2016 Fourth International Symposium on Computing and Networking (2016)
Nov. 22, 2016 to Nov. 25, 2016
We have proposed a processor which can exploit valuelocalityinprogramsbyautomaticallyapplyingcomputation reuse. The processor which we call auto-memoization processor dynamically detects functions and loop iterations as reusable blocks, and stores their input sequences and results into a lookup table. When the current input sequence matches one of the stored input sequences on the table, the stored result associated with the matched input sequence is written back to the registers and caches. In the previous implementation, a part of the table is implemented with a CAM for achieving associative search for input matching with small overhead. However, CAMs consumeconsiderablylargeenergy, areaandmanufacturingcost. Therefore, CAM size should be as small as possible for improving practicality of the auto-memoization processor. In this paper, we propose a low-power implementation of the auto-memoization processor by utilizing a RAM and a Bloom filter. The result of the simulation experiment shows that power consumption of the table is reduced by 67.5% at a maximum and by 50.4% on average.
Impedance matching, Random access memory, Indexes, Manufacturing, Parallel processing, Engines
M. Fujii, Y. Sato, T. Tsumura and Y. Nakashima, "Exploiting Bloom Filters for Saving Power Consumption of Auto-Memoization Processor," 2016 Fourth International Symposium on Computing and Networking(CANDAR), Hiroshima, Japan, 2016, pp. 354-360.