1 @node Representation Limits, System Configuration Limits, System Information, Top
2 @chapter Representation Limits
4 This chapter contains information about constants and parameters that
5 characterize the representation of the various integer and
6 floating-point types supported by the GNU C library.
9 * Integer Representation Limits:: Determining maximum and minimum
10 representation values of
11 various integer subtypes.
12 * Floating-Point Limits :: Parameters which characterize
13 supported floating-point
14 representations on a particular
18 @node Integer Representation Limits, Floating-Point Limits , , Representation Limits
19 @section Integer Representation Limits
20 @cindex integer representation limits
21 @cindex representation limits, integer
22 @cindex limits, integer representation
24 Sometimes it is necessary for programs to know about the internal
25 representation of various integer subtypes. For example, if you want
26 your program to be careful not to overflow an @code{int} counter
27 variable, you need to know what the largest representable value that
28 fits in an @code{int} is. These kinds of parameters can vary from
29 compiler to compiler and machine to machine. Another typical use of
30 this kind of parameter is in conditionalizing data structure definitions
31 with @samp{#ifdef} to select the most appropriate integer subtype that
32 can represent the required range of values.
34 Macros representing the minimum and maximum limits of the integer types
35 are defined in the header file @file{limits.h}. The values of these
36 macros are all integer constant expressions.
41 @deftypevr Macro int CHAR_BIT
42 This is the number of bits in a @code{char}, usually eight.
47 @deftypevr Macro int SCHAR_MIN
48 This is the minimum value that can be represented by a @code{signed char}.
53 @deftypevr Macro int SCHAR_MAX
54 This is the maximum value that can be represented by a @code{signed char}.
59 @deftypevr Macro int UCHAR_MAX
60 This is the maximum value that can be represented by a @code{unsigned char}.
61 (The minimum value of an @code{unsigned char} is zero.)
66 @deftypevr Macro int CHAR_MIN
67 This is the minimum value that can be represented by a @code{char}.
68 It's equal to @code{SCHAR_MIN} if @code{char} is signed, or zero
74 @deftypevr Macro int CHAR_MAX
75 This is the maximum value that can be represented by a @code{char}.
76 It's equal to @code{SCHAR_MAX} if @code{char} is signed, or
77 @code{UCHAR_MAX} otherwise.
82 @deftypevr Macro int SHRT_MIN
83 This is the minimum value that can be represented by a @code{signed
84 short int}. On most machines that the GNU C library runs on,
85 @code{short} integers are 16-bit quantities.
90 @deftypevr Macro int SHRT_MAX
91 This is the maximum value that can be represented by a @code{signed
97 @deftypevr Macro int USHRT_MAX
98 This is the maximum value that can be represented by an @code{unsigned
99 short int}. (The minimum value of an @code{unsigned short int} is zero.)
104 @deftypevr Macro int INT_MIN
105 This is the minimum value that can be represented by a @code{signed
106 int}. On most machines that the GNU C system runs on, an @code{int} is
112 @deftypevr Macro int INT_MAX
113 This is the maximum value that can be represented by a @code{signed
119 @deftypevr Macro {unsigned int} UINT_MAX
120 This is the maximum value that can be represented by an @code{unsigned
121 int}. (The minimum value of an @code{unsigned int} is zero.)
126 @deftypevr Macro {long int} LONG_MIN
127 This is the minimum value that can be represented by a @code{signed long
128 int}. On most machines that the GNU C system runs on, @code{long}
129 integers are 32-bit quantities, the same size as @code{int}.
134 @deftypevr Macro {long int} LONG_MAX
135 This is the maximum value that can be represented by a @code{signed long
141 @deftypevr Macro {unsigned long int} ULONG_MAX
142 This is the maximum value that can be represented by an @code{unsigned
143 long int}. (The minimum value of an @code{unsigned long int} is zero.)
146 @strong{Incomplete:} There should be corresponding limits for the GNU
147 C Compiler's @code{long long} type, too. (But they are not now present
150 The header file @file{limits.h} also defines some additional constants
151 that parameterize various operating system and file system limits. These
152 constants are described in @ref{System Parameters} and @ref{File System
157 @node Floating-Point Limits , , Integer Representation Limits, Representation Limits
158 @section Floating-Point Limits
159 @cindex floating-point number representation
160 @cindex representation, floating-point number
161 @cindex limits, floating-point representation
163 Because floating-point numbers are represented internally as approximate
164 quantities, algorithms for manipulating floating-point data often need
165 to be parameterized in terms of the accuracy of the representation.
