Reducing the Repairing Penalty on Misspeculation in Thread-Level Speculation

Programmer Speculative execution Speculative multithreading
DOI: 10.1145/3468081.3471120 Publication Date: 2021-10-22T16:14:12Z
ABSTRACT
We have been developing a software-based Speculative Memory (SM) library to enlarge the opportunity for program parallelization. It enables tasks having possible unknown dependencies on each other be executed speculatively in parallel while dynamically inspecting among those tasks. When violation of (misspeculation) is detected, it aborts execution violating task and re-executes it. But re-execution from its beginning may bring large performance losses. To reduce such losses, therefore, this paper, we introduced checkpoint repair mechanism into SM. The idea itself not new, but implemented enabling programmer optimize overhead data saving at repairing. also present programming technique enhance quality speculative result by using task's uncommitted data. This can contribute reducing frequency aborts. evaluated our SSM sample constructing binary search tree. Consequently, could verify effectiveness both enhancement execution.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (26)
CITATIONS (2)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....