Mitrionics Looks Beyond FPGAs

By Nicole Hemsoth

November 18, 2009

At SC09 this week, Mitrionics announced it has started to work on an experimental compiler that aims to make parallel programming architecture-agnostic. The goal of the work is to extend the Mitrion-C platform for FPGAs to multicore CPUs, cluster architectures, and eventually even GPGPUs.  We asked Stefan Möhl, Mitrionics’ chief science officer and co-founder, to explain what’s behind the new technology and what prompted the decision to add support for other parallel architectures.

HPCwire: Can you tell us about the new programming capabilities of the Mitrionics platform that you announced here at the Supercomputing Conference?

Stefan Möhl: Well, we haven’t added new programming capabilities to the Mitrionics’ Accelerated Computing Platform yet. We are still in the proof-of-concept stage with this new compiler, but things look very promising. For this proof-of-concept compiler, the news is that existing Mitrion-C code, originally written for the MVP on FPGAs, will now also run on multicores and clusters. This initial proof-of-concept was made only to prove that the basic principles work, so there are limits to what code we can currently run. A production version of a portable programming language will require changes to Mitrion-C to make it less focused on what is needed for FPGA acceleration.

HPCwire: How does it work?

Möhl: The main challenge when porting between parallel architectures is that the level of granularity of the parallelism differs. For example, to parallelize code for vector processors, you would have to parallelize inner-most loops. To parallelize code for clusters, you would have to parallelize the outer-most loops. Doing general automatic parallelization (parallelization without re-writing the code) has not been solved, even after decades of research. Nor is there a general automatic way to transform one kind of parallelism into another.

Mitrion-C was originally developed as a programming language for the Mitrion Virtual Processor (MVP). The MVP is a hardware design for a compute engine specifically developed for high-performance execution in FPGAs. As such, it is full MIMD ((Multiple Instruction stream, Multiple Data stream) at the individual instruction level, so it potentially executes every single instruction of the program in parallel. This can be thought of as a limit-case for parallelism. Mitrion-C is a C-family language that supports and aids the programmer in specifying the kind of parallelism that the MVP requires. It is roughly as similar to ANSI-C as Java or C# are, so it isn’t too unusual to use.

The trick that makes Mitrion-C work for parallel portability comes from an important asymmetry in parallelization. Though automatic parallelization without code re-writes is very hard to achieve, general automatic sequentialization is much, much easier. Trivially, operating systems have run multiple programs in parallel on sequential processors for many years. For efficient execution, there are of course many optimization considerations, but it is still much easier than automatic parallelization. This property is what we use to port Mitrion-C between platforms. Since the code is fully parallel from the start, we never parallelize at all, we only sequentialize. So for a cluster, instead of parallelizing outer-most loops, we sequentialize everything except the outer-most loops. And for a vector processor, we sequentialize everything except for the inner-most loops.

HPCwire: So if you don’t have the parallelization problem, how can you handle the various memory architectures of multicore CPUs, GPGPUs and clusters, and so on?

Möhl: Our FPGA background has required us to consider these issues carefully from the start. FPGAs are usually connected to the system on data buses designed for devices with an order of magnitude less performance than FPGAs. So Mitrion-C was designed from the start to allow programmers to manage both memory latency and raw memory bandwidth in an effective manner. This issue will become increasingly important also for multicores and manycores, since increasing core counts without increasing clock-frequencies of data buses will put them in the same situation FPGAs have always been in.

Another important aspect comes from the diversity of FPGA cards. There are almost no two FPGA cards with the same memory sub-system, so we had to design Mitrion-C to have a memory model that addresses this from the start.

In Mitrion-C, there is no assumption of a single monolithic memory space. Instead, each collection may have its own address space, and different ones for different memory size and bandwidth requirements. This allows programmers to manually stage data from few, large and slow memories to many, small and fast memories in any number of levels. There are also several different built-in types for multi-dimensional data collections that let programmers specify what kind of access patterns a collection should permit. This helps the programmer in making correct and efficient programs, and also lets the compiler know what types of memory to place the data collection in. Of course, you can still write a program that requires more, larger or faster memories than a particular system has, but Mitrion-C will at least make you aware of what you demand of the system.

HPCwire: Can the exact same Mitrion-C source code be compiled to any of target architectures?

Möhl: Yes. You will need to parameterize for the number of cores you want to run on in a cluster or multicore, or how much unroll you want for loops in an FPGA, but other than that, the same code works without changes. However, not all algorithms will be efficient on all architectures, so the programmer will in some cases need to consider what platform to run the algorithms in, or change the algorithms to suit the available platform.

HPCwire: Mitrionics has focused on FPGA software development since its inception in 2001. What prompted the decision to target other architectures?

Möhl: Well, we are actually still focused on FPGAs. What prompted this is a customer interest in running Mitrion-C on standard processors and not only the Mitrion Virtual Processor. Customers want to be able to write an algorithm once and make efficient use of it on systems with and without FPGAs. They would also like to avoid having their code “locked in” to FPGAs. So we set up an experiment at Mitrionics to see what can be done with Mitrion-C on other platforms. And, as it turns out, very much can be done!

HPCwire: There are already a number of programming environments and languages that target multicore CPUs and GPGPUs and clusters. What does Mitrionics brings to the table?

Möhl: Three main things. First, Mitrion-C is a single, coherent language that maintains the same style of programming regardless of what platform you run it on. Programming languages like MPI, OpenMP, OpenCL and CUDA are really several different languages mixed together. There is the base-line C-code which is purely sequential, then there are added parts for clusters (in the case of MPI), multicores (in the case of OpenMP), or GPUs (in the case of OpenCL and CUDA). Often, you even have to combine them, such as with MPI+OpenMP. The additions introduce completely different ways of doing things than what the sequential C code does. They are not just added syntax in the sequential C paradigm. That means that you are really writing in several different languages at the same time, and need to learn them all to be able to do it properly. It also complicates the code dramatically.

Second, the fact that you have separate syntax for each architecture means that you need to re-write your code to move it between architectures. With our solution, software developers can make a single investment in writing code, and then use it on any architecture depending on what is optimal under current circumstances. With a universal programming language that can be used to target any architecture without changing syntax, it also becomes possible to explore the benefits and possibilities of different architectures much faster, in the end resulting in more efficient code.

Finally, and perhaps most tantalizing, is that the portability is not limited to the architectures that are popular today. History has seen a wide range of architectures — from the old scalar processors, vector processors, Thinking Machines, MasPar and SIMD, the Multi-Threaded Architecture, large shared memory machines, MPPs, and clusters to today’s FPGAs, GPUs, Cell, multicores and several others. Each new generation has required code re-writes. Though this is not yet proven, there is good hope that Mitrion-C would be efficient without re-writes on most of the historical popular parallel architectures. If that is the case, it bodes well for parallel architectures of the future too. Though we probably won’t be able to say “Never again!” to re-writes for all eternity, Mitrion-C holds the promise of dramatically reduce the number of re-writes we will need to do in the future.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire