We are happy to announce the availability of Arm Allinea Studio 18.2 version with key new enhancements in Compiler and Libraries.
In 18.2, license management is enabled by default for Arm Compiler. You will need to add a valid license file to run the Arm Compiler and Libraries. You can use named users only license on a single node, or you can use Arm Licence Server to implement floating licenses for multiple users. For more details about licensing, refer to the Arm Allinea Studio licensing page. If you need help with licensing, our team of Arm experts are there to support you - Get support.
For Arm Performance Libraries, one of the main areas of improvement is the performance of FFT routines. We have focused on the FFTW interface, and we are recommending going forward that all users should use this API to get the best performance.
In this release, particular attention has been placed on complex-to-complex transforms in both single and double precision. The underlying algorithms have been completely overhauled to provide much higher performing implementations than we had provided previously. We have paid particular attention to the lengths of FFT cases that we have been seeing used in real HPC cases, where real users have been collecting data using perf-libs-tools logging library. This means that we have high performing implementations on multiples of all primes up to 19, which encompasses the vast majority of FFT usage we have seen. This can be seen in the below graph where our double precision cases are compared against FFTW's 3.3.7 release on Arm as well. Each point represents the relative advantage of using Arm Performance Libraries for FFTs of that length, where being at 1 would represent parity with FFTW, and higher is better.
These optimizations will be seen in the 18.2 release by users using the FFTW 1-d interface. This is, therefore, our foundational work on which many further FFT optimizations will be built in the following releases. In particular, we will be extending these optimizations to be evident from the advanced and multi-dimensional interfaces.
Array processing is an important feature of Fortran HPC applications. Arrays can be passed to subroutines or functions in various forms. In previous releases of the Arm HPC compiler, debugging of arrays was limited to fixed length arrays. This release enables debugging of different kinds of Fortran array arguments, such as adjustable arrays, assumed size arrays and assumed shape arrays.
Modules are used in Fortran for packaging functions and data so that they can be shared or reused by different program units. The 18.0 release supported debug of functions/subroutines inside a module scope, and 18.1 extended support for module variables. This release adds preliminary support for import of modules. Specifically, imported module variables are visible to the debugger at the scope where they are used.
Porting and tuning guides on the Arm Developer website provide instructions on how to build and run popular HPC and scientific workloads with Arm Compilers. These guides enable a smooth transition when porting HPC applications to the Armv8-A architecture.
We are excited to announce the availability of Arm Allinea Studio 18.2 with FFT and debug improvements. Please get in touch to request a trial or buy a license. We plan to provide the next release 18.3 towards the end of May 2018 with more features and improvements.
Release history More about Arm HPC tools.