Program transformations as the base for optimizing parallelizing compilers

The paper deals with program transformations leading to acceleration and summarize the publications on various parallel computing architectures and tools for developing effective programs for them. The discussion focuses on a combination of parallelization and optimization of access to memory modules of different levels. It highlights that the lag of automatic program optimization from the needs of new architectures restrains the development of new promising computing systems. The development of the theory of program transformation and optimizing (parallelizing) compilers could lead to a significant increase in the productivity of programmers. The article substantiates the call for the modernization of the optimizing compilation and presents new problem statements.