A direct array injection technique in a fine-grain multithreading execution model

We propose on optimized array handling scheme called the direct array injection technique (DAIT) which can be used in conjunction with I-Structures to provide non-strict and parallel array accesses in a fine/medium grain multithreaded computing environment. By overlapping array production and consumption, better processor utilization can be achieved resulting in higher performance. The direct array injection technique differs from the I-Structure approach in that array elements are forwarded directly from the producer activation to the consumer activation without being buffered in a structure store. The advantages of this technique over I-Structures are: network traffic is reduced; and dynamic memory allocation from the global heap space to store arrays is avoided