ARM Research Summit 2017 Workshop
The ARM Research Summit is an academic summit to discuss future trends and disruptive technologies across all sectors of computing. On the first day of the Summit, ARM Research will host a gem5 workshop to give a brief overview of gem5 for computer engineers who are new to gem5 and dive deeper into some of gem5's more advanced capabilities. The attendees will learn what gem5 can and cannot do, how to use and extend gem5, as well as how to contribute back to gem5.
The ARM Research Summit will take place in Cambridge (UK) over the days of 11-13 September 2017. The gem5 workshop will be a full day event on the 11th September.
Contents
- 1 Streaming & Offline viewing
- 2 Target Audience
- 3 Registration
- 4 Schedule
- 5 Talks
- 6 Short Talks
- 6.1 Debugging a target-agnostic JIT compiler with GEM5
- 6.2 COSSIM: An Integrated Solution to Address the Simulator Gap for Parallel Heterogeneous Systems
- 6.3 Simulation of Complex Systems Incorporating Hardware Accelerators
- 6.4 Generating Synthetic Traffic for Heterogeneous Architectures
- 6.5 ARM Research Starter Kit: System Modeling using gem5
- 6.6 Interacting with gem5 using workload-automation & devlib
- 6.7 gem5: empowering the masses
- 6.8 Integrating and quantifying the impact of low power modes in the DRAM controller in gem5
Streaming & Offline viewing
The workshop is being streamed live and all talks will be available on YouTube after the workshop. See the main summit page for details.
Target Audience
The primary audience is researchers who are using, or planning to use, gem5 for architecture research.
Prerequisites: Attendees are expected to have a working knowledge of C++, Python, and computer systems.
Registration
See the main ARM Research Summit website for details about registration.
Schedule
The workshop will take place on Monday the 11th September 2017 at Robinson College in Cambridge (UK). The workshop starts at 9.00 and runs in parallel with the main Summit program until 16.30 when it joins the main program.
Talks
Trace-driven simulation of multithreaded applications in gem5
The gem5 modular simulator provides a rich set of CPU models which permits balancing simulation speed and accuracy. The growing interest in using gem5 for design-space exploration however requires higher simulation speeds so as to enable scalability analysis with systems comprising tens to hundreds of cores. One relevant approach for enabling significant speedups lies in using trace-driven simulation, in which CPU cores are abstracted away thereby enabling to refocus simulation effort on memory/interconnect subsystems which play a key role on performance. This talk describes some of the work carried out on the Mont-Blanc european projects on trace-driven simulation and discusses the related challenges for multicore architectures in which trace injection requires to account for the API synchronization of the underlying running application. The ElasticSimMATE tool is presented as an initiative towards combining Elastic Traces and SimMATE so as to enable fast and accurate simulation of multithreaded applications on ARM multicore systems.
Dr Gilles Sassatelli is a CNRS senior scientist at LIRMM, a CNRS-University of Montpellier academic research unit with a staff of over 400. He is vice-head of the microelectronics department and leads a group of 20 researchers working in the area of smart embedded digital systems. He has authored over 200 peer-reviewed papers and has occupied key roles in a number of international conferences. Most of his research is conducted in the frame of international EU-funded projects such as the DreamCloud and Mont-Blanc projects.
Alejandro Nocua received the Ph.D. degree in Microelectronics from the University of Montpellier, France, in 2016. Currently, he is a postdoctoral researcher at the French National Center for Scientific Research (CNRS). His research interests include the analysis of high-performance and energy-efficiency design methodologies. He received his Master degree in Science from the National Institute of Astrophysics, Optics and Electronics (INAOE), Mexico, in 2013. Alejandro was awarded his BS degree in Electronics Engineering from Industrial University of Santander (UIS), Colombia in 2011.
Florent Bruguier received the M.S. and Ph.D. degrees in microelectronics from the University of Montpellier, France, in 2009 and 2012, respectively. From 2012 to 2015, he was a Scientific Assistant with the Montpellier Laboratory of Informatics, Robotics, and Microelectronics, University of Montpellier. Since 2015, he is a Permanent Associate Professor. He has co-authored over 30 publications. His research interests are focused on self-adaptive and secure approaches for embedded systems.
Anastasiia Butko, Ph.D. is a Postdoctoral Fellow in the Computational Research Division at Lawrence Berkeley National Laboratory (LBNL), CA. Her research interests lie in the general area of computer architecture, with particular emphasis on high-performance computing, emerging and heterogeneous technologies, associated parallel programming and architectural simulation techniques. Broadly, her reasearch addresses the question of how alternative technologies can provide continuing performance scaling in the approaching Post-Moore’s Law era. Her primary research projects include development of the EDA tools for fast superconducting logic design, development of the classical ISA for quantum processor control, development of the fast and flexible System-on-Chip generators using Chisel DSL. Dr. Butko co-leads Open Source Supercomputing project and is a technical committee member of the RISC-V foundation.
