Whenever I cover a trade show, I always have my antenna up for companies I haven’t seen or heard about. The recent Design Automation Conference, held at the Anaheim Convention Center, did not disappoint as two companies fit the bill: Compaan Design and Menta.
Parallelizing Sequential Code
Compaan has developed a technology for hotspot parallelization of legacy/certified ISO C applications. The resulting product, HotSpot Parallelizer for ISO C, translates C-code hotspots to data-streaming Kahn Process Networks (KPNs) that robustly and efficiently utilize highly parallel heterogeneous multicore chip architectures.
Much legacy C code does not take advantage of today’s multicore processors. Compann’s product helps engineers use this code more efficiently on these processors, especially for jobs like streaming video. The product examines the C code and finds hotspots within the code where parallel processing on multicore processors can achieve higher throughput or lower energy consumption.
If you’re not familiar with KPNs, Wikipedia describes them as “a distributed model of computation (MoC) where a group of deterministic sequential processes are communicating through unbounded FIFO channels. The resulting process network exhibits deterministic behavior that does not depend on the various computation or communication delays.”
Wikipedia adds, “The model was originally developed for modeling distributed systems but has proven its convenience for modeling signal processing systems. As such, KPNs have found many applications in modeling embedded systems and high-performance computing systems.”
“On an FPGA, you have lots of parallelism and distributed memory. But C code is sequential and uses global memory, and that doesn’t work very well,” said Bart Kienhuis, CEO of Compaan, describing the problem of mapping legacy C code to an FPGA. “We do a step in the middle called the Kahn Process Networks, which we derive automatically out of C code.”
The product analyzes the C code, finds all the data flow in the code, and converts that code into KPNs. These independent processes talk over FIFO channels and can be easily mapped into FPGAs.
Pointing to the computer, Kienhuis said, “What you see here are all kinds of memory references. If you look at this code, you might think, ‘How can I ever parallelize it?’ That’s why you have tools. The tools will figure it out for you.”
The HotSpot Parallelizer works in conjunction with Xilinx FPGAs, but Compaan expects to support other architectures in the future. Check out Engineering TV at www.engineeringtv.com/video/Code-to-FPGAs-Made-Easy-by-Comp for more.
Meanwhile, Menta showed its eFPGA Creator and eFPGA Core. The company touts the former as the first complete design environment for creating customizable programmable logic architectures. The latter is a high-density embedded programmable-logic core designed for systems-on-a-chip (SoCs), ASICs, or systems-in-a-package (SiPs).
Just prior to the show, Menta announced the tapeout of the world’s first MRAM-based FPGA. According to the company, it leverages key innovations including nonvolatile magnetic memory and proprietary circuitry that enables compact integration of MRAM and embedded-FPGA solutions.
Developed in joint collaboration with the Microelectronic department of LIRMM (Laboratory of Informatics, Robotics and Microelectronics of Montpellier), this tapeout validates the possibility of stacking MRAM technology over traditional CMOS logic and introduces to the market a new type of robust nonvolatile FPGA. MRAM-based FPGAs are being validated for defense, aerospace, automotive, and consumer applications.
“This world-first MRAM-based FPGA demonstrates the versatility of our eFPGA Core technology,” said Laurent Rougé, Menta’s CEO and founder. “One of the key benefits of developing an FPGA with MRAM technology is that it enables high-density nonvolatile FPGAs based on leading-edge CMOS technology nodes, unlike traditional flash-based approaches only available on mature CMOS processes.”
Back in the office, I mentioned these two companies to my editorial colleagues, David Maliniak and Bill Wong. David said that Compaan’s product reminded him of a product from Critical Blue called Prism, which helps analyze code to implement and optimize it on multicore architectures.
But the guys had a “wait and see” attitude about eFPGA Core. Menta plans to implement a business licensing model like the one ARM uses for processor cores. How this pans out remains to be seen.