Reply-To: From: "Arthur" To: "DJGPP Mailing List" Subject: RE: Floating/fixed point Date: Sun, 13 Sep 1998 14:15:45 +0100 Message-ID: <000101bddf18$9db5fa00$d54b08c3@arthur> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit In-Reply-To: <35FA9D74.E138AD72@unb.ca> Importance: Normal Precedence: bulk > Arthur wrote: > > Although floating point math is just as fast as integer math on > a Pentium, > > it sucks on anything less than a Pentium, on any AMD processor > (apart from > > the K6-2) and any Cyrix processor (apart from the Cyrix MediaGX > chip). Also > > conditionals and integer-float conversion are unfeasibly slow > on any chip. > I don't know where you got your info, but floating point > math is MUCH slower > than int's even on a Pentium. Pentium's do floating point math > much faster > than 486/387 chips did, but they are still slower than integer. > You can't add > 2 floats in 1 clock cycle. Are you sure you're using the FPU? Are you sure you're using a genuine Intel? Besides, the real difference is in multiplications, divisions and trigonomic functions, which are quite a bit faster than integer math on a standard Pentium. On a PII, the FPU is generally the same as a Pentium, but integer math is improved quite a bit over a Pentium. > > However, a 32-bit float (32.32) is much more accurate than a > 32-bit (16.16) > > fixed number. > That is very misleading (or wrong). 32-bit float's (also > known as double's) > aren't 32.32. You make it sound like there are 32 bits before > the decimal and > 32 bits after (9 digits.9 digits). double's are 12 digits of > precision with a > base 10 exponent that can range from +-308. Sorry, I was harking back to my Motorola days where a basic float was 32 bits before the decimal and 32 bits after (encoded in two longwords). But this just goes further to prove my point that floating point has a higher precision than fixed point math. James Arthur jaa AT arfa DOT clara DOT net ICQ#15054819