This is the mail archive of the
glibc-bugs@sourceware.org
mailing list for the glibc project.
[Bug math/15396] Vast inaccuracies of casin/casinh/cacos/cacosh with IBM long double
- From: "jsm28 at gcc dot gnu.org" <sourceware-bugzilla at sourceware dot org>
- To: glibc-bugs at sourceware dot org
- Date: Thu, 09 May 2013 21:59:07 +0000
- Subject: [Bug math/15396] Vast inaccuracies of casin/casinh/cacos/cacosh with IBM long double
- Auto-submitted: auto-generated
- References: <bug-15396-131 at http dot sourceware dot org/bugzilla/>
http://sourceware.org/bugzilla/show_bug.cgi?id=15396
Joseph Myers <jsm28 at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |SUSPENDED
--- Comment #1 from Joseph Myers <jsm28 at gcc dot gnu.org> 2013-05-09 21:59:07 UTC ---
I investigated the Real part of: casinh (0x1.fp-1025 +
0x0.ffffffffffffffffffffffffffcp0 i) error. It appears to trace back to a bug
in libgcc's __gcc_qdiv when working with subnormal values. If I compute
0x1p-1024L / 0x3p-53L, I get a long double with high part
0x1.5555555555555p-973 and low part 0 (a big loss of precision). If instead I
compute 1 / 0x3p-53L and then multiply by 0x1p-1024L, I get a long double with
high part 0x1.5555555555555p-973 and low part 0x0.0aaaaaaaaaaabp-1022.
So libgcc will need fixing (preferably on all active GCC release branches as
well as trunk) and then this bug will need revisiting to see if that explains
all the failures or only some. Given the dependence on a GCC fix, marking
SUSPENDED until that fix is done.
--
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.