166 Some of the functions in the C library itself need this information; for
167 example, the algorithms for printing and reading floating-point numbers
168 (@pxref{I/O on Streams}) and for calculating trigonometric and
169 irrational functions (@pxref{Mathematics}) use information about the
170 underlying floating-point representation to avoid round-off error and
171 loss of accuracy. User programs that implement numerical analysis
172 techniques also often need to be parameterized in this way in order to
173 minimize or compute error bounds.
175 The specific representation of floating-point numbers varies from
176 machine to machine. The GNU C library defines a set of parameters which
177 characterize each of the supported floating-point representations on a
181 * Floating-Point Representation:: Definitions of terminology.
182 * Floating-Point Parameters:: Descriptions of the library
184 * IEEE Floating Point:: An example of a common
188 @node Floating-Point Representation, Floating-Point Parameters, , Floating-Point Limits
189 @subsection Floating-Point Representation
191 This section introduces the terminology used to characterize the
192 representation of floating-point numbers.
194 You are probably already familiar with most of these concepts in terms
195 of scientific or exponential notation for floating-point numbers. For
196 example, the number @code{123456.0} could be expressed in exponential
197 notation as @code{1.23456e+05}, a shorthand notation indicating that the
198 mantissa @code{1.23456} is multiplied by the base @code{10} raised to
201 More formally, the internal representation of a floating-point number
202 can be characterized in terms of the following parameters:
206 The @dfn{sign} is either @code{-1} or @code{1}.
207 @cindex sign (of floating-point number)
210 The @dfn{base} or @dfn{radix} for exponentiation; an integer greater
211 than @code{1}. This is a constant for the particular representation.
212 @cindex base (of floating-point number)
213 @cindex radix (of floating-point number)
216 The @dfn{exponent} to which the base is raised. The upper and lower
217 bounds of the exponent value are constants for the particular
219 @cindex exponent (of floating-point number)
221 Sometimes, in the actual bits representing the floating-point number,
222 the exponent is @dfn{biased} by adding a constant to it, to make it
223 always be represented as an unsigned quantity. This is only important
224 if you have some reason to pick apart the bit fields making up the
225 floating-point number by hand, which is something for which the GNU
226 library provides no support. So this is ignored in the discussion that
228 @cindex bias (of floating-point number exponent)
231 The value of the @dfn{mantissa} or @dfn{significand}, which is an
233 @cindex mantissa (of floating-point number)
234 @cindex significand (of floating-point number)
237 The @dfn{precision} of the mantissa. If the base of the representation
238 is @var{b}, then the precision is the number of base-@var{b} digits in
239 the mantissa. This is a constant for the particular representation.
241 Many floating-point representations have an implicit @dfn{hidden bit} in
242 the mantissa. Any such hidden bits are counted in the precision.
243 Again, the GNU library provides no facilities for dealing with such low-level
244 aspects of the representation.
245 @cindex precision (of floating-point number)
246 @cindex hidden bit (of floating-point number mantissa)
249 The mantissa of a floating-point number actually represents an implicit
250 fraction whose denominator is the base raised to the power of the
251 precision. Since the largest representable mantissa is one less than
252 this denominator, the value of the fraction is always strictly less than
253 @code{1}. The mathematical value of a floating-point number is then the
254 product of this fraction; the sign; and the base raised to the exponent.
256 If the floating-point number is @dfn{normalized}, the mantissa is also
257 greater than or equal to the base raised to the power of one less
258 than the precision (unless the number represents a floating-point zero,
259 in which case the mantissa is zero). The fractional quantity is
260 therefore greater than or equal to @code{1/@var{b}}, where @var{b} is
262 @cindex normalized floating-point number
264 @node Floating-Point Parameters, IEEE Floating Point, Floating-Point Representation, Floating-Point Limits
265 @subsection Floating-Point Parameters
267 @strong{Incomplete:} This section needs some more concrete examples
268 of what these parameters mean and how to use them in a program.