Dr. Butko received her Ph.D. in Microelectronics from the University of Montpellier, France (2015). Her doctoral thesis developed fast and accurate simulation techniques for many-core architectures exploration. Her graduate work has been conducted within the European project MontBlanc, which aims to design a new supercomputer architecture using low-power embedded technologies.
Dr. Butko received her MSc. Degree in Microelectronics from UM2, France and MSc and BSc Degrees in Digital Electronics from NTUU "KPI", Ukraine. During her Master she participated on the international program of double diploma between Montpellier and Kiev universities.
Modeling Cache Coherence with gem5
Correctly implementing cache coherence protocols is hard and these implementation details can affect the system's performance. Therefore, it is important to robustly model the detailed cache coherence implementation. The popular computer architecture simulator gem5 uses Ruby as its cache coherence model providing higher fidelity cache coherence modeling than many other simulators.
In this talk, I will give a brief overview of Ruby, including SLICC: the domain-specific language Ruby uses to specify cache protocols. I will show the extreme flexibility of this model and details of a simple cache coherence protocol. After this talk, you will be able to dive in and begin writing your own coherence protocols!
Jason Lowe-Power is an Assistant Professor at University of California, Davis in the Computer Science department. Jason's research focuses on increasing the energy efficiency and performance of end-to-end applications like analytic database operations used by Amazon, Google, Target, etc. One important aspect of this research is adding hardware mechanisms to systems that enable all programmers to use emerging hardware accelerators like GPUs. Additionally, Jason is a leader of the open-source architectural simulator, gem5, used by over 1500 academic papers. Jason received his PhD from University of Wisconsin-Madison in Summer 2017. He was awarded the Wisconsin Distinguished Graduate Fellowship Cisco Computer Sciences Award in 2014 and 2015.
A Detailed On-Chip Network Model inside a Full-System Simulator
Compute systems are ubiquitous, with form factors ranging from smartphones at the edge to datacenters in the cloud. Chips in all these systems today comprise 10s to 100s of homogeneous/heterogeneous cores or processing elements. The growing emphasis on parallelism, distributed computing, heterogeneity, and energy-efficiency across all these systems makes the design of the Network-on-Chip (NoC) fabric connecting the cores critical to both high-performance and low power consumption.
It is imperative to model the details of the NoC when architecting and exploring the design-space of a complex many-core system. If ignored, an inaccurate NoC model could lead to over-design or under-design due to incorrect trade-off choices, causing performance losses at runtime. To this end, we have designed and integrated a detailed on-chip network model called Garnet inside the gem5 (www.gem5.org) full-system architectural simulator which is being used extensively by both industry and academia. Together with Garnet, gem5 provides plug-and-play models of cores, caches, cache coherence protocols, NoC, memory controller, and DRAM, with varying levels of details, enabling computer architects and designers to trade-off simulation speed and accuracy.
In this talk, we will first introduce the basic building blocks of NoCs and present the state-of-the-art used in chips today. We will then present Garnet, and demonstrate how it faithfully models the state-of-the-art, while also offering immense flexibility in modifying various parts of the microarchitecture to serve the needs of both homogeneous many-cores and heterogeneous accelerator-based systems of the future via case studies and code-snippets. Finally, we will demonstrate how Garnet works within the entire gem5 ecosystem.
Tushar Krishna is an Assistant Professor in the Schools of ECE and CS at Georgia Tech. He received a Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology in 2014. Prior to that he received a M.S.E from Princeton University in 2009, and a B.Tech from the Indian Institute of Technology (IIT) Delhi in 2007, both in Electrical Engineering.
Before joining Georgia Tech in 2015, Dr. Krishna was a post-doctoral researcher in the VSSAD Group at Intel, Massachusetts, and then at the Singapore-MIT Alliance for Research and Technology at MIT.
Dr. Krishna's research interests are in computer architecture, interconnection networks, networks-on-chip, deep learning accelerators, and FPGAs.
System Simulation with gem5, SystemC and other Tools
SystemC TLM based virtual prototypes have become the main tool in industry and research for concurrent hardware and software development, as well as hardware design space exploration. However, there exists a lack of accurate, free, changeable and realistic SystemC models of modern CPUs. Therefore, many researchers use the cycle accurate open source system simulator gem5, which has been developed in parallel to the SystemC standard. In this tutorial we present the coupling of gem5 with SystemC that offers full interoperability between both simulation frameworks, and therefore enables a huge set of possibilities for system level design space exploration. Furthermore, we show several examples for coupling gem5 with SystemC and other tools.
