[Octopus-devel] buildbot failure in x86_64_ifort_single_build

[Xavier Andrade]

Hi Florian,

What I did was to replace (1) and (3) so nrm2 is always calculated in double
precision and returns real(8). The performance impact is not so large, as
lalg_nrm2 is not used much in comparison with lalg_dot.

Cheers,

Xavier


On Fri, 14 Sep 2007, Florian Lorenzen wrote:

> Hi,
>
> this failure is a bit stupid: we need lalg_nrm2 with the following
> types:
>
> (1) real(4) -> real(4)
> (2) real(8) -> real(8)
> (3) complex(4) -> real(4)
> (4) complex(8) -> real(8)
>
> and, here comes the problem,
>
> (5) real(4) -> real(8)
> (6) complex(4) -> real(8)
>
> The first four are the usual stuff in our BLAS binding, the last two
> are necessary, I suspect, for the single prcision binary that need
> double precision somewhere to converge.
>
> Unfortunately, F90 only does type dispatch on arguments, not on
> results, so we have to decide, if we have (1), (3) or (5), (6) in the
> interface. As (5), (6) are for special cases, I'd opt for the former
> but as I don't know where they are needed, someone else should
> decide. But now, you don;t have to look for the error anymore, at
> least.
>
> Best regards,
>
> Florian
>
> lorenzen at physik.fu-berlin.de writes:
>
>> The Buildbot has detected a failed build of x86_64_ifort_single_build.
>> Full details are available at:
>>  http://www.tddft.org/trac/octopus/tracbb/x86_64_ifort_single_build/builds/55
>>
>> Buildbot URL: http://www.tddft.org/trac/octopus/tracbb/
>>
>> Buildslave for this Build: octopus
>>
>> Build Reason:
>> Build Source Stamp: HEAD
>> Blamelist: acastro
>>
>> BUILD FAILED: failed shell_4
>>
>> sincerely,
>>  -The Buildbot
>>
>> _______________________________________________
>> Octopus-devel mailing list
>> Octopus-devel at tddft.org
>> http://www.tddft.org/mailman/listinfo/octopus-devel
>>
>