d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Oops, r336412 undid the fix of the overflow threshold in r323003. Restore

Browse Source 
the previous overflow threshold and adjust comments.
This commit is contained in:
Bruce Evans 2018-07-19 14:24:48 +00:00
parent 17e98da88e
commit 50c8bd4e53
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=336484
1 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
lib/msun/src/s_csqrtl.c
View File

@ -36,17 +36,22 @@ __FBSDID("$FreeBSD$");
#include "math_private.h"
/*
* THRESH is now calculated portably (up to 113-bit precision). However,
* the denormal threshold is hard-coded for a 15-bit exponent with the usual
* bias. s_logl.c and e_hypotl have less hard-coding but end up requiring
* the same for the exponent and more for the mantissa.
* Several thresholds require a 15-bit exponent and also the usual bias.
* s_logl.c and e_hypotl have less hard-coding but end up requiring the
* same for the exponent and more for the mantissa.
*/
#if LDBL_MAX_EXP != 0x4000
#error "Unsupported long double format"
#endif
/* For avoiding overflow for components >= LDBL_MAX / (1 + sqrt(2)). */
#define THRESH (LDBL_MAX / 2.414213562373095048801688724209698L)
/*
* Overflow must be avoided for components >= LDBL_MAX / (1 + sqrt(2)).
* The precise threshold is nontrivial to determine and spell, so use a
* lower threshold of approximaely LDBL_MAX / 4, and don't use LDBL_MAX
* to spell this since LDBL_MAX is broken on i386 (it overflows in 53-bit
* precision).
*/
#define THRESH 0x1p16382L
long double complex
csqrtl(long double complex z)