Welcome to DRT !

Many runtime libraries exist to easy the programmability of current multi-cores and future multi-multi-cores. Most of them expose to the programmers a high level interface, therefore leaving a lot of internal overhead to the runtime libraries: e.g. OpenMP, Cilk, TBB, SWARM ... just to name a few.

However, as of my knowledge, none of them is designed to offer a very low level API that could be targeted on one side by compilers and that could be easily migrated in future processor through, e.g., a simple Instruction Set Extension (ISE) of an existing ISA (e.g. X86 or ARM).

Recent efforts like OCR (Open Community Runtime) are going in the same direction like our DRT; however, they bring a complexity that is necessary to cover the needs of many requirements.

With DRT, the complexity is reduced to a minimum (current implementation is about 250 lines of C code). The current efforts are directed towards a compiler chain from pure C and toward the evaluation of an efficent ISE that we call T-STAR through the HPLabs COTSon simulator. Preliminary tests brought us very nice scalability: we tested up to 1024 cores organized as a single-chip multi-multi-core, but there is no limitation that we have foreseen except, of course, availability of resources, availability of input data, limitations in the algorithm.

Please enjoy and contribute by this blog or by directing emails to me.
You are very welcome to expose any criticism or appreciation :-) and to contribute to this software as well.

Roberto Giorgi.