This is the mail archive of the
mailing list for the GCC project.
Re: real.c on unicosmk
- From: dewar at gnat dot com (Robert Dewar)
- To: rth at redhat dot com, steve at moshier dot net
- Cc: Joe dot Buck at synopsys dot com, gcc at gcc dot gnu dot org, rl at cs dot tu-berlin dot de,zack at codesourcery dot com
- Date: Thu, 23 May 2002 10:20:37 -0400 (EDT)
- Subject: Re: real.c on unicosmk
> For counterexample, the sun2 and sun3 assemblers do not accept
> that for immediate double operands; you have to give them a decimal
> string. There are other counterexamples also, but this
> conversation is now making me too depressed to continue.
We have found in the GNAT technology that relying on the assembler to do
proper nearest bit generation of fpt values, including properly rounded
denormals etc, is much too unreliable. Unlike C, Ada requires as part
of the language definition that such things be done exactly.