Matthias Jung received his PhD degree in Electrical Engineering from the University of Kaiserslautern Germany in 2017. His research interest are SystemC based virtual prototypes, especially with the focus on the modeling of memory systems and memory controller design. Since may 2017 he is a researcher at Fraunhofer IESE, Kaiserslautern, Germany.
Christian Menard received a Diploma degree in Information Systems Technology from TU Dresden in Germany in 2016 and joined the chair for compiler construction as a Ph.D. student within the excellence cluster cfaed in TU Dresden. His current research includes system-level modeling of widely heterogeneous hardware as well dataflow compilers for heterogeneous MPSoC platforms.
CPU power estimation using PMCs and its application in gem5
Fast and accurate estimation of CPU power consumption is necessary to inform run-time power management approaches and allow effective design space exploration. Power simulators, combined with a full-system architectural simulator such as gem5, enable power-performance trade-offs to be investigated early in the design of a system. However, the accuracy of existing power simulators is known to be low, and this can lead to incorrect conclusions being made. In this talk, I will present our statistically rigorous methodology for building accurate run-time power models using Performance Monitoring Counters (PMCs) for mobile and embedded devices, and demonstrate how our models make more efficient use of limited training data and better adapt to unseen scenarios by uniquely considering stability. Models built using the methodology for both ARM Cortex-A7 and Cortex-A15 CPUs exhibit a 3.8% and 2.8% average error respectively. I will also present online resources that we have made available from the work, including software tools, documentation, raw data and further results. I will also present results from an investigation into the correlation between gem5 activity statistics and hardware PMCs. Based on this, a gem5 power model for a simulated quadcore ARM Cortex-A15 has been created, built using the above methodology, and its accuracy compared against experimental results obtained from hardware.
Geoff Merrett is an Associate Professor in the Department of Electronics and Computer Science at the University of Southampton. He received the BEng (1st, Hons) and PhD degrees in Electronic Engineering from Southampton in 2004 and 2009 respectively. His research interests are in energy-aware and self-powered computing systems, with application across the spectrum from highly constrained IoT devices to many-core mobile and embedded systems. He has published over 100 peer-reviewed articles in these areas, and given invited talks at a number of international events. Dr Merrett is a Co-Investigator on the EPSRC-funded £5.6M PRiME Programme Grant (where he leads the applications and cross-layer interaction theme), "Continuous on-line adaptation in many-core systems: From graceful degradation to graceful amelioration", and deputy-lead on the "Wearable and Autonomous Computing for Future Smart Cities" Platform Grant. He is technical manager of Southampton’s ARM-ECS Research Centre, an award-winning industry-academia collaboration between the University of Southampton and ARM. He coordinates IoT research at the University, and leads the wireless sensing theme of its Pervasive Systems Centre. He is an Associate Editor for the IET CDS journal, serves as a reviewer for a number of leading journals, and on TPCs for a range of conferences. He co-manages the UK’s Energy Harvesting Network, was General Chair of the ACM Workshop on Energy-Harvesting and Energy-Neutral Sensing Systems in 2013, 2014, and 2015, and was the General Chair of the European Workshop on Microelectronics Education 2016. He is a member of the IEEE, IET and Fellow of the HEA.
Short Talks
Debugging a target-agnostic JIT compiler with GEM5
Author: Boris Shingarov - LabWare
We explain how GEM5 enabled us to develop a target-agnostic JIT compiler, in which no knowledge about the target ISA is coded by the human programmer; instead, the backend is inferred, using logic programming, from a formal machine description written in a Processor Description Language. Debugging such a JIT presents some challenges which can not be addressed using traditional approaches. One such challenge is the impedance mismatch between the high-level abstractions in the PDL and the low-level inferred implementation. In this talk, we present a new debugger based on simulating the execution of the target runtime VM in GEM5; the debugger frontend connects to this simulation using the RSP wire protocol.
COSSIM: An Integrated Solution to Address the Simulator Gap for Parallel Heterogeneous Systems
In an era of complex networked heterogeneous systems, simulating independently only parts, components or attributes of a system-under-design is not a viable, accurate or efficient option. The interactions are too many and too complicated to produce meaningful results and the optimization opportunities are severely limited when considering each part of a system in an isolated manner. COSSIM offers a framework that can handle the simulation of a complete system-of-systems including processors, peripherals and networks that can appeal to Parallel (Heterogeneous) Systems designers and application developers in an integrated way.
