|
|
Arcane
4.1.12.0
User documentation
|
|
|
|
Arcane
4.1.12.0
User documentation
|
This page describes the tools available in ARCANE to perform bit-by-bit comparisons of the values of variables managed by ARCANE. The following comparisons are possible:
It is important to note that comparison is only possible on variables managed by Arcane (such as #VariableCellReal, #VariableScalarInt32, ...). In the current implementation, only variables with a numeric data type are compared (so, for example, not by the 'String' data types).
This mechanism allows determining the list of Arcane variables that are different between two executions. It only works on Arcane variables.
The operating principle is as follows:
All Arcane variables are compared except those that have the IVariable::PExecutionDepend property set to true. In parallel/sequential comparison, those that also have the IVariable::PSubDomainDepend property are not compared.
To save the reference information, simply run the case after setting the environment variable STDENV_VERIF to the value WRITE. In this case, the values of all variables will be saved at each iteration. It is possible to do this for each entry point by setting the environment variable STDENV_VERIF_ENTRYPOINT, but this greatly increases the volume of information to save.
By default, the information is saved in the directory /tmp/$USER/verif, but it is possible to change this by specifying another path in the environment variable STDENV_VERIF_PATH
/tmp directory.Since version 3.16 of Arcane, it is possible to choose the comparison method used to calculate the difference. This is done via the environment variable STDENV_VERIF_DIFF_METHOD. The possible values are:
Once the reference has been executed, simply set the environment variable STDENV_VERIF to READ and run a new execution. It is possible to change the number of subdomains compared to the reference execution and thus perform comparisons between parallel and sequential (in this case, the sequential execution must be the reference and be executed first) or parallel/parallel comparisons.
In the execution listing, the list of variables that are different between this execution and the reference, along with their values, will then appear for each iteration (or for each entry point if STDENV_VERIF_ENTRYPOINT is defined), as follows:
For each variable and each subdomain, its name, the Item::uniqueId() and the Item::localId() of the entity, as well as whether it is ghost ((G)) or belongs to the subdomain ((O)), the current value (val), the reference value (ref), and the relative difference (rdiff) are indicated. To avoid cluttering the listing, only the 10 most significant differences in absolute value are displayed.
When the variable is an array variable and is not on a mesh entity, (G) or (O) does not appear, and instead of the local entity number, the index of the element in the array is displayed.
In parallel, it may be normal for the values on ghost cells to be different from the reference if the variable is not synchronized. Since this can be the case for many variables, it is possible to display differences only on cells belonging to the subdomain by setting the environment variable STDENV_VERIF_SKIP_GHOSTS.
Just as it is possible to perform bit-by-bit comparisons, it is possible to verify that variables are properly synchronized between subdomains. To do this, simply specify the value CHECKSYNC to the environment variable STDENV_VERIF. Values with the IVariable::PNoNeedSync attribute and partial variables are not compared.
It is also possible to verify that the values of a variable are the same on all replicas of a subdomain. To do this, you must specify the value CHECKREPLICA to the environment variable STDENV_VERIF. Variables with the IVariable::PNoReplicaSync attribute and partial variables are not compared.
1.16.1