TY - GEN
T1 - Enhancing the performance of a multiplayer game by using a parallelizing compiler
AU - Al-Dosary, Yasir I.M.
AU - Kimura, Keiji
AU - Kasahara, Hironori
AU - Narita, Seinosuke
PY - 2012/11/19
Y1 - 2012/11/19
N2 - Video Games have been a very popular form of digital entertainment in recent years. They have been delivered in state of the art technologies that include multi-core processors that are known to be the leading contributor in enhancing the performance of computer applications. Since parallel programming is a difficult technology to implement, that field in Video Games is still rich with areas for advancements. This paper investigates performance enhancement in Video Games when using parallelizing compilers and the difficulties involved in achieving that. This experiment conducts several stages in attempting to parallelize a well-renowned sequentially written Video Game called ioquake3. First, the Game is profiled for discovering bottlenecks, then examined by hand on how much parallelism could be extracted from those bottlenecks, and what sort of hazards exist in delivering a parallel-friendly version of ioquake3. Then, the Game code is rewritten into a hazard-free version while also modified to comply with the Parallelizable-C rules, which crucially aid parallelizing compilers in extracting parallelism. Next, the program is compiled using a parallelizing compiler called OSCAR (Optimally Scheduled Advanced Multiprocessor) to produce a parallel version of ioquake3. Finally, the performance of the newly produced parallel version of ioquake3 on a Multi-core platform is analyzed. The following is found: (1) the parallelized game by the compiler from the revised sequential program of the game is found to achieve a 5.1 faster performance at 8-threads than original one on an IBM Power 5+ machine that is equipped with 8-cores, and (2) hazards are caused by thread contentions over globally shared data, and as well as thread private data, and (3) AI driven players are represented very similarly to Human players inside ioquake3 engine, which gives an estimation of the costs for parallelizing Human driven sessions, and (4) 70% of the costs of the experiment is spent in analyzing ioquake3 code, 30% in implementing the changes in the code.
AB - Video Games have been a very popular form of digital entertainment in recent years. They have been delivered in state of the art technologies that include multi-core processors that are known to be the leading contributor in enhancing the performance of computer applications. Since parallel programming is a difficult technology to implement, that field in Video Games is still rich with areas for advancements. This paper investigates performance enhancement in Video Games when using parallelizing compilers and the difficulties involved in achieving that. This experiment conducts several stages in attempting to parallelize a well-renowned sequentially written Video Game called ioquake3. First, the Game is profiled for discovering bottlenecks, then examined by hand on how much parallelism could be extracted from those bottlenecks, and what sort of hazards exist in delivering a parallel-friendly version of ioquake3. Then, the Game code is rewritten into a hazard-free version while also modified to comply with the Parallelizable-C rules, which crucially aid parallelizing compilers in extracting parallelism. Next, the program is compiled using a parallelizing compiler called OSCAR (Optimally Scheduled Advanced Multiprocessor) to produce a parallel version of ioquake3. Finally, the performance of the newly produced parallel version of ioquake3 on a Multi-core platform is analyzed. The following is found: (1) the parallelized game by the compiler from the revised sequential program of the game is found to achieve a 5.1 faster performance at 8-threads than original one on an IBM Power 5+ machine that is equipped with 8-cores, and (2) hazards are caused by thread contentions over globally shared data, and as well as thread private data, and (3) AI driven players are represented very similarly to Human players inside ioquake3 engine, which gives an estimation of the costs for parallelizing Human driven sessions, and (4) 70% of the costs of the experiment is spent in analyzing ioquake3 code, 30% in implementing the changes in the code.
KW - OSCAR
KW - Quake
KW - Video Games
KW - ioquake3
KW - parallel Computing
KW - parallelizing compilers
UR - http://www.scopus.com/inward/record.url?scp=84869006395&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84869006395&partnerID=8YFLogxK
U2 - 10.1109/CGames.2012.6314554
DO - 10.1109/CGames.2012.6314554
M3 - Conference contribution
AN - SCOPUS:84869006395
SN - 9781467311199
T3 - Proceedings of CGAMES'2012 USA - 17th International Conference on Computer Games: AI, Animation, Mobile, Interactive Multimedia, Educational and Serious Games
SP - 67
EP - 75
BT - Proceedings of CGAMES'2012 USA - 17th International Conference on Computer Games
T2 - 17th International Conference on Computer Games, CGAMES 2012
Y2 - 30 July 2012 through 1 August 2012
ER -