allinea - Now part of ARM

What's new in the version 7 release of the Allinea toolset

Wed February 8, 2017 by Mark O'Connor, Director Product Management, Server & HPC Tools, ARM
Image shows breakdown of memory usage for HPC code

Version 7 update Webinar 

View slides

Last summer we announced Knights Landing support in Allinea Forge and Performance Reports and CUDA support on the OpenPOWER platform. 

Multi-platform support has always been a key strength of our tools and we remain as committed today as we have ever been. Our 7.0 release is a prime example - we've added support for IBM's Spectrum MPI for little-endian OpenPOWER architectures and further memory debugging enhancements for High-Bandwidth Memory (HBM) available with Intel Xeon Phi processors.

The thing that's exciting most users is the introduction of PAPI counters for Allinea MAP.  We always believed in showing simple, easy-to-interpret metrics in MAP and allowing users to dive down more deeply when they find something interesting.

With the addition of several carefully-selected sets of PAPI counters you can now explore memory bandwidth utilization, vectorization, branch prediction and more - without having to be an expert in CPU architectures or to pick from a list of a hundred conflicting possibilities.

Naturally, this is also supported on the POWER8 and ARMv8 architectures - and we have exciting news about NVIDIA CUDA coming soon!

Many users have expressed interest in MAP's custom metrics feature, allowing anyone to add their own system - or application-specific counters to MAP. With the 7.0 release we've further extended this to allow custom Performance Report sections to be added with just a handful of lines of XML. 

All this is available as part of our Advanced Metrics Pack, which combines custom metrics, energy metrics, PAPI metrics and the new Lustre metadata performance metrics. Existing users of the Energy Pack naturally be upgraded to benefit from all these new features as standard.

The following screenshot of MAP shows the new Lustre metrics in action. The profile was generated as part of the SAGE Horizon2020 research project in which we are involved, and has helped to identify causes for bottlenecks on large production I/O systems:

Shows Lustre metrics in Allinea Forge Profiler for HPC code 

We've also moved to providing monthly feature preview releases in addition to our monthly patch releases.

These contain new features that are then rolled up into our major long-term support releases each year. Our first feature release will be coming at the end of this month, adding support for PAPI counters to Performance Reports.

We also have a range of exciting improvements coming to DDT. Look out for more news on this soon in our newsletter! - sign up.

Not a user? Take a one week software trial 

Need a user upgrade? Request more details.