Re: [PATCH] PR fortran/31593: Speed up loops with DO variables as procedure arguments

[Tobias Schlüter <Tobias dot Schlueter at physik dot uni-muenchen dot de>]

Hi Dominique,

Dominique Dhumieres wrote:
> Here are the timings for capacita.f90 (unpatched, patched, and without
> the call to gfc_trans_simple_do:
>
> [ibook-dhum] lin/test% gfc -m64 -O3 -ffast-math -funroll-loops -fomit-frame-pointer capacita.f90
> [ibook-dhum] lin/test% time a.out > /dev/null
> 49.467u 1.360s 0:51.08 99.4%	0+0k 0+0io 39pf+0w
> [ibook-dhum] lin/test% time a.out > /dev/null
> 49.724u 1.348s 0:51.23 99.6%	0+0k 0+0io 0pf+0w
> [ibook-dhum] lin/test% /Volumes/MacBook/opt/gcc/gcc4.5w-0/bin/gfortran  -m64 -O3 -ffast-math -funroll-loops -fomit-frame-pointer capacita.f90
> [ibook-dhum] lin/test% time a.out > /dev/null
> 53.174u 1.380s 0:54.74 99.6%	0+0k 0+0io 0pf+0w
> [ibook-dhum] lin/test% time a.out > /dev/null
> 53.273u 1.377s 0:54.83 99.6%	0+0k 0+0io 0pf+0w
> [ibook-dhum] lin/test% /Volumes/MacBook/opt/gcc/gcc4.5w-do/bin/gfortran -m64 -O3 -ffast-math -funroll-loops -fomit-frame-pointer capacita.f90
> [ibook-dhum] lin/test% time a.out > /dev/null
> 48.640u 1.356s 0:50.16 99.6%	0+0k 0+0io 0pf+0w
> [ibook-dhum] lin/test% time a.out > /dev/null
> 48.732u 1.366s 0:50.26 99.6%	0+0k 0+0io 0pf+0w
>
> so it is ~8% slower with your patch and ~2% faster without
> the call to gfc_trans_simple_do.

The other way, gcc4.5w-0 was with the 'if (0 && ...' according to your 
other mail.  I didn't expect the 2% speedup, but as we saw yesterday, 
escaping pointers can have very global effects.  The 'if (0 &&' still 
lets pointers to loop variables escape, so any effects of my patch that 
are not local to the affected loop are not to be expected.  That said, 
I've yet to understand why capacita.f90 loses so badly, its inner loop 
isn't complicated and therefore the optimizers should be able to handle 
it well even without gfc_trans_simple_do.

I'll revise my patch along the lines RTH suggested, and resubmit.  The 
problem with air.f90 definitely lies somewhere else, and out of reach 
for me.

Cheers,
- Tobi
--
Tobias Schlüter
Am Coulombwall 1, Zi. 326
85748 Garching b. München
Tel.: +49/89/289-14139