Zastosowanie standardu OpenMP do projektowania systemów wbudowanych

Artykuł prezentuje nowe podejście do projektowania systemów wbudowanych z użyciem języka C z dyrektywami OpenMP. Opisano w nim motywację użycia standardu OpenMP do syntezy sprzętowo-programowej. Przedstawiono proponowane rozwiązanie oraz porównano je z klasycznym projektowaniem systemów sprzętowych. Przedstawiono także konstrukcje równoległe standardu OpenMP, syntezowane do postaci współbieżnych układów cyfrowych. Pokazano przykładowy program w języku OpenMP wraz z jego przekładem do kodu SystemC oraz schemat RTL układu będącego wynikiem syntezy opisanego źródła.

The embedded system is a special-purpose computer that performs one or a few dedicated tasks. It contains hardware and software parts [3]. The paper presents a new approach to embedded system design using C language with OpenMP directives. It is different from classic hardware design (Fig. 1a) because it allows describing both hardware and software using a common language (Fig. 1b). OpenMP is a standard that specifies parallel programs using a shared memory architecture. It is the collection of compiler directives and runtime library functions in C/C++ and Fortran languages [11]. Support for concurrency that corresponds to hardware performance is the main motivation of using OpenMP to embedded system design. OpenMP enables describing chips on high level of abstraction without knowledge about details of its structure. It improves flexibility of the software/hardware migration. OpenMP offers simulation, verification and estimation of the system performance. There is sufficient amount of legacy C libraries which facilitate the task of system modeling. Fig. 2a shows an example of OpenMP code that adds two matrixes A and B using a parallel loop. The systemC program being the results of behavioral synthesis of the example 2a is presented in Fig. 2b. Parallel regions in OpenMP have been transformed to SC_METHODS processes in SystemC. Fig. 3 shows the RTL schematic diagram of the chip synthesized from a code 2b. It contains three blocks proc1, proc2, proc3 that are equivalent to threads in OpenMP program. A schematic diagram of the single block is presented in Fig. 4. The unit consists of an adder, a FDE flip-flop that realizes barrier synchronization and two FDR flip-flops representing signals S and R.