Floating point problems (accuracy).

Joseph S. Myers joseph@codesourcery.com
Thu Feb 10 12:24:00 GMT 2005


On Wed, 9 Feb 2005, Vincent Lefevre wrote:

> There are no reasons not to fix it. For those who fear a performance

The x86 back-end needs changing so it no longer pretends that x87 floating 
point operates other than on "long double".  The front ends need to model 
excess precision explicitly so that they can handle an expression of type 
"double" being represented as long double and a cast from double to double 
removing excess precision; even -ffloat-store only deals with assignment, 
not casts, and I don't know whether precision will be discarded properly 
by function call and return either.  Then the front ends need some way of 
representing the present vague semantics for -ffast-math when 
down-conversions to double on assignment aren't required.

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    joseph@codesourcery.com (CodeSourcery mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)



More information about the Gcc-help mailing list