270 These macro definitions can be accessed by including the header file
271 @file{float.h} in your program.
274 Macro names starting with @samp{FLT_} refer to the @code{float} type,
275 while names beginning with @samp{DBL_} refer to the @code{double} type
276 and names beginning with @samp{LDBL_} refer to the @code{long double}
277 type. (In implementations that do not support @code{long double} as
278 a distinct data type, the values for those constants are the same
279 as the corresponding constants for the @code{double} type.)@refill
280 @cindex @code{float} representation limits
281 @cindex @code{double} representation limits
282 @cindex @code{long double} representation limits
284 Of these macros, only @code{FLT_RADIX} is guaranteed to be a constant
285 expression. The other macros listed here cannot be reliably used in
286 places that require constant expressions, such as @samp{#if}
287 preprocessing directives or array size specifications.
289 Although the ANSI C standard specifies minimum and maximum values for
290 most of these parameters, the GNU C implementation uses whatever
291 floating-point representations are supported by the underlying hardware.
292 So whether GNU C actually satisfies the ANSI C requirements depends on
293 what machine it is running on.
297 @deftypevr Macro int FLT_ROUNDS
298 This value characterizes the rounding mode for floating-point addition.
299 The following values indicate standard rounding modes:
303 The mode is indeterminable.
305 Rounding is towards zero.
307 Rounding is to the nearest number.
309 Rounding is towards positive infinity.
311 Rounding is towards negative infinity.
315 Any other value represents a machine-dependent nonstandard rounding
321 @deftypevr Macro int FLT_RADIX
322 This is the value of the base, or radix, of exponent representation.
323 This is guaranteed to be a constant expression, unlike the other macros
324 described in this section.
329 @deftypevr Macro int FLT_MANT_DIG
330 This is the number of base-@code{FLT_RADIX} digits in the floating-point
331 mantissa for the @code{float} data type.
336 @deftypevr Macro int DBL_MANT_DIG
337 This is the number of base-@code{FLT_RADIX} digits in the floating-point
338 mantissa for the @code{double} data type.
343 @deftypevr Macro int LDBL_MANT_DIG
344 This is the number of base-@code{FLT_RADIX} digits in the floating-point
345 mantissa for the @code{long double} data type.
350 @deftypevr Macro int FLT_DIG
351 This is the number of decimal digits of precision for the @code{float}
352 data type. Technically, if @var{p} and @var{b} are the precision and
353 base (respectively) for the representation, then the decimal precision
354 @var{q} is the maximum number of decimal digits such that any floating
355 point number with @var{q} base 10 digits can be rounded to a floating
356 point number with @var{p} base @var{b} digits and back again, without
357 change to the @var{q} decimal digits.
359 The value of this macro is guaranteed to be at least @code{6}.
364 @deftypevr Macro int DBL_DIG
365 This is similar to @code{FLT_DIG}, but is for the @code{double} data
366 type. The value of this macro is guaranteed to be at least @code{10}.
371 @deftypevr Macro int LDBL_DIG
372 This is similar to @code{FLT_DIG}, but is for the @code{long double}
373 data type. The value of this macro is guaranteed to be at least
379 @deftypevr Macro int FLT_MIN_EXP
380 This is the minimum negative integer such that the mathematical value
381 @code{FLT_RADIX} raised to this power minus 1 can be represented as a
382 normalized floating-point number of type @code{float}. In terms of the
383 actual implementation, this is just the smallest value that can be
384 represented in the exponent field of the number.
389 @deftypevr Macro int DBL_MIN_EXP
390 This is similar to @code{FLT_MIN_EXP}, but is for the @code{double} data
396 @deftypevr Macro int LDBL_MIN_EXP
397 This is similar to @code{FLT_MIN_EXP}, but is for the @code{long double}
403 @deftypevr Macro int FLT_MIN_10_EXP
404 This is the minimum negative integer such that the mathematical value
405 @code{10} raised to this power minus 1 can be represented as a
406 normalized floating-point number of type @code{float}. This is
407 guaranteed to be no greater than @code{-37}.
