David Nickerson wrote:
> Hi Andrew,
>
> Saw this in today's meeting minutes and I just wanted to check how it
> might affect how I'm using the CCGS?
>
Hi Andre,
There are two main changes:
1) The separation of code generation into two steps:
a) generating a DAG of objects representing steps in a procedural
process.
b) DFS visit of the DAG, generating C code as a side effect.
2) Support for optimising numerical Jacobian calculations. This works by
splitting the DAG visit into two steps: firstly visit starting from all
equations used to compute variables directly used by the rate equations
(so the DFS will catch all procedural steps needed to compute variables
needed to compute the rates), and secondly (wrapped in #ifndef
VARIABLES_FOR_RATES_ONLY) visits all remaining equations.
Change 1 should not affect the output of the functions that get
generated (it may change the ordering of assignments, but only to the
extent that ordering does not matter, so you will still get the exact
same results).
Change 2 only makes any difference at all if you define
VARIABLES_FOR_RATES_ONLY in your variableCodeFragment. Since this will
not normally be defined, you don't have to do anything as a result of my
changes.
However, if you wanted, you could use my changes to get a more efficient
Jacobian function, or to avoid computing variables that you don't need
(if you define VARIABLES_FOR_RATES_ONLY, the code will not evaluate
variables which do not directly affect the time evolution of your model.
When computing the Jacobian, this is obviously a good idea, as it saves
computing variables which will just be thrown away unused anyway. You
might also want to do this for intermediate steps of the evaluation if
you are going to look at the final result). The CIS utilises this by
using variableCodeFragment twice, once with VARIABLES_FOR_RATES_ONLY
defined, and once with it not defined.
Best regards,
Andrew
_______________________________________________
cellml-discussion mailing list
[email protected]
http://www.cellml.org/mailman/listinfo/cellml-discussion