The framework is based on gem5 as the main simulation engine for processor-based systems and extends its capabilities by integrating it with the OMNET++ network simulator. This integration allows independent gem5 instances to be networked with all network protocols and hierarchies that can be supported by OMNET++, thus creating a very flexible solution. The integration of the two main simulation tools is realized through the IEEE 1516 High-Level Architecture standard (HLA), through which all communication tasks are performed. Through HLA and custom libraries, a two-level (per node and global) synchronization scheme is also implemented to ensure a coherent notion of time between all nodes.
Since HLA is IP-based all gem5 instances and OMNET++ can be executed on the same physical machine or on any distributed system (or any combination in between). The overall framework – the set of gem5 nodes, the OMNET++ simulator and the CERTI HLA – are integrated in a unified Eclipse-based GUI that has been developed to provide easy simulation set-up, execution and visualization of results. McPAT is also integrated in a semi-automated way through the GUI in order to provide power and energy estimations for each node, while OMNET++ provides power estimations for networking-related components (NICs and network devices).
Andreas Brokalakis is a senior hardware engineer at Synelixis Solutions Ltd. At the same time he is pursuing a PhD degree at the Technical University of Crete, Greece. He holds a Bachelor degree in Computer Engineering from University of Patras, Greece and a Master’s Degree on Hardware/Software Co-design from the same university. Current work and research interests involve computer architecture and arithmetic, as well as design of ASIC and FPGA systems and accelerators.
Nikolaos Tampouratzis is a PhD student at Technical University of Crete, working on simulation tools for computing systems. He has joined Telecommunication Systems Institute, Technical University of Crete since October 2012 as a research associate, providing research and development services to several EU-funded research projects. He received his Computer Science diploma from the University of Crete (UOC, Greece), with specialization in Hardware Design and FPGAs. He continued his studies in the Technical University of Crete (TUC Greece) where he received his Master Diploma in Electronic and Computer Engineering in which he specialized in Computer Architecture and Hardware Design.
Simulation of Complex Systems Incorporating Hardware Accelerators
The breakdown of Dennard scaling coupled with the persistently growing transistor counts increased the importance of application-specific hardware acceleration; such an approach offers significant performance and energy benefits compared to general-purpose solutions. In order to thoroughly evaluate such architectures, the designer should perform a quite extensive design space exploration so as to evaluate the trade-offs across the entire system. The design, until recently, has been predominantly done using Register Transfer Level languages such as Verilog and VHDL, which, however, lead to a prohibitively long and costly design effort. In order to reduce the design time a wide range of both commercial and academic High-Level Synthesis (HLS) tools have emerged; most of these tools, handle hardware accelerators that are described in synthesizable SystemC. The problem today, however, is that most simulators used for evaluating the complete user applications (i.e. full-system CPU/Mem/Peripheral simulators) lack any type of SystemC accelerator support.
Within this context, we extend gem5 to support the simulation of generic SystemC accelerators. We introduce a novel flow that enables us to rapidly prototype synthesisable SystemC hardware accelerators in conjunction with gem5. The proposed solution handles automatically all communication and synchronisation issues.
Compared to a standard gem5 system, several changes at different levels are required, from the OS and device drivers level down to the implementation of a device model in the gem5 simulator. Instead of using files to write data for an external accelerator, perform the simulation and then read back the results, our approach communicates with the SystemC simulator through programmed I/Os and DMA engines, supporting full global synchronisation. Apart from the apparent benefits concerning the implementation and simulation accuracy, the proposed solution is also orders of magnitude faster.
Nikolaos Tampouratzis is a PhD student at Technical University of Crete, working on simulation tools for computing systems. He has joined Telecommunication Systems Institute, Technical University of Crete since October 2012 as a research associate, providing research and development services to several EU-funded research projects. He received his Computer Science diploma from the University of Crete (UOC, Greece), with specialization in Hardware Design and FPGAs. He continued his studies in the Technical University of Crete (TUC Greece) where he received his Master Diploma in Electronic and Computer Engineering in which he specialized in Computer Architecture and Hardware Design.
Generating Synthetic Traffic for Heterogeneous Architectures
Modern system-on-chip architectures consist of many heterogeneous processing elements. The communication fabric and memory hierarchy supporting these processing elements heavily influence the system’s overall performance. Exploring the design space of these heterogeneous architectures with detailed models of each processing element can be time-consuming. Statistical simulation has been shown to be an effective tool for quickly evaluating architectures by abstracting away complexity.
