This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: IEEE 754 and fused-madd (was: Re: patch: rs6000 specific)

On Thursday, December 6, 2001, at 01:41 PM, David Edelsohn wrote:

> 	The PowerPC port of GCC does use FMA by default unless
> -mno-fused-madd is invoked.
> 	What Dale and others at Apple seemed to be requesting was that GCC
> perform some of the -ffast-math transformations by default by using
> additional combiner patterns.  Their argument seems to be that use of FMA
> already is non-compliant, so anything is fair game.
> 	I believe that if GCC normally only performs those transformations
> with -ffast-math, then both the transformations should not be duplicated
> in the MD file and GCC should not perform those transformations by
> default.
> 	I am not an expert in this field.  I am glad that FMA may be
> compliant with IEEE 754.  My belief is that basic use of FMA matching the
> original expression is different than use of FMA with implicit expression
> transformations.

I'm not an expert either.  I have an e-mail in to somebody who is, but
he hasn't gotten back to me yet.  It seems clear this patch isn't going
to be accepted as is, so I'll look at it further.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]