A feasibility study of hierarchical multithreading
Mohamed Zahran, M. Franklin
Abstract
Mohamed Zahran, M. Franklin
Abstract
Many studies have shown that significant amounts of parallelism exist at different granularities. Execution models such as superscalar and VLIW exploit parallelism from a single thread. Multithreaded processors make a step towards exploiting parallelism from different threads, but are not geared to exploit parallelism at different granularities (fine and medium grain). We present a feasibility study of a new execution model for exploiting both adjacent and distant parallelism in the dynamic instruction stream. Our model, called hierarchical multithreading, uses a two-level hierarchical arrangement of processing elements. The lower level of the hierarchy exploits instruction-level parallelism and fine-grain thread-level parallelism, whereas the upper level exploits more distant parallelism. Detailed simulation studies with a cycle accurate simulator are presented, showing the feasibility of hierarchical multithreading. Conclusions are drawn about the best ways to obtain the most front the hierarchical multithreading scheme.
OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Many studies have shown that significant amounts of parallelism exist at different granularities. Execution models such as superscalar and VLIW exploit parallelism from a single thread. Multithreaded processors make a step towards exploiting parallelism from different threads, but are not geared to exploit parallelism at different granularities (fine and medium grain). We present a feasibility study of a new execution model for exploiting both adjacent and distant parallelism in the dynamic instruction stream. Our model, called hierarchical multithreading, uses a two-level hierarchical arrangement of processing elements. The lower level of the hierarchy exploits instruction-level parallelism and fine-grain thread-level parallelism, whereas the upper level exploits more distant parallelism. Detailed simulation studies with a cycle accurate simulator are presented, showing the feasibility of hierarchical multithreading. Conclusions are drawn about the best ways to obtain the most front the hierarchical multithreading scheme.
Key concepts: Multithreading, Computer science, Exploit, Parallel computing, Task parallelism, Thread (computing), Simultaneous multithreading, Instruction-level parallelism