This talk describes work done on modelling the spatial and temporal behaviour of a processing element’s address stream. We present a methodology that can automatically characterize a processing element by observing its reads and writes. Using these characteristics we can stimulate a communication fabric connecting many different processing elements by synthetically recreating their addresses. These addresses arrive at their destination in the memory hierarchy, spawning new messages and responses to read and write requests. Architects can now combine ynthetic processing elements that represent various different components on current and future systems-on-chip to evaluate the impact of changes at the interconnection network and memory hierarchy.
Mario Badr is a PhD Candidate at the University of Toronto working under the supervision of Dr. Natalie Enright Jerger. He received his B.A.Sc. and M.A.Sc from the University of Toronto in Electrical Engineering and Computer Engineering, respectively. He has interned with Qualcomm Research Silicon Valley and received the Roberto Padovani Scholarship for his outstanding technical contributions. In addition, he has been recognized at the university and departmental levels for excellence as a teaching assistant. His research interests include performance evaluation in computer architecture, heterogeneous architectures, and multi-threaded workloads.
ARM Research Starter Kit: System Modeling using gem5
ARM Research Enablement aims to enhance computing research by enabling researchers worldwide to easily access ARM-based IP and technologies, and helping them to increase their research impact. As a part of our research enablement activities, we provide a System Modeling Research Starter Kit using gem5. We have released a High Performance In-order (HPI) CPU timing model based on ARMv8-A in gem5. I will present a high-level overview of the released system, its documentation and benchmark scripts. This talk will target those who are new to gem5 as well as those who would like to promote gem5 in research.
Ashkan Tousi is a Senior Research Engineer at ARM Cambridge and an Honorary Lecturer at the University of Glasgow. He received his PhD in computing science (parallel computing) in 2015. He currently leads research enablement activities at ARM, which cover a range of different research areas from SoC design to IoT and data science.
Interacting with gem5 using workload-automation & devlib
Running workloads on gem5 is often not straightforward. This talk will discuss workload-automation and devlib, 2 new open-source tools to interact with gem5. These frameworks, written to interact with various hardware platforms, have recently been extended to include gem5 as a platform. We will discuss use cases and advantages/disadvantages of each tool and show how they can make your gem5 work easier.
Anouk Van Laer is a Modelling Engineer in Architecture: Systems & Technology group at ARM. She obtained her PhD at University College London, where she investigated the effects of optical interconnects on the performance of chip multiprocessors, using gem5.
gem5: empowering the masses
This talk will give an overview of the state of power modelling in gem5. After discussing the basic power modelling infrastructure, it will cover the state of CPU DVFS as well as recent improvements in how CPU power states are controlled for the ARM architecture in gem5. The talk will cover these improvements in power modelling, highlighting the way in which the accuracy and versatility of the simulator have been improved.
Sascha Bischoff is a Senior Software Engineer in the Architecture: Systems & Technology group at ARM in Cambridge. Whilst completing his PhD with the University of Southampton, he spent 3.5 years based in ARM Research in Cambridge. He has spent a large part of the last 6 years working with gem5, typically with a focus on power management, ideally without impacting the delivered performance.
Integrating and quantifying the impact of low power modes in the DRAM controller in gem5
Across applications, DRAM is a significant contributor to the overall system power, with the DRAM access energy per bit up to three orders of magnitude higher compared to on-chip memory accesses. To improve the power efficiency, DRAM technology incorporates multiple low power modes, each with different trade-offs between achievable power savings and performance impact due to entry and exit delay requirements. Accurate modeling of these low power modes and entry and exit control is crucial to analyze the trade-offs across controller configurations and workloads with varied memory access characteristics.
In this talk, we will give an overview of the decision making logic we added to the DRAM controller in gem5 that triggers transitions to/from the power-down modes. Integrating this functionality makes gem5 the first publicly available DRAM low power full-system simulator, providing the research community a tool for DRAM power analysis for a breadth of use cases. We will conclude with simulation data that characterises the low power behaviour and shows energy and performance trade-offs for realistic workloads.
Note: This talk is based on a paper accepted at MEMSYS 17. Authors from ARM: Radhika Jagtap, Wendy Elsasser and Andreas Hansson. Authors from University of Kaiserslautern: Matthias Jung and Norbert Wehn.
Radhika Jagtap is a Senior Research Engineer working in the Memory & Systems research group. She has plenty of experience with gem5 (elastic traces, interconnect, memory controller) and is involved in several collaborative research projects, especially with academics. Currently she is exploring the problem of energy efficient data movement for sparse data workloads.
