rounding upward, downward, toward zero) occurs in addition. If:

IRND = 0 if floating-point addition rounds upward, downward or toward zero;

IRND = 1 if floating-point addition rounds to nearest, but not in the IEEE style;

IRND = 2 if floating-point addition rounds in the IEEE style.

IUFLOW (OUTPUT, OPTIONAL) integer(i4b)

A parameter indicating whether underflow is full or partial:

IUFLOW = 0 if there is full underflow (flush to zero, etc);

IUFLOW = 1 if there is partial underflow.

IGRD (OUTPUT, OPTIONAL) integer(i4b)

The number of guard digits for multiplication with chopping arithmetic (IRND = 0). If:

IGRD = 0 if floating-point arithmetic rounds, or if it chops and only BASEDIGITS digits participate in the post-normalization shift of the floating-point significand in multiplication;

IGRD = 1 if floating-point arithmetic chops and more than BASEDIGITS digits participate in the post-normalization shift of the floating-point significand in multiplication.

IEXP (OUTPUT, OPTIONAL) integer(i4b)

A guess for the number of bits dedicated to the representation of the exponent of a floating point number if BASE is a power of two and -1 otherwise.

IFLOAT (OUTPUT, OPTIONAL) integer(i4b)

A guess for the number of bits dedicated to the representation of a floating point number if BASE is a power of two and -1 otherwise.

EXPEPSPOS (OUTPUT, OPTIONAL) integer(i4b)

The largest in magnitude negative integer such that

1.0 + float(base)**(expepspos) /= 1.

EXPEPSNEG (OUTPUT, OPTIONAL) integer(i4b)

The largest in magnitude negative integer such that

1.0 - float(base)**(expepsneg) /= 1.

MINEXPBASE (OUTPUT, OPTIONAL) integer(i4b)

The largest in magnitude negative integer such that float(base)**minexpbase is positive and normalized.

MAXEXPBASE (OUTPUT, OPTIONAL) integer(i4b)

The largest in magnitude positive integer such that float(base)**(maxexpbase) is positive and normalized.

EPSPOS (OUTPUT, OPTIONAL) real(stnd)

The smallest power of BASE whose sum with 1. is greater than 1. That is, float(base)**(expepspos).

EPSNEG (OUTPUT, OPTIONAL) real(stnd)

The smallest power of BASE whose difference with 1. is less than 1. That is, float(base)**(expepsneg).

EPSILPOS (OUTPUT, OPTIONAL) real(stnd)

The smallest positive floating point number whose sum with 1. is greater than 1.

EPSILNEG (OUTPUT, OPTIONAL) real(stnd)

The smallest positive floating point number whose difference with 1. is less than 1.

RNDUNIT (OUTPUT, OPTIONAL) real(stnd)

Unit roundoff of the machine.

Further Details¶

This subroutine is based on the routines MACHAR by Cody and DLAMCH in LAPACK77 (version 3). For further details, See:

Malcolm M.A., 1972: Algorithms to reveal properties of

floating-point arithmetic. Comms. of the ACM, 15, 949-951.

Gentleman, W.M., and Marovich, S.B., 1974: More on algorithms

that reveal properties of floating point arithmetic units. Comms. of the ACM, 17, 276-277.

Cody, W.J., 1988: MACHAR: A subroutine to dynamically

determine machine parameters,TOMS 14, No. 4, 303-311.

`function test_ieee ( )`¶

Purpose¶

TEST_IEEE try to determine if the computer follows the IEEE standard 754 for binary floating-point arithmetic.

Arguments¶

none

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to determine if the computer follows the IEEE standard 754 for binary floating-point arithmetic.

`function test_nan ( )`¶

Purpose¶

TEST_NAN returns TRUE if NaNs exist, and FALSE otherwise.

Arguments¶

None

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to determine if NaNs exist as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`function is_nan ( x )`¶

Purpose¶

This function returns TRUE if the scalar X is a NaN, and FALSE otherwise.

Arguments¶

X (INPUT) real(stnd)

The floating point number to be tested.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

Finally, if the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this function returns TRUE if the scalar X is equal to huge(X).

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`function is_nan ( x )`¶

Purpose¶

This function returns the value TRUE if any of the elements of the vector X is a NaN, and FALSE otherwise.

Arguments¶

X (INPUT) real(stnd), dimension(:)

The floating point vector to be tested.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

If the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this function returns TRUE if any of the elements of the vector X is equal to huge(X).

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`function is_nan ( x )`¶

Purpose¶

This function returns the value TRUE if any of the elements of the matrix X is a NaN, and FALSE otherwise.

Arguments¶

X (INPUT) real(stnd), dimension(:,:)

The floating point matrix to be tested.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

If the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this function returns TRUE if any of the elements of the matrix X is equal to huge(X).

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`subroutine replace_nan ( x, missing )`¶

Purpose¶

This subroutine replaces the scalar X with the scalar MISSING, if X is a NaN on input.

Arguments¶

X (INPUT/OUTPUT) real(stnd)

The floating point number to be tested.

MISSING (INPUT) real(stnd)

The floating point number used to replace NaNs.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

If the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this subroutine replaces X with the scalar MISSING, if X is equal to huge(X).

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`subroutine replace_nan ( x, missing )`¶

Purpose¶

This subroutine replaces the elements of the vector X which are NaNs by the scalar MISSING.

Arguments¶

X (INPUT/OUTPUT) real(stnd), dimension(:)

The floating point vector to be tested.

MISSING (INPUT) real(stnd)

The floating point number used to replace the NaNs.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

Finally, if the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this subroutine replaces the elements of the vector X which are equal to huge(X) with the scalar MISSING.

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`subroutine replace_nan ( x, missing )`¶

Purpose¶

This subroutine replaces the elements of the matrix X which are NaNs by the scalar MISSING.

Arguments¶

X (INPUT/OUTPUT) real(stnd), dimension(:,:)

The floating point matrix to be tested.

MISSING (INPUT) real(stnd)

The floating point number used to replace the NaNs.

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to detect NaNs as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

Finally, if the computer does not follow the IEEE standard 754 for binary floating-point arithmetic, this subroutine replaces the elements of the matrix X which are equal to huge(X) with the scalar MISSING.

For further details, see:

Cody, W.J., and Coonen, J.T., 1993: Algorithm 722,

TOMS 19, No. 4, 443-451.

`function nan ( )`¶

Purpose¶

NAN returns as a scalar function, the bit pattern corresponding to a quiet NaN in the IEEE standard 754 for binary floating-point arithmetic if the machine recognizes NaNs or the maximum floating point number of kind STND otherwise.

Arguments¶

None

Further Details¶

If the compiler follows the Fortran 2003 standard, the facilities provided by the IEEE_ARITHMETIC module are used to create a quiet NaN as defined in the IEEE standard 754 for binary floating-point arithmetic.

Otherwise, the routine exploits the IEEE requirement that NaNs compare as unequal to all values, including themselves.

Finally, NAN returns the maximum floating point number of kind STND, if the computer does not follow the IEEE standard 754 for binary floating-point arithmetic.

`function true_nan ( )`¶

Purpose¶

TRUE_NAN returns as a scalar function, the bit pattern corresponding to a quiet NaN in the IEEE standard 754 for binary floating-point arithmetic.

Arguments¶

None

Module_Num_Constants¶

function lamch ( cmach )¶

Purpose¶

Arguments¶

Further Details¶

subroutine mach ( basedigits, irnd, iuflow, igrd, iexp, ifloat, expepspos, expepsneg, minexpbase, maxexpbase, epspos, epsneg, epsilpos, epsilneg, rndunit )¶

Purpose¶

Arguments¶

Further Details¶

function test_ieee ( )¶

Purpose¶

Arguments¶

Further Details¶

function test_nan ( )¶

Purpose¶

Arguments¶

Further Details¶

function is_nan ( x )¶

Purpose¶

Arguments¶

Further Details¶

function is_nan ( x )¶

Purpose¶

Arguments¶

Further Details¶

function is_nan ( x )¶

Purpose¶

Arguments¶

Further Details¶

subroutine replace_nan ( x, missing )¶

Purpose¶

Arguments¶

Further Details¶

subroutine replace_nan ( x, missing )¶

Purpose¶

Arguments¶

Further Details¶

subroutine replace_nan ( x, missing )¶

Purpose¶

Arguments¶

Further Details¶

function nan ( )¶

Purpose¶

Arguments¶

Further Details¶

function true_nan ( )¶

Purpose¶

Arguments¶

`function lamch ( cmach )`¶

`subroutine mach ( basedigits, irnd, iuflow, igrd, iexp, ifloat, expepspos, expepsneg, minexpbase, maxexpbase, epspos, epsneg, epsilpos, epsilneg, rndunit )`¶

`function test_ieee ( )`¶

`function test_nan ( )`¶

`function is_nan ( x )`¶

`function is_nan ( x )`¶

`function is_nan ( x )`¶

`subroutine replace_nan ( x, missing )`¶

`subroutine replace_nan ( x, missing )`¶

`subroutine replace_nan ( x, missing )`¶

`function nan ( )`¶

`function true_nan ( )`¶