[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [Gcl-devel] Re: [Maxima] Re: m68k gcl/maxima: Minor Floating point errors



Greetings!  I think its a matter of when the registers are written to
memory.  Apparently the m68k can optimize in such a way that many
intermediate results are held in registers before having to be written
out.  There is no -ffloat-store on x86.  The difference is perhaps
related to the paucity of registers on x86, and the fact that m68k
registers are *96* bits wide, AFAICR, instead of 80.  But in general,
one does get more precision on x86 FPU calculations than the strictly
64bit SSE2, for example.

Take care,

Raymond Toy <toy@rtp.ericsson.se> writes:

> >>>>> "Camm" == Camm Maguire <camm@enhanced.com> writes:
> 
>     Camm> Greetings!
>     Camm> Rick Younie <rick@def.debian.net> writes:
> 
>     >> Camm Maguire wrote:
>     >> > Greetings!  I have one Debian machine (m68k) which is producing very
>     >> > small numerical discrepancies on the results of the two floating point
>     >> > intensive tests in rtest8.mac:
>     >> 
>     >> Hi,
>     >> 
>     >> That rang a bell.  There's this in the gcc man page:
>     >> 
>     >> -ffloat-store
>     >> Do  not  store floating point variables in registers.  This pre-
>     >> vents undesirable excess precision on machines such as the 68000
>     >> where  the floating registers (of the 68881) keep more precision
>     >> than a double is supposed to have.
>     >> 
>     >> For most programs, the excess precision does only  good,  but  a
>     >> few  programs  rely  on  the precise definition of IEEE floating
>     >> point.  Use `-ffloat-store' for such programs.
> 
>     Camm> That did it -- thanks!  As Richard Fateman observed, its quite
>     Camm> surprising the magnitude of the difference was so large.  Apparently
>     Camm> m68k has the more accurate answer, right?  In any case, I'm adding
>     Camm> -ffloat-store to gcl's compile flags on m68k to avoid testing
>     Camm> problems. 
> 
> But does gcl on x86 also do a -ffloat-store?  x86 also keeps extra
> precision, so I would have expected similar issues on x86.
> 
> Ray
> 
> 
> _______________________________________________
> Gcl-devel mailing list
> Gcl-devel@gnu.org
> http://mail.gnu.org/mailman/listinfo/gcl-devel
> 
> 

-- 
Camm Maguire			     			camm@enhanced.com
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah



Reply to: