The research work is dedicated to short SIMD processing and the integration of these new processing facilities into high-level programming languages. Currently, the most significant activities are development of syntax and semantics extensions to the ANSI C in such a way that we could use short SIMD processing facilities in C programming language and development of its vectorizing compiler which is capable to automatically extract short SIMD parallelism from loops. We call it MMC (multimedia C) language. We have also introduced a new data-dependence test for array references with linear subscripts. It is a simple method that can be used to filter out data dependences with a sufficiently large distance between memory references. By checking the distance between the memory references we can filter out data dependences that do not prohibit potential SIMD vectorization and that can be excluded from further, possibly more accurate and expensive, tests. The method is based on checking the minimum and maximum distances between two memory references within the iteration space rather than searching for the existence of an integer solution to the dependency equation. This results in greater accuracy and also in a reduction of the time cost as it computes only the bounds of an integer affine function.

The part of research work is directed toward the development of special computer hardware which, through its programmable ALU, is capable of performing custom selected functions. We are also developing an embedded system with two RISC microprocessors, which is implemented in field-programmable gate.