[daip] AIPS text version woes
Eric Greisen
egreisen at nrao.edu
Wed Aug 5 16:59:07 EDT 2015
On 07/30/2015 10:02 AM, Michael Bietenholz wrote:
> What a mess! Given what seems to be the general flakiness of gfortran,
> I wonder if one really should make it all work
> in some particular version of gfortran, since it seems pretty
> likely it will break with the next version.
>
> I'll poke around at HartRAO to see what versions of gfortran
> are around there.
>
> Not respecting the "fno-automatic" seems like a fairly serious
> bug - the existence of the compiler flag suggests that people
> other AIPS use this feature. Unless that were a particular bug
> in 4.4.7 which subsequently got fixed, although you'd think
> there would be reference to that around and likely a fix
> in the redhat version.....
I am now testing my personal version with gfortran (4.8.3 but I think
4.9.2 and 5.1.0 are essentially the same) with the -fno-automatic option
omitted. I had to find several interesting spots in the code - 2 Q
routines today with required SAVEs not fully done. Also really large
buffers are more efficient when SAVEd.
But Y2k took 110 seconds with Intel (LNX64 version), 112 seconds with
gfortran 4.8.3 with automatic variables, and 123 seconds with
-fno-automatic. So something is faster with automatic.
Tomorrow's MNJ will include the latest fixes to this automatic business.
Eric
More information about the Daip
mailing list