Extra-P 4.2.0 Release
Scaling analysis support
- Automatic detection and conversion of strong scaling
- Improved strong scaling support for CUBE files
- Enhanced weak scaling modeling for CUBE profiles of multi-threaded applications
More options for selecting the measure of aggregation
- Modeling the minimum and maximum of the measured point is now possible
New modelers
- Segmented modeling: modeling of two behaviors in one model
- Support for segmented performance models with a single or multiple model parameters.
- Extension of the available plotting features to support segmented models.
- Adaptive modeling: optionally uses a neural network to improve the noise resiliency of the models
Measurement point suggestions based on Gaussian Process Regression
- The GUI features a view for measurement point suggestions.
- Several suggestion approaches help the user to complete or extend the set of available measurement points
to improve model accuracy further and enhance the noise resilience of the model while optimizing modeling cost.
Enhancements
- Functions can be formatted in Python and Latex format
- Function names are shortened in plots
- Keeps track of number of repetitions
- Optionally keeps all measured values in the experiment
Full Changelog: v4.1.2...v4.2.0