United States-English |
|
|
HP-UX Reference > Ffesetflushtozero(3M)HP-UX 11i Version 3: February 2007 |
|
NAMEfesetflushtozero() — set floating-point underflow mode DESCRIPTIONThe fesetflushtozero() function sets the current underflow mode. If the argument is 1, the underflow mode is set to flush-to-zero mode. If the argument is zero, the underflow mode is set to IEEE-754-compliant (gradual) underflow mode. For arguments other than 1 or zero, the effect is undefined. The default underflow mode is IEEE-754-compliant. Flush-to-zero mode, also known as fast underflow mode, is supported on most PA1.1 systems and on all PA2.0 and HP Integrity servers. In IEEE-754-compliant mode, cases that might underflow may be handled by trapping into the kernel, where the IEEE-mandated conversion of the result into a denormalized value or zero is accomplished by software emulation. On PA-RISC systems, flush-to-zero mode allows the substitution of a zero for denormal operands and operation results, without trapping into the kernel. On Integrity servers, flush-to-zero mode causes a zero to be substituted for denormal results (but not operands), without trapping into the kernel. Flush-to-zero mode may offer a significant performance improvement for some applications. USAGETo use this function, compile either with the default -Ae option or with the -Aa and -D_HPUX_SOURCE options. Make sure your program includes <fenv.h>. For Integrity servers, specify +Ofenvaccess on the compiler command line or place the call to this function under the effect of an affirmative FENV_ACCESS pragma: #pragma STDC FENV_ACCESS ON If the FENV_ACCESS pragma is placed outside of any top-level declarations in a file, the pragma will apply to all declarations in the compilation following the pragma until another FENV_ACCESS pragma is encountered or until the end of the file is reached. If the FENV_ACCESS pragma is placed at the beginning of a block (compound statement), the pragma will apply until another FENV_ACCESS pragma is encountered or until the end of the block is reached. For PA-RISC, you might need to use the +Onomoveflops compiler option in order to prevent optimizations that can undermine the specified behavior of this function. Link in the math library by specifying -lm on the compiler or linker command line. For more information, see the HP-UX floating-point guide for HP Integrity servers at the following site: http://www.hp.com/go/fp. |
Printable version | ||
|