[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Gnucap-devel] faster queueing
From: |
al davis |
Subject: |
Re: [Gnucap-devel] faster queueing |
Date: |
Fri, 26 Feb 2016 12:06:41 -0500 |
On Wed, 24 Feb 2016 11:52:09 +0100
Felix Salfelder <address@hidden> wrote:
> On Wed, Feb 24, 2016 at 09:07:48AM +0100, Felix Salfelder wrote:
> > indeed, in -uf there's a "precalc_last()" call after setup(cmd) in
> > s__init.cc... iirc due to some issues with temperature dependency. need
> > to recheck...
>
> fixed and backported the fix. see the precalc_last branch on savannah.
Do you have some test cases that demonstrate the problem, and that it
is fixed?
Doing a code walk-through .. I see the problem, but this patch doesn't
fix it. In some cases, temperature is handled in precalc_first().
I think a better change to the code is to not use the flags, but
instead move precalc_last() to always happen after setup().
Then also revisit the decisions about what is in precalc_first() vs
precalc_last().
From the wiki ...
>>>>>Some values must be calculated before determining structure. These
>>>>>go in precalc_first.
>>>>>Some values must be calculated after structure is determined.
>>>>>These go in precalc_last.
>>>>>Some values don't matter whether they are calculated before or
>>>>>after structure is determined. It strictly doesn't matter whether
>>>>>they are in precalc_first or precalc_last. By convention,
>>>>>evaluation of parameter expressions goes in precalc_first,
>>>>>calculation of other values that are done only once goes in
>>>>>precalc_last.
The choice "by convention" is probably wrong. It probably should be
that unless the calculations are needed before expand() and setup(), it
they should be in precalc_last(). A lot of devices need to change.