Resource management in dataflow-based multithreaded execution

Due to the large amount of potential parallelism, resource management is a critical issue in multithreaded execution. The challenge in code generation is to control the parallelism without reducing the machine's ability to ex ploit it. Controlled parallelism reduces idle time, communication, and dela y caused by synchronization. Ar tile same time it increases the potential f or exploitation of program data structure locality. In this paper, we evalu ate the performance of methods to control program parallelism and resource usage in the context of the fine-grain dataflow execution model. The method s are in themselves not new, but their performance analysis is. The two met hods to control parallelism here are slicing and chunking. We present the m ethods and their compilation strategy and evaluate their effectiveness in t erms of run time and matching store occupancy. Communication is categorized in memory, loop, call, and expression communication. Input and output mess age locality is measured. Two techniques to reduce communication are introd uced. Grouping allocates loop and function bodies on one processor and bund ling combines messages with the same sender and receiver into one. Their ef fects on the total communication volume are quantified. (C) 2001 Academic P ress.