On Sat, 23 Aug 2008 10:32:46 +1200
Ian Collins <[EMAIL PROTECTED]> wrote:
> Mike Meyer wrote:
> > On Fri, 22 Aug 2008 16:49:25 +1200
> > Ian Collins <[EMAIL PROTECTED]> wrote:
> >
> >
> >> Mike Meyer wrote:
> >>
> >>> On Thu, 21 Aug 2008 18:59:24 +1200
> >>> Ian Collins <[EMAIL PROTECTED]> wrote:
> >>>
> >>>> Alfred Monticello wrote:
> >>>>
> >>>>> The one in /opt/sfw can produce 64bit code but is a 32bit binary.
> >>>>>
> >>>> What's wrong with that?
> >>>> I can't see any reason for wanting a 64 bit compiler binary, unless you
> >>>> have a file which takes more the 4GB or RAM to compile..
> >>>>
> >>> Is the compiler that built the compiler really so bad that having
> >>> twice as many general-purpose registers available doesn't make any
> >>> difference in performance? Or do you simply not care how long it takes
> >>> to compile things?
> >>>
> >> If life was that simple, everything would be built 64 bit on Solaris.
> >> But it isn't. Some applications run slower, some faster.
> >>
> >> What were your benchmark results when you compared 32 and 64 bit
> >> compiler binaries?
> >>
> > My benchmarks with gcc 4 showed an that 64 bit code ran an average of
> > about 33% faster than 32 bit code.
> By code, is that generated code or the compiler its self?
Yes. Both simple benchmarks and compilation times for medium-sized
code bases (compile times measured in single-digit hours) with 32 & 64
bit compilers.
> > Once you get beyond simple
> > benchmarks to real-world applications, you get into interesting
> > questions like whether you should compare your 32 bit compiler running
> > on a 64 or a 32 bit kernel (I went with the latter because the systems
> > I was benchmarking came with compilers that matched the kernel, and
> > the results matched the rest of the benchmarks).
> I'm surprised there would be any difference. The same application code
> is running on either kernel.
I was interested in total CPU time, which included system time. I
suppose I could have just checked user time, but I was (and am) more
interested in the behavior of the system as a whole.
> > What benchmarks did you get that show there's really no need for a 64
> > bit build of the compiler on OpenSolaris?
> I don't have any, but the Studio developers have commented in the past
> that they don't see any advantages in building a 64 bit version of their
> compilers.
Well, if you can rebuild your entire package in less time than it
takes to get a cup of coffee, then I don't suppose there is any
advantage :-).
> I guess I should have a try with gcc. My concern was that a 64 bit
> build would require a larger working memory set for 64 bit pointers,
> which would result in more cache thrashing than a 32 bit build.
I haven't looked at the cache hit behavior in a *long* time (since
like v7), so maybe I should just stop there.
<mike
--
Mike Meyer <[EMAIL PROTECTED]> http://www.mired.org/consulting.html
Independent Network/Unix/Perforce consultant, email for more information.
O< ascii ribbon campaign - stop html mail - www.asciiribbon.org
_______________________________________________
opensolaris-discuss mailing list
[email protected]
