Mailinglist Archive: opensuse-programming (27 mails)

[Jerry Feldman <gaf@xxxxxxx>]

On 01/01/2010 01:27 PM, Per Jessen wrote:
> Ah, yes, I agree (of course) - I thought you meant just wrapping the
> assembler code in C-functions, which didn't seem to give me much. 
>
> A complete rewrite in C is not an option - the vast majority of this
> code does math and such, and is highly optimized/specialized/tweaked
> (no pipeline stalls etc). 
On 01/02/2010 04:11 AM, Per Jessen wrote:
> Petr Baudis wrote:
>
>   
> > On Fri, Jan 01, 2010 at 07:27:49PM +0100, Per Jessen wrote:
> >     
> > > A complete rewrite in C is not an option - the vast majority of this
> > > code does math and such, and is highly optimized/specialized/tweaked
> > > (no pipeline stalls etc).
> > >       
> > Are you _sure_ that what was highly optimized in pentium times will
> >     
> It has been kept up-to-date with modern architectures - we've stuck to
> tasm for as long as possible, and also 'enhanced' it with a few macros
> for support of newer instructions. 
>   
For the x86 series that strategy is ok, but you are running into dead
ends. Hopefully, yasm will fill your needs.
I would actually consider using FORTRAN. FORTRAN compilers today provide
very good optimizations where C is more difficult to optimize.


One possible test is to take a reletively small module, recode it in
FORTRAN and C, aggressively optimize it and run some benchmark tests.
While GCC is a good compiler, other commercial compilers, like Intel,
take the same factors into account, such as latency and pipeline stalls.

-- 
Jerry Feldman <gaf@xxxxxxx>
Boston Linux and Unix
PGP key id: 537C5846
PGP Key fingerprint: 3D1B 8377 A3C0 A5F2 ECBB  CA3B 4607 4319 537C 5846