Opteron speed

[John Edward Scott]

> 
> gmp-list at daedalus-solutions.co.uk writes:
> >
> > I was just wondering if anyone had a (rough) ETA for a release with
> > Opteron optimizations?
> 
> Perhaps a couple of months.
> 
> > The reason I ask is that we have a very intensive modelling program
> > which we've just switched over to using GMP (rather than calls to
> > sinl, sqrtl etc) and have noticed a *huge* speed decrease,
> 
> All the various overheads make a slowdown fairly inevitable on
> smallish operands.  If you only need quads or something then there
> might be a fixed-precision library that could go faster.
> 
> > presumably due to the fact that GMP is being built as "generic".
> 
> It won't be helping :).
> 

Just as a follow-up, I did some profiling of my app and the results
are below -

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total
 time   seconds   seconds    calls  ms/call  ms/call  name
 20.53     20.19    20.19 341823446     0.00     0.00  mpfr_mul
 19.78     39.64    19.45 550784802     0.00     0.00  mpfr_round_raw_generic
 10.85     50.31    10.67 120318096     0.00     0.00  mpfr_sub1
  8.69     58.86     8.55 24588477     0.00     0.00  spose::fill(lattice const&)
  6.30     65.06     6.20      639     9.70     9.70  mpfr_add1
  4.91     69.89     4.83 117175960     0.00     0.00  mpfr_add
  3.67     73.50     3.61 500682285     0.00     0.00  mpfr_check_range
  3.08     76.53     3.03 707753567     0.00     0.00  mpfr_set_si
  2.96     79.44     2.91 120318096     0.00     0.00  mpfr_cmp2
  2.41     81.81     2.37 512970801     0.00     0.00  mpfr_add_ui

So, given that my app has a runtime of *months*...any Opteron
optimisation in mpfr_mul or mpfr_round_raw_generic would *greatly*
help.

I know a *little* of Assembler, but I really don't want to re-invent
the wheel if some work is currently going on in this area. I'm also
happy to run Alpha code if it's in a state to be tested, otherwise
I guess I'll just have to give up.

Thanks,

John.