We are pleased to announce the availability of Parallelware Analyzer 0.17 which includes many new exciting features. New vectorization support introduces detection of opportunities for vectorization and the generation of compiler-specific and portable directives. A lot of effort has also been put into improving usability through a redesigned reporting that reflects the function and loop structure of your code along with more filtering capabilities and increased analysis performance.
- Vector opportunities are now reported for innermost loops that fulfill a strict set of requirements (eg. countable, single entry/exit).
- Generation of OpenMP, GCC, Clang or ICC directives to vectorize loops with the
- Automatically add directives for all vector opportunities in the code with
- Hybrid parallelization support: use
--openmp multi+simdto parallelize nested loops using both multi-threading and SIMD vectorization.
- Defects and recommendations report has been redesigned to reflect the function and loop structure of your code. Additionally, three levels of detail are now available and can be controlled with the
- Loop nesting is very important when considering which loop to parallelize or when analyzing data scoping, memory footprint, etc. This information was missing and is now made available by printing loops as trees in the reports.
- Filtering by function, loop or variable has been extended to all reports.
Other new features
- A new array memory access pattern analysis is available through
- Improved integration with third-party tools thanks to CSV and JSON output formats now being available for all reports.
- Thanks to an extensive redesign, tools are now leaner and analysis time has been reduced by removing some passes.
Follow Parallelware Analyzer’s Quick Start Guide and try the tool on your own code.
Building performance into the code from day one with Codee