"Roger Stafford" <ellieandrogerxyzzy@mindspring.com.invalid> wrote in message <g4hk75$jd$1@fred.mathworks.com>... "Ned A" <ned456@hotmail.com> wrote in message <g4e1pb$2qt $1@fred.mathworks.com>... ....... When I add q+r+s and compare it to s+q+r, I get a different result for about 20%% of values of q,r,s. ........ Ned, I stated that matlab could get different

Michael Hofmann OK, ich glaube es Dir, wenn Du mir noch zeigst, dass m = sqrt(2)-1 rational ist. sqrt(2) ist ein Kettenbruch und somit rational. Beweis: sqrt(2) = 1 + 1/(2 + 1/(2 + 1/(2 + 1/(2 + 1/(2 + 1/(2 + ...)))))) Ausgeschrieben: .. Wurzel(2) = 1.4142135623730950 0: 1/1 = 1.0000000000000000 err = 4.14e-001 1: 3/2

And here is the same code in unit/module form with tables commented out: The possible table lookups/optimizations are left in in case I or any one else wants to test those ;) // *** Begin of Module Code *** unit UnitStringToUint64_version_003; interface function TryStrToUint64( S : string; var Value : Uint64 ) : boolean; implementation { Skybuck's TryStrToUint64 conversion

Hello, Sometimes I google for my own code and then it sucks if I one can't find it or too it's incomplete. So one more time the complete code as it now is, all 3 versions embedded. Version 0.03 is now active... big 1600 bytes lookup tables still active but those could be commented out. Maybe somebody else could write a nice benchmark program for it ? Tomorrow maybe I write one, but

David Marcus wrote: David Marcus wrote: There is a well-defined mathematical function from real numbers to real numbers that rounds numbers to a given number of digits. There is no reason why a computer can't implement the restriction of this function to the set of binary floating point numbers. My point was that implementing such a function by rescaling is unlikely