412 @deftypevr Macro int DBL_MIN_10_EXP
413 This is similar to @code{FLT_MIN_10_EXP}, but is for the @code{double}
419 @deftypevr Macro int LDBL_MIN_10_EXP
420 This is similar to @code{FLT_MIN_10_EXP}, but is for the @code{long
428 @deftypevr Macro int FLT_MAX_EXP
429 This is the maximum negative integer such that the mathematical value
430 @code{FLT_RADIX} raised to this power minus 1 can be represented as a
431 floating-point number of type @code{float}. In terms of the actual
432 implementation, this is just the largest value that can be represented
433 in the exponent field of the number.
438 @deftypevr Macro int DBL_MAX_EXP
439 This is similar to @code{FLT_MAX_EXP}, but is for the @code{double} data
445 @deftypevr Macro int LDBL_MAX_EXP
446 This is similar to @code{FLT_MAX_EXP}, but is for the @code{long double}
452 @deftypevr Macro int FLT_MAX_10_EXP
453 This is the maximum negative integer such that the mathematical value
454 @code{10} raised to this power minus 1 can be represented as a
455 normalized floating-point number of type @code{float}. This is
456 guaranteed to be at least @code{37}.
461 @deftypevr Macro int DBL_MAX_10_EXP
462 This is similar to @code{FLT_MAX_10_EXP}, but is for the @code{double}
468 @deftypevr Macro int LDBL_MAX_10_EXP
469 This is similar to @code{FLT_MAX_10_EXP}, but is for the @code{long
476 @deftypevr Macro double FLT_MAX
477 The value of this macro is the maximum representable floating-point
478 number of type @code{float}, and is guaranteed to be at least
484 @deftypevr Macro double DBL_MAX
485 The value of this macro is the maximum representable floating-point
486 number of type @code{double}, and is guaranteed to be at least
492 @deftypevr Macro {long double} LDBL_MAX
493 The value of this macro is the maximum representable floating-point
494 number of type @code{long double}, and is guaranteed to be at least
501 @deftypevr Macro double FLT_MIN
502 The value of this macro is the minimum normalized positive
503 floating-point number that is representable by type @code{float}, and is
504 guaranteed to be no more than @code{1E-37}.
509 @deftypevr Macro double DBL_MIN
510 The value of this macro is the minimum normalized positive
511 floating-point number that is representable by type @code{double}, and
512 is guaranteed to be no more than @code{1E-37}.
517 @deftypevr Macro {long double} LDBL_MIN
518 The value of this macro is the minimum normalized positive
519 floating-point number that is representable by type @code{long double},
520 and is guaranteed to be no more than @code{1E-37}.
526 @deftypevr Macro double FLT_EPSILON
527 This is the minimum positive floating-point number of type @code{float}
528 such that @code{1.0 + FLT_EPSILON != 1.0} is true. It's guaranteed to
529 be no greater than @code{1E-5}.
534 @deftypevr Macro double DBL_EPSILON
535 This is similar to @code{FLT_EPSILON}, but is for the @code{double}
536 type. The maximum value is @code{1E-9}.
541 @deftypevr Macro {long double} LDBL_EPSILON
542 This is similar to @code{FLT_EPSILON}, but is for the @code{long double}
543 type. The maximum value is @code{1E-9}.
547 @node IEEE Floating Point, , Floating-Point Parameters, Floating-Point Limits
548 @subsection IEEE Floating Point
549 @cindex IEEE floating-point representation
550 @cindex floating-point, IEEE
554 Here is an example showing how these parameters work for a common
555 floating point representation, specified by the @cite{IEEE Standard for
556 Binary Floating-Point Arithmetic (ANSI/IEEE Std 754-1985)}. Nearly
557 all computers today use this format.
559 The IEEE single-precision float representation uses a base of 2. There
560 is a sign bit, a mantissa with 23 bits plus one hidden bit (so the total
561 precision is 24 base-2 digits), and an 8-bit exponent that can represent
562 values in the range -125 to 128, inclusive.
564 So, for an implementation that uses this representation for the
565 @code{float} data type, appropriate values for the corresponding
576 FLT_MIN 1.17549435E-38F
577 FLT_MAX 3.40282347E+38F
578 FLT_EPSILON 1.19209290E-07F
581 Here are the values for the @code{double} data type:
590 DBL_MAX 1.7976931348623157E+308
591 DBL_MIN 2.2250738585072014E-308
592 DBL_EPSILON 2.2204460492503131E-016