Arm is at the heart of the world’s most advanced digital products, its technology enables the creation of new markets and the transformation of industries and society. Arm has a stake in the advancement of digital technology, and so a strong commitment towards leading and partnering in research initiatives that aim to achieve this common goal. Arm Research division from the headquarters in Cambridge (UK) will participate in the project. In the UK offices, 70+ researchers are involved in cutting-edge research activities spanning multi-core architectures, energy-efficient low-power servers, memory systems, machine learning and HPC systems.
In MAX, Arm will bring expertise in computer architectures, in-depth understanding of workload, system-on-chip design space exploration using open-source tools like gem5 or DynamoRIO. Arm has been working for quite some time to couple DynamoRIO with the Arm Instruction Emulator (Arm IE), which allows to the instrument and execute workload compiled with Scalable Vector Extension (SVE) support on the current generation of Arm platforms. In particular, 3 new clients will be soon available in next public release: SVE memory tracing, SVE instruction tracing and SVE instruction counter. Instructions and memory traces can provide terrific insights about architectural choices, driving co-design considerations such as sizing of vector units, sizing of caches, memory utilisation, estimated vs measured executed instructions per cycle. Lessons learned and co-design efforts will be used to design future features, which will be invaluable to current and future Arm partners. Traces and instruction counters are two great metrics to understand the ability of a compiler to convert the high-level code written by a domain expert and into properly vectorized machine code. Via a close interaction with MAX application developers, it will be possible to identify common best practices in code development that will help to write scalable and vectorizable code able to efficiently exploit future multicore and many-core systems. Arm has recently lunch an initiative called Arm HPC Software Grants to help researchers and partners growing Arm software ecosystem. This programme provides access to premium tools and support for experimenting with and porting codes to Arm hardware. Eligible projects are limited in time and scope but, once granted, full access to Arm HPC Suite which includes compilers, performance libraries, Arm Forge and Arm Performance Reports. Arm will make it accessible to all the partners during the duration of the project, supporting porting and optimisation in-kind in addition to the effort spent in more focus co-design activities.
