1FENV(3) Linux Programmer's Manual FENV(3)
2
6 feclearexcept, fegetexceptflag, feraiseexcept, fesetexceptflag, fetes‐
7 texcept, fegetenv, fegetround, feholdexcept, fesetround, fesetenv,
8 feupdateenv, feenableexcept, fedisableexcept, fegetexcept - floating-
9 point rounding and exception handling
10
12 #include <fenv.h>
13 int feclearexcept(int excepts);
15 int fegetexceptflag(fexcept_t *flagp, int excepts);
16 int feraiseexcept(int excepts);
17 int fesetexceptflag(const fexcept_t *flagp, int excepts);
18 int fetestexcept(int excepts);
19 int fegetround(void);
21 int fesetround(int rounding_mode);
22 int fegetenv(fenv_t *envp);
24 int feholdexcept(fenv_t *envp);
25 int fesetenv(const fenv_t *envp);
26 int feupdateenv(const fenv_t *envp);
27 Link with -lm.
29
31 These eleven functions were defined in C99, and describe the handling
32 of floating-point rounding and exceptions (overflow, zero-divide etc.).
33 Exceptions
35 The divide-by-zero exception occurs when an operation on finite numbers
36 produces infinity as exact answer.
37 The overflow exception occurs when a result has to be represented as a
39 floating-point number, but has (much) larger absolute value than the
40 largest (finite) floating-point number that is representable.
41 The underflow exception occurs when a result has to be represented as a
43 floating-point number, but has smaller absolute value than the smallest
44 positive normalized floating-point number (and would lose much accuracy
45 when represented as a denormalized number).
46 The inexact exception occurs when the rounded result of an operation is
48 not equal to the infinite precision result. It may occur whenever
49 overflow or underflow occurs.
50 The invalid exception occurs when there is no well-defined result for
52 an operation, as for 0/0 or infinity - infinity or sqrt(-1).
53 Exception handling
55 Exceptions are represented in two ways: as a single bit (exception
56 present/absent), and these bits correspond in some implementation-
57 defined way with bit positions in an integer, and also as an opaque
58 structure that may contain more information about the exception (per‐
59 haps the code address where it occurred).
60 Each of the macros FE_DIVBYZERO, FE_INEXACT, FE_INVALID, FE_OVERFLOW,
62 FE_UNDERFLOW is defined when the implementation supports handling of
63 the corresponding exception, and if so then defines the corresponding
64 bit(s), so that one can call exception handling functions, for example,
65 using the integer argument FE_OVERFLOW|FE_UNDERFLOW. Other exceptions
66 may be supported. The macro FE_ALL_EXCEPT is the bitwise OR of all
67 bits corresponding to supported exceptions.
68 The feclearexcept() function clears the supported exceptions repre‐
70 sented by the bits in its argument.
71 The fegetexceptflag() function stores a representation of the state of
73 the exception flags represented by the argument excepts in the opaque
74 object *flagp.
75 The feraiseexcept() function raises the supported exceptions repre‐
77 sented by the bits in excepts.
78 The fesetexceptflag() function sets the complete status for the excep‐
80 tions represented by excepts to the value *flagp. This value must have
81 been obtained by an earlier call of fegetexceptflag() with a last argu‐
82 ment that contained all bits in excepts.
83 The fetestexcept() function returns a word in which the bits are set
85 that were set in the argument excepts and for which the corresponding
86 exception is currently set.
87 Rounding mode
89 The rounding mode determines how the result of floating-point opera‐
90 tions is treated when the result cannot be exactly represented in the
91 signifcand. Various rounding modes may be provided: round to nearest
92 (the default), round up (towards positive infinity), round down
93 (towards negative infinity), and round towards zero.
94 Each of the macros FE_TONEAREST, FE_UPWARD, FE_DOWNWARD, and
96 FE_TOWARDZERO is defined when the implementation supports getting and
97 setting the corresponding rounding direction.
98 The fegetround() function returns the macro corresponding to the cur‐
100 rent rounding mode.
101 The fesetround() function sets the rounding mode as specified by its
103 argument and returns zero when it was successful.
104 C99 and POSIX.1-2008 specify an identifier, FLT_ROUNDS, defined in
106 <float.h>, which indicates the implementation-defined rounding behavior
107 for floating-point addition. This identifier has one of the following
108 values:
109 -1 The rounding mode is not determinable.
111 0 Rounding is towards 0.
113 1 Rounding is towards nearest number.
115 2 Rounding is towards positive infinity.
117 3 Rounding is towards negative infinity.
119 Other values represent machine-dependent, non-standard rounding modes.
121 The value of FLT_ROUNDS should reflect the current rounding mode as set
123 by fesetround() (but see BUGS).
124 Floating-point environment
126 The entire floating-point environment, including control modes and sta‐
127 tus flags, can be handled as one opaque object, of type fenv_t. The
128 default environment is denoted by FE_DFL_ENV (of type const fenv_t *).
129 This is the environment setup at program start and it is defined by ISO
130 C to have round to nearest, all exceptions cleared and a non-stop (con‐
131 tinue on exceptions) mode.
132 The fegetenv() function saves the current floating-point environment in
134 the object *envp.
135 The feholdexcept() function does the same, then clears all exception
137 flags, and sets a non-stop (continue on exceptions) mode, if available.
138 It returns zero when successful.
139 The fesetenv() function restores the floating-point environment from
141 the object *envp. This object must be known to be valid, for example,
142 the result of a call to fegetenv() or feholdexcept() or equal to
143 FE_DFL_ENV. This call does not raise exceptions.
144 The feupdateenv() function installs the floating-point environment rep‐
146 resented by the object *envp, except that currently raised exceptions
147 are not cleared. After calling this function, the raised exceptions
148 will be a bitwise OR of those previously set with those in *envp. As
149 before, the object *envp must be known to be valid.
150
152 These functions return zero on success and non-zero if an error
153 occurred.
154
156 These functions first appeared in glibc in version 2.1.
157
159 IEC 60559 (IEC 559:1989), ANSI/IEEE 854, C99, POSIX.1-2001.
160
162 Glibc Notes
163 If possible, the GNU C Library defines a macro FE_NOMASK_ENV which rep‐
164 resents an environment where every exception raised causes a trap to
165 occur. You can test for this macro using #ifdef. It is only defined
166 if _GNU_SOURCE is defined. The C99 standard does not define a way to
167 set individual bits in the floating-point mask, for example, to trap on
168 specific flags. glibc 2.2 supports the functions feenableexcept() and
169 fedisableexcept() to set individual floating-point traps, and fegetex‐
170 cept() to query the state.
171 #define _GNU_SOURCE
173 #include <fenv.h>
174 int feenableexcept(int excepts);
176 int fedisableexcept(int excepts);
177 int fegetexcept(void);
178 The feenableexcept() and fedisableexcept() functions enable (disable)
180 traps for each of the exceptions represented by excepts and return the
181 previous set of enabled exceptions when successful, and -1 otherwise.
182 The fegetexcept() function returns the set of all currently enabled
183 exceptions.
184
186 C99 specifies that the value of FLT_ROUNDS should reflect changes to
187 the current rounding mode, as set by fesetround(). Currently, this
188 does not occur: FLT_ROUNDS always has the value 1.
189
191 feature_test_macros(7), math_error(7)
192
194 This page is part of release 3.22 of the Linux man-pages project. A
195 description of the project, and information about reporting bugs, can
196 be found at http://www.kernel.org/doc/man-pages/.
197Linux 2008-08-11 FENV(3)