Title :
Automatic generation of tuners for intel Concurrent Collections programs
Author :
Peiyi Tang ; Moulliet, Ryan J.
Author_Institution :
Dept. of Comput. Sci., Univ. of Arkansas at Little Rock, Little Rock, AR, USA
Abstract :
Tuners of Concurrent Collections programs (CnC) are essential for the scalable performance. But writing correct codes for tuners of CnC programs is difficult, as it requires programmers to do accurate data flow analysis of the applications. In this paper, we present a template tuner for generating tuners for linear algebra CnC programs automatically. Programmers only need to write a trace function to pass to the template tuner and do not need to write any codes for the tuner callback functions. The instantiated tuner does the data flow analysis during its construction and provides all the callback functions. Performance evaluation shows that the automatically generated tuners are as efficient as hand-coded tuners.
Keywords :
concurrency control; data flow analysis; linear algebra; parallel programming; program testing; software performance evaluation; Intel concurrent collections programs; automatic tuner generation; callback functions; data flow analysis; hand-coded tuners; instantiated tuner; linear algebra CnC programs; parallel programming; performance evaluation; template tuner; trace function; Linear algebra; Tuners; Automatic Tuners; Data Flow Analysis; Intel Concurrent Collections; Parallel Programming;
Conference_Titel :
SoutheastCon 2015
Conference_Location :
Fort Lauderdale, FL
DOI :
10.1109/SECON.2015.7133026