UPMEM’s processing-in-memory technology, set to launch in 2019, is designed to speed up big-data analytics by a factor of 20 or more while also reducing power consumption, making it highly attractive for high-throughput data-intensive applications, such as human genome sequencing/mapping or string searching on compressed document databases. UPMEM will use Amsterdam-based Solid Sands' SuperTest to test and validate the LLVM compiler it is developing for its embedded processor architecture.
“The compiler is absolutely critical, because if you have any kind of bug in your compiler it has a very serious impact on your customers. So we need to test it very intensively. After evaluating all the options, we chose SuperTest for two main reasons: firstly, because of its very exhaustive test coverage, and secondly, because its fine-grain reporting structure allows us to quickly pin-point the underlying cause of compiler errors,” said David Furodet, Software Manager at UPMEM. “An added advantage was that it was very easy to get SuperTest up and running and delivering results.”
When executing high-throughput data-intensive applications such as genome analysis or string searching, conventional hardware architectures require large amounts of data to be moved from system memory into on-chip cache, consuming a considerable amount of time and power.
UPMEM’s processing-in-memory technology shifts the highly repetitive tasks typical of these applications into a programmable processor embedded within the system memory DRAM chips. Application developers can continue to write their programs in C or C++, with appropriately compiled executable code segments loaded to these embedded processors at run-time. In tests, UPMEM’s processing-in-memory solutions have been shown to increase the execution speed of many applications by a factor of 20 or more.
“SuperTest has long been used to validate compilers for conventional processor architectures such as x86 and ARM, revealing bugs in virtually all of the so-called standard compiler collections. However, it’s equally good at testing and validating compilers for custom processor architectures like those being developed by UPMEM,”