@node Floating-Point Limits 
@chapter Floating-Point Limits
@pindex <float.h>
@cindex floating-point number representation
@cindex representation of floating-point numbers

Because floating-point numbers are represented internally as approximate
quantities, algorithms for manipulating floating-point data often need
to be parameterized in terms of the accuracy of the representation.
Some of the functions in the C library itself need this information; for
example, the algorithms for printing and reading floating-point numbers
(@pxref{I/O on Streams}) and for calculating trigonometric and
irrational functions (@pxref{Mathematics}) use information about the
underlying floating-point representation to avoid round-off error and
loss of accuracy.  User programs that implement numerical analysis
techniques also often need to be parameterized in this way in order to
minimize or compute error bounds.

The specific representation of floating-point numbers varies from
machine to machine.  The GNU C Library defines a set of parameters which
characterize each of the supported floating-point representations on a
particular system.

@menu
* Floating-Point Representation::   Definitions of terminology.
* Floating-Point Parameters::	    Descriptions of the library facilities.
* IEEE Floating-Point::		    An example of a common representation.
@end menu

@node Floating-Point Representation
@section Floating-Point Representation

This section introduces the terminology used to characterize the
representation of floating-point numbers.

You are probably already familiar with most of these concepts in terms
of scientific or exponential notation for floating-point numbers.  For
example, the number @code{123456.0} could be expressed in exponential
notation as @code{1.23456e+05}, a shorthand notation indicating that the
mantissa @code{1.23456} is multiplied by the base @code{10} raised to
power @code{5}.

More formally, the internal representation of a floating-point number
can be characterized in terms of the following parameters:

@itemize @bullet
@item
The @dfn{sign} is either @code{-1} or @code{1}.
@cindex sign (of floating-point number)

@item
The @dfn{base} or @dfn{radix} for exponentiation; an integer greater
than @code{1}.  This is a constant for the particular representation.
@cindex base (of floating-point number)
@cindex radix (of floating-point number)

@item
The @dfn{exponent} to which the base is raised.  The upper and lower
bounds of the exponent value are constants for the particular
representation.
@cindex exponent (of floating-point number)

Sometimes, in the actual bits representing the floating-point number,
the exponent is @dfn{biased} by adding a constant to it, to make it
always be represented as an unsigned quantity.  This is only important
if you have some reason to pick apart the bit fields making up the
floating-point number by hand, which is something for which the GNU
library provides no support.  So this is ignored in the discussion that
follows.
@cindex bias, in exponent (of floating-point number)

@item
The value of the @dfn{mantissa} or @dfn{significand}, which is an
unsigned quantity.
@cindex mantissa (of floating-point number)
@cindex significand (of floating-point number)

@item 
The @dfn{precision} of the mantissa.  If the base of the representation
is @var{b}, then the precision is the number of base-@var{b} digits in
the mantissa.  This is a constant for the particular representation.

Many floating-point representations have an implicit @dfn{hidden bit} in
the mantissa.  Any such hidden bits are counted in the precision.
Again, the GNU library provides no facilities for dealing with such low-level
aspects of the representation.
@cindex precision (of floating-point number)
@cindex hidden bit, in mantissa (of floating-point number)
@end itemize

The mantissa of a floating-point number actually represents an implicit
fraction whose denominator is the base raised to the power of the
precision.  Since the largest representable mantissa is one less than
this denominator, the value of the fraction is always strictly less than
@code{1}.  The mathematical value of a floating-point number is then the
product of this fraction; the sign; and the base raised to the exponent.

If the floating-point number is @dfn{normalized}, the mantissa is also
greater than or equal to the base raised to the power of one less
than the precision (unless the number represents a floating-point zero,
in which case the mantissa is zero).  The fractional quantity is
therefore greater than or equal to @code{1/@var{b}}, where @var{b} is
the base.
@cindex normalized floating-point number

@node Floating-Point Parameters
@section Floating-Point Parameters

@strong{Incomplete:}  This section needs some more concrete examples
of what these parameters mean and how to use them in a program.

These macro definitions can be accessed by including the header file
@file{<float.h>} in your program.

Macro names starting with @samp{FLT_} refer to the @code{float} type,
while names beginning with @samp{DBL_} refer to the @code{double} type
and names beginning with @samp{LDBL_} refer to the @code{long double}
type.  (In implementations that do not support @code{long double} as
a distinct data type, the values for those constants are the same
as the corresponding constants for the @code{double} type.)@refill

Note that only @code{FLT_RADIX} is guaranteed to be a constant
expression, so the other macros listed here cannot be reliably used in
places that require constant expressions, such as @samp{#if}
preprocessing directives and array size specifications.

Although the ANSI C standard specifies minimum and maximum values for
most of these parameters, the GNU C implementation uses whatever
floating-point representations are supported by the underlying hardware.
So whether GNU C actually satisfies the ANSI C requirements depends on
what machine it is running on.

@comment float.h
@comment ANSI
@defvr Macro FLT_ROUNDS
This value characterizes the rounding mode for floating-point addition.
The following values indicate standard rounding modes:

@table @code
@item -1
The mode is indeterminable.
@item 0
Rounding is towards zero.
@item 1
Rounding is to the nearest number.
@item 2
Rounding is towards positive infinity.
@item 3
Rounding is towards negative infinity.
@end table

@noindent
Any other value represents a machine-dependent nonstandard rounding
mode.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_RADIX
This is the value of the base, or radix, of exponent representation.
This is guaranteed to be a constant expression, unlike the other macros
described in this section.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_MANT_DIG
This is the number of base-@code{FLT_RADIX} digits in the floating-point
mantissa for the @code{float} data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MANT_DIG
This is the number of base-@code{FLT_RADIX} digits in the floating-point
mantissa for the @code{double} data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MANT_DIG
This is the number of base-@code{FLT_RADIX} digits in the floating-point
mantissa for the @code{long double} data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_DIG
This is the number of decimal digits of precision for the @code{float}
data type.  Technically, if @var{p} and @var{b} are the precision and
base (respectively) for the representation, then the decimal precision
@var{q} is the maximum number of decimal digits such that any floating
point number with @var{q} base 10 digits can be rounded to a floating
point number with @var{p} base @var{b} digits and back again, without
change to the @var{q} decimal digits.

The value of this macro is guaranteed to be at least @code{6}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_DIG
This is similar to @code{FLT_DIG}, but is for the @code{double} data
type.  The value of this macro is guaranteed to be at least @code{10}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_DIG
This is similar to @code{FLT_DIG}, but is for the @code{long double}
data type.  The value of this macro is guaranteed to be at least
@code{10}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_MIN_EXP
This is the minimum negative integer such that the mathematical value
@code{FLT_RADIX} raised to this power minus 1 can be represented as a
normalized floating-point number of type @code{float}.  In terms of the
actual implementation, this is just the smallest value that can be
represented in the exponent field of the number.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MIN_EXP
This is similar to @code{FLT_MIN_EXP}, but is for the @code{double} data
type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MIN_EXP
This is similar to @code{FLT_MIN_EXP}, but is for the @code{long double}
data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_MIN_10_EXP
This is the minimum negative integer such that the mathematical value
@code{10} raised to this power minus 1 can be represented as a
normalized floating-point number of type @code{float}.  This is
guaranteed to be no greater than @code{-37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MIN_10_EXP
This is similar to @code{FLT_MIN_10_EXP}, but is for the @code{double}
data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MIN_10_EXP
This is similar to @code{FLT_MIN_10_EXP}, but is for the @code{long
double} data type.
@end defvr



@comment float.h
@comment ANSI
@defvr Macro FLT_MAX_EXP
This is the maximum negative integer such that the mathematical value
@code{FLT_RADIX} raised to this power minus 1 can be represented as a
floating-point number of type @code{float}.  In terms of the actual
implementation, this is just the largest value that can be represented
in the exponent field of the number.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MAX_EXP
This is similar to @code{FLT_MAX_EXP}, but is for the @code{double} data
type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MAX_EXP
This is similar to @code{FLT_MAX_EXP}, but is for the @code{long double}
data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro FLT_MAX_10_EXP
This is the maximum negative integer such that the mathematical value
@code{10} raised to this power minus 1 can be represented as a
normalized floating-point number of type @code{float}.  This is
guaranteed to be at least @code{37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MAX_10_EXP
This is similar to @code{FLT_MAX_10_EXP}, but is for the @code{double}
data type.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MAX_10_EXP
This is similar to @code{FLT_MAX_10_EXP}, but is for the @code{long
double} data type.
@end defvr


@comment float.h
@comment ANSI
@defvr Macro FLT_MAX
The value of this macro is the maximum representable floating-point
number of type @code{float}, and is guaranteed to be at least
@code{1E+37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MAX
The value of this macro is the maximum representable floating-point
number of type @code{double}, and is guaranteed to be at least
@code{1E+37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MAX
The value of this macro is the maximum representable floating-point
number of type @code{long double}, and is guaranteed to be at least
@code{1E+37}.
@end defvr


@comment float.h
@comment ANSI
@defvr Macro FLT_MIN
The value of this macro is the minimum normalized positive
floating-point number that is representable by type @code{float}, and is
guaranteed to be no more than @code{1E-37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_MIN
The value of this macro is the minimum normalized positive
floating-point number that is representable by type @code{double}, and
is guaranteed to be no more than @code{1E-37}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_MIN
The value of this macro is the minimum normalized positive
floating-point number that is representable by type @code{long double},
and is guaranteed to be no more than @code{1E-37}.
@end defvr


@comment float.h
@comment ANSI
@defvr Macro FLT_EPSILON
This is the minimum positive floating-point number of type @code{float}
such that @code{1.0 + FLT_EPSILON != 1.0} is true.  It's guaranteed to
be no greater than @code{1E-5}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro DBL_EPSILON
This is similar to @code{FLT_EPSILON}, but is for the @code{double}
type.  The maximum value is @code{1E-9}.
@end defvr

@comment float.h
@comment ANSI
@defvr Macro LDBL_EPSILON
This is similar to @code{FLT_EPSILON}, but is for the @code{long double}
type.  The maximum value is @code{1E-9}.
@end defvr



@node IEEE Floating Point
@section IEEE Floating Point

Here is an example showing how these parameters work for a common
floating point representation, specified by the @cite{IEEE Standard for
Binary Floating-Point Arithmetic (ANSI/IEEE Std 754-1985)}.

The IEEE single-precision float representation uses a base of 2.  There
is a sign bit, a mantissa with 23 bits plus one hidden bit (so the total
precision is 24 base-2 digits), and an 8-bit exponent that can represent
values in the range -125 to 128, inclusive.

So, for an implementation that uses this representation for the
@code{float} data type, appropriate values for the corresponding
parameters are:

@example
FLT_RADIX                         2
FLT_MANT_DIG                     24
FLT_DIG                           6
FLT_MIN_EXP                    -125
FLT_MIN_10_EXP                  -37
FLT_MAX_EXP                     128
FLT_MAX_10_EXP                  +38
FLT_MIN             1.17549435E-38F
FLT_MAX             3.40282347E+38F
FLT_EPSILON         1.19209290E-07F
@end example