Does HP Fortran not continue to support the etime() function when compiled and linked with -U77? Omitting cpu_time is a serious
omission, but I believe they did not intend to make f90 work correctly under HPUX10.20.
The intrinsic SUM() no doubt uses a different order of operations from those which you would expect from your source code. The
differences shown are to be expected in single precision.
----- Original Message -----
From: "Jürgen v.Hagen" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Tuesday, June 06, 2000 5:00 AM
Subject: Re: Results for Friedrich Hertweck's little timing test
> HP-UX 10.20, f90 1.0, C240 HPPA 2.0
>
> has anybody a reliable subroutine for CPU-time measurements
> for the HP-UC f90 compiler? I attach the code I used (C-wrapper
> for mytime...). CPU_TIME is not part of the f90 HP-UX compiler.
>
> as expected, the compiler is able to do a good job for loops.
> Puzzling for me is, that the SUM() call is much faster than the
> others, and also that the result is different.
>
> I also included a BLAS call to sasum like in the code below.
> Run times are in the range of the f77 loops.
>
> juergen
>
> 1.) -g
> f77 loop, contiguous array, sum =32811.492188 time = 9.620000
> f77 loop, stride 2x2 array, sum =32811.492188 time = 11.590000
> f90 loop, contiguous array, sum =32811.492188 time = 18.459999
> f90 loop, stride 2x2 array, sum =32811.492188 time = 19.039997
> f90 SUM(stride 2x2 array), sum =32811.437500 time = 0.700001
>
>
> 2.)
> +O3 +Odataprefetch +Ofastaccess +Onoinitcheck +Oloop_unroll \
> +Onofltacc +FPD +Oregionsched +Oentrysched
> f77 loop, contiguous array, sum =32811.492188 time = 0.850000
> f77 loop, stride 2x2 array, sum =32811.492188 time = 1.020000
> f90 loop, contiguous array, sum =32811.492188 time = 1.340000
> f90 loop, stride 2x2 array, sum =32811.492188 time = 1.010000
> f90 SUM(stride 2x2 array), sum =32811.437500 time = 0.700000
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|