From e5ec494194c5585e2736354fd0bf284db0ba729b Mon Sep 17 00:00:00 2001 From: enge Date: Fri, 2 Sep 2011 12:41:35 +0000 Subject: norm.c: rewritten to handle all cases of over- and underflow and to obtain a correct ternary return value norm.dat: new tests with over- and underflow mpc-impl.h: added macro MPC_MIN BUGS, TODO, mpc.texi updated to reflect these changes git-svn-id: svn://scm.gforge.inria.fr/svn/mpc/trunk@1071 211d60ee-9f03-0410-a15a-8952a2c7a4e4 --- BUGS | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'BUGS') diff --git a/BUGS b/BUGS index 8c74438..14ef613 100644 --- a/BUGS +++ b/BUGS @@ -1,8 +1,7 @@ - Overflow and underflow are not considered in all functions, which might lead to some infinite loops. -- Except for mpc_mul, - no checks are made for intermediate over- or underflows, which may occur - in the middle of the algorithms although the final result may be +- Also, checks are not always made for intermediate over- or underflows, which + may occur in the middle of the algorithms although the final result may be representable. For instance, if in the computation of Im (cos(x+i*y)) = -sin(x)*sinh(y) an overflow occurs in sinh(y), the value sign(Im (cos(x+i*y))) * inf @@ -10,3 +9,7 @@ sin(x)*sinh(y) is representable. If furthermore an underflow occurred in sin(x) (which has not been observed in practice), then the return value would be NaN*(+-inf)=NaN. + +Currently, the following functions have been rewritten to solve these +problems: +mul, norm -- cgit v1.2.1