Name
Abstract | ||
|---|---|---|
Dynamic parallelism (DP) is a promising feature for GPUs, which allows on-demand spawning of kernels on the GPU without any CPU intervention. However, this feature has two major drawbacks. First, the launching of GPU kernels can incur significant performance penalties. Second, dynamically-generated kernels are not always able to efficiently utilize the GPU cores due to hardware-limits. To address these two concerns cohesively, we propose SPAWN, a runtime framework that controls the dynamically-generated kernels, thereby directly reducing the associated launch overheads and queuing latency. Moreover, it allows a better mix of dynamically-generated and original (parent) kernels for the scheduler to effectively hide the remaining overheads and improve the