HPUX fpgetmask[3m]
fpgetround(3M) fpgetround(3M)
NAME
fpgetround(), fpsetround(), fpgetmask(), fpsetmask(), fpgetsticky(),
fpsetsticky(), fpsetdefaults(), fpgetcontrol(), fpsetcontrol(),
fpgetfastmode(), fpsetfastmode() - floating-point mode-control
functions
SYNOPSIS
#include <math.h>
fp_rnd fpgetround(void);
fp_rnd fpsetround(fp_rnd mode);
fp_except fpgetmask(void);
fp_except fpsetmask(fp_except value);
fp_except fpgetsticky(void);
fp_except fpsetsticky(fp_except value);
int fpgetfastmode(void);
int fpsetfastmode(int value);
void fpsetdefaults(void);
fp_control fpgetcontrol(void);
fp_control fpsetcontrol(fp_control value);
DESCRIPTION
The fpgetround() suite of functions allows programmers to manipulate
the floating-point control register (also called the floating-point
status register).
fpgetround() returns the current rounding mode. The type of the
returned value, fp_rnd, is defined as follows in <math.h>:
typedef enum {
FP_RZ=0, /* Round toward zero */
FP_RN, /* Round to nearest */
FP_RP, /* Round toward positive infinity */
FP_RM, /* Round toward negative infinity */
} fp_rnd;
The default value is FP_RN. Round-to-nearest mode rounds to the
representable value closest to the true value. If two representable
values are equally close to the true value, the system chooses the one
whose least significant bit is zero.
Hewlett-Packard Company - 1 - HP-UX Release 9.0: August 1992
fpgetround(3M) fpgetround(3M)
fpsetround() sets the rounding mode to the specified value of type
fp_rnd and returns the previous rounding mode.
There are five floating-point exceptions: divide-by-zero, overflow,
underflow, imprecise (inexact) result, and invalid operation. If a
floating-point exception occurs and the corresponding exception trap
enable bit is set to 1, the trap takes place. If an exception occurs
and the exception trap enable bit is set to 0, the corresponding
exception flag is set to 1 and no trap takes place. The exception-
trap-enable bits are sometimes called mask bits; the exception flags
are sometimes called sticky bits. The routines fpgetmask() and
fpgetsticky() return the current settings of these bits. To change
the settings of these bits, use fpsetmask() and fpsetsticky().
fpgetmask() returns the current exception trap enable bits. The type
of the returned value, fp_except, is defined as int in <math.h>. The
floating-point exception types are defined as follows in <math.h>:
#define FP_X_INV 0x10 /* invalid operation exception */
#define FP_X_DZ 0x08 /* divide-by-zero exception */
#define FP_X_OFL 0x04 /* overflow exception */
#define FP_X_UFL 0x02 /* underflow exception */
#define FP_X_IMP 0x01 /* imprecise (inexact result) */
#define FP_X_CLEAR 0x00 /* simply zero to clear all flags */
fpsetmask() sets or clears the exception trap enable bits and returns
the previous setting. The argument is an expression of type
fp_except. (To set or clear the exception trap enable bits at compile
time, use the compiler option +FPstring).
fpgetsticky() returns the current exception flags.
fpsetsticky() sets or clears the exception flags and returns the
previous setting. The argument is an expression of type fp_except.
fpgetfastmode() and fpsetfastmode() allow the programmer to change the
way the system handles underflow. Fast underflow mode, also known as
fastmode, is an alternative to IEEE-754-compliant underflow mode. On
Series 700/800 systems, underflow involves a fault into the kernel,
where the IEEE-mandated conversion of the result into a denormalized
value or zero is accomplished by software emulation. On some PA1.1
-based systems, fastmode causes the hardware to simply substitute a
zero for the result of an operation, with no fault occurring. This
may be a significant performance optimization for applications that
underflow frequently. Fastmode also causes denormalized floating-
point operands to be treated as if they were true zero operands.
fpgetfastmode() returns the current fastmode setting: 1 if fastmode is
set, 0 if the default IEEE-754-compliant underflow mode is set. On
systems that do not support fastmode, this function returns an
undefined value.
Hewlett-Packard Company - 2 - HP-UX Release 9.0: August 1992
fpgetround(3M) fpgetround(3M)
On systems that support fastmode, fpsetfastmode() sets fastmode to
either 1 (fastmode) or 0 (IEEE-754-compliant underflow mode) and
returns the previous setting. On systems that do not support
fastmode, this function has no effect.
fpsetdefaults() changes the default environment on Series 700
workstations, which is
Round to nearest (FP_RN)
All exception flags cleared (FP_X_CLEAR)
All exception traps disabled
Fast underflow mode disabled
fpsetdefaults() changes these defaults to more useful values.
Specifically, it enables traps for the invalid operation, divide-by-
zero, and overflow exceptions, while leaving the underflow and
inexact-result exception traps disabled. It sets the environment as
follows:
Round to nearest (FP_RN)
All exception flags cleared (FP_X_CLEAR)
All exception traps enabled except underflow and inexact result
(FP_X_INV+FP_X_DZ+FP_X_OFL)
Fast underflow mode enabled (if the system supports it)
fpgetcontrol() and fpsetcontrol() access fp0, the floating-point
unit's control register (also called the status register).
fpgetcontrol() returns the value of fp0. The type of the returned
value, fp_control, is defined as long in <math.h>.
fpsetcontrol() sets the value of fp0 and returns the previous value.
For the format of fp0, see the HP-UX Floating-Point Guide or the PA-
RISC 1.1 Architecture and Instruction Set Reference Manual.
DEPENDENCIES
Series 300/400
These functions are not supported on Series 300/400 systems.
Series 700/800
All of these functions are provided in the PA1.1 versions of the math
library only. The +DA1.1 linker option (default on Series 700
systems) links in a PA1.1 version automatically. A PA1.1 library can
be linked in explicitly. For more information, see the HP-UX
Floating-Point Guide.
WARNINGS
fpsetsticky() modifies all exception flags. fpsetmask() modifies all
exception trap enable bits.
Hewlett-Packard Company - 3 - HP-UX Release 9.0: August 1992
fpgetround(3M) fpgetround(3M)
Both C and FORTRAN require truncation (rounding to zero) for
floating-point to integer conversions. The current rounding mode has
no effect on these conversions.
Hewlett-Packard Company - 4 - HP-UX Release 9.0: August 1992