Message-Id: <1.5.4.16.19960730075544.206fa44e@vub.vub.ac.be> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Date: Tue, 30 Jul 1996 09:55:44 +0200 To: djgpp-workers AT delorie DOT com From: Pieter Vankeerberghen Subject: Re: Long double support At 12:34 29/7/96 PDT, you wrote: > >How useful is long double to most applications? > >double provides such high precision, 15k of bloat to support something >which is: > not ANSI > not used (by my estimate 99.999% of the time) >I would say if it is needed have a special libldbl.a which has >to be used to handle long doubles. > >I'm of the feeling to ignore it...the application writer should write >special routines to handle long doubles and other non-ansi features. Those who are doing numerical simulations need speed, memory and accuracy. DJGPP delivers speed and memory. Increasing the numerical precision may e.g. reveal possible cancellation errors and therefore be useful for development of the algorithm. If no cancellation errors are present, one may continue to use good old double precision. One should have a look at the increase in code size, both in scanf and printf, after compression. If the code size of the hello world app is important, include in the distribution a hello world.c with puts instead of printf. What we see here is the burden of the run time evaluation of the printf format instead at link time, nothing to be done about it. Imagine DJGPP obtaining higher ratings on Moshier's test bed IEETST than commercial stuff ? DJGPP is not only used for Quake.... Pieter Vankeerberghen ---------------------------------------------------------------------------- Pieter Vankeerberghen tel: +00 32 2 477 43 29 Fabi/ChemoAc, Vrije Universiteit Brussel fax: +00 32 2 477 47 35 Laarbeeklaan 103, 1090 Brussels, Belgium email: pietervk AT vub DOT vub DOT ac DOT be ----------------------------------------------------------------------------