Hitachi developed an integrated array processor (IAP) as an add-on unit to boost the scientific and technical computation performance of its general-purpose HITAC M-180 computer. The HITAC M-180 IAP was completed in 1978. Hitachi developed its own internal vector arithmetic methodology (a set of about 40 vector instructions) with the objective of providing an inherently usable device with excellent cost effectiveness. As a result, the HITAC M-180 IAP was the first general-purpose machine to reach an effective performance speed of over 10 megaFLOPS.
Users could make use of existing standard Fortran programs as is, as well as TSS, because code was compiled automatically by a sophisticated vector compiler that made vector calculations on data in a virtual space, without using vector registers. The compiler detected code in a Fortran program that could be vectorized and generated the applicable machine instruction code using the M-180 IAP’s vector instructions. Fortran do loops are usually thought of as candidates for vectorization, but not all do loops can be vectorized. This is because executing a standard Fortran program written assuming sequential execution with vector instructions will end up changing the execution order and possibly the expected outcome. The data referent parsing technique developed for this compiler, however, made it possible to automatically vectorize do loops. One special feature of the M-180 IAP’s vector instruction set was an instruction that calculated inner products. The M-180 IAP’s Fortran compiler detected do loops calculating inner products and vectorized them by applying the inner product vector instruction.
The M-180 IAP was widely used by large-scale computation centers at universities and by the Meteorological Agency and others in real-world applications, thereby helping to improve the application technology for vector computers. Hitachi’s experience with vector computers was put to use in the development and application of later supercomputers.
Later IAPs extended the vector instruction set for greater code vectorization and worked as add-on units for Hitachi’s M-200H, M-280H, and M-680H models. The M-200H IAP, in particular, supported functions for sums, inner products, and first-order recurrence relations, as well as indirect referencing — functions that the Cray-1 did not completely support at the time. The M-280H IAP added control vectors and became the world’s first unit to automatically vectorize conditional do loops using these control vectors.
The application of vector computers expanded after this time from numeric computations to symbolic computations. This was brought about by the development of the HITAC M-680H IDP (integrated database processor) and other vector computers that processed relational databases.
