Summary: | Inaccuracies in tgammaf | ||
---|---|---|---|
Product: | glibc | Reporter: | Thomas Koenig <tkoenig> |
Component: | math | Assignee: | Not yet assigned to anyone <unassigned> |
Status: | RESOLVED FIXED | ||
Severity: | normal | CC: | glibc-bugs |
Priority: | P2 | Flags: | fweimer:
security-
|
Version: | unspecified | ||
Target Milestone: | --- | ||
Host: | Target: | ||
Build: | Last reconfirmed: |
Description
Thomas Koenig
2007-10-10 18:48:12 UTC
Two possible implementations at http://gcc.gnu.org/ml/fortran/2007-10/msg00197.html http://gcc.gnu.org/ml/fortran/2007-10/msg00201.html There now is a C implementaton of the [tl]gamma* functions in libgfortran, contributed by FX Coudert: http://gcc.gnu.org/viewcvs/trunk/libgfortran/intrinsics/c99_functions.c?r1=128648&r2=130245 This might be OK for inclusion in glibc. You didn't mention what architecture this was on. Anyway, confirmed with current sources on x86, and with more and greater errors on x86_64 (I suppose excess precision is helping a bit on x86). Presumably the same cause as other tgamma bugs - all five versions (flt-32, dbl-64, ldbl-96, ldbl-128, ldbl-128ibm) follow the same approach of exp (lgamma) which is problematic for results with large (positive or negative) exponent and needs better implementations for all five cases. |