.png){kind=icon}
.png){kind=icon}
Return
A Task-Parallel Pipeline Programming Model with Token Dependency
.png){kind=icon}
PRE
AI
DOI：10.1007/978-3-031-97196-9_1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Task-parallel pipeline framework explores pipeline parallelism in applications and is critical in many parallel and heterogeneous areas, such as VLSI static timing analysis and data similarity search. However, existing solutions only deal with certain types of applications in which data dependency exists between preceding data and succeeding data in a forward direction. Some applications, such as video encoding, exhibit data dependency in both forward and backward directions and cannot be processed with existing solutions. To address the limitation, we introduce a token dependency-aware pipeline framework. Our framework associates each data element with a token as its identifier, supports explicit definitions of forward and backward token dependency with an expressive programming model, resolves token dependency using simple data structures, and schedules tokens with lightweight atomic counters. We have evaluated the framework on applications that exhibit both forward and backward token dependency. For example, our framework is 8.6% faster than PARSEC's implementation in x.264 video encoding applications.
Keywords:
Task-parallel pipeline
Token dependency
Video encoding
Parallel programming
Data dependency
Journal
A
IF:
0
Papers:
13
Citations:
0
