We’re glad to announce the release of Parallelware Trainer 1.6 which features a significant overhaul of how you interact with parallelization opportunities. We’ve also extended the GPU offloading support by allowing the copy of multi-dimensional arrays using OpenMP as well as through new GPU-specific defects and recommendations. In total, 6 new defects and 8 new recommendations are now available along with the latest version of the underlying Parallelware technology.
Parallelization opportunity redesign.
Until now, parallelization opportunities took the visual cue of a green circle. When a loop could not be analyzed, this was shown as a red circle which you could click to get insights into what is preventing the analysis from succeeding. However, there were missing cases in between: when, while a loop was successfully analyzed, it did not constitute a parallelization opportunity supported by Parallelware Trainer. To address such cases, this release introduces two new circles: orange circles will be shown when the opportunity analysis is missing some information; blue circles will be shown for loops that can not be automatically parallelized by Parallelware Trainer. In both cases, you can click the circle to see the exact causes along with a detailed data scoping report.
Multi-dimensional array copy to the GPU.
OpenMP enter/exit data statements are now supported to perform deep-copy of non-contiguous multidimensional arrays when offloading to the GPU.
New defects and recommendations.
The catalog of defects and recommendations has been expanded with 6 new defects and 8 recommendations. Of these, 2 new defects and 3 recommendations are specific to GPU offloading.
Latest Parallelware technology.
Each new release includes the latest version of the Parallelware core technology which is in constant evolution to support more code bases and parallelization features.
More info:
- Customizing Parallelware Trainer through environment variables
- Support resources
- Parallelware Trainer help sheet
- Appentra Knowledge Base – Checks
Try Parallelware Trainer for free today
Previous release notes:
Leave a Reply