Re: GNAT compiler switches and optimization

[Stephen Leake <stephen_leake@stephe-leake.org>]

Duncan Sands <duncan.sands@math.u-psud.fr> writes:

> On Friday 20 October 2006 12:47, tkrauss wrote:
>> I'm a bit stuck trying to figure out how to coax more performance
>> out of some Ada code.  I suspect there is something simple (like
>> compiler switches) but I'm missing it.  As an example I'm using
>> a simple matrix multiply and comparing it to similar code in
>> Fortran.  Unfortunately the Ada code takes 3-4 times as long.
>
> GNAT GPL 2006 is based on gcc 3.4.6.  For fortran you are using
> gcc 4.2.0.  Try using comparable compiler versions, eg: an Ada
> aware gcc 4.2.0 (several linux distributions provide this) or
> a gcc 3.4.6 version of fortran (i.e. some version of g77).


I have a pre-release of GNAT 5.05, which is based on gcc 4.2. I tried
tst_array with both -O2 and -O3; -O2 was faster. AdaCore says -O3 has
"dangerous experimental optimizations; don't use it". Here's a
comparison:

GNAT 5.04, -O2 -gnatp:

./tst_array.exe 800
Init Time:           0.557314686

Mult Time:          20.139092538

 1.99202E+02 1.96854E+02
 1.88844E+02 1.84498E+02


GNAT 5.05, -O2 -gnatp:

./tst_array.exe 800
Init Time:           0.575690486

Mult Time:          12.160329316

 1.99202E+02 1.96854E+02
 1.88844E+02 1.84498E+02


Note that I timed the initialize with random separate from the matrix
multiply; it could easily be that the random number generators are
different. But that's a small effect.

I don't have gfortran installed.

-- 
-- Stephe