Lahey Support
08-15-2003, 01:24 AM
In a message dated 2/25/2000 6:09:42 AM Pacific Standard Time,
[address removed] writes:
> the two MATMUL and EXP functions take up
> > half of the total CPU time. The two lines with EXP account for 1/3 of
> > the CPU time.
>
> This is probably not help, but sympathy. In my experience, exponentiation
> is very costly in terms of execution time. This has been the case on
> various platforms.
My copies of lf90 and lf95 are quite slow in MATMUL. I note that the version
of Livermore Fortran Kernel 21 which I have in the g77 version in
ftp://members.aol.com/n8tm/lloops.shar.gz is over 4 times as fast as MATMUL
in lf90. I know that lf95 is better; I now have a discarded original copy of
lf95 left over from my job termination, so I could compare on this box, but
I'm fairly sure that you could still write a significantly faster version.
Lahey's in-lined exponentiations, however, are quite fast compared to any x86
alternative.
I didn't catch how you timed your code. For me, the lack of normal profiling
ability on Windows is a serious deficiency in Lahey.
Tim
[address removed]
----------------------------------------------------------
To unsubscribe, send to [address removed] the following
as the first and only line of the message body:
unsubscribe fortran
----------------------------------------------------------
[address removed] writes:
> the two MATMUL and EXP functions take up
> > half of the total CPU time. The two lines with EXP account for 1/3 of
> > the CPU time.
>
> This is probably not help, but sympathy. In my experience, exponentiation
> is very costly in terms of execution time. This has been the case on
> various platforms.
My copies of lf90 and lf95 are quite slow in MATMUL. I note that the version
of Livermore Fortran Kernel 21 which I have in the g77 version in
ftp://members.aol.com/n8tm/lloops.shar.gz is over 4 times as fast as MATMUL
in lf90. I know that lf95 is better; I now have a discarded original copy of
lf95 left over from my job termination, so I could compare on this box, but
I'm fairly sure that you could still write a significantly faster version.
Lahey's in-lined exponentiations, however, are quite fast compared to any x86
alternative.
I didn't catch how you timed your code. For me, the lack of normal profiling
ability on Windows is a serious deficiency in Lahey.
Tim
[address removed]
----------------------------------------------------------
To unsubscribe, send to [address removed] the following
as the first and only line of the message body:
unsubscribe fortran
----------------------------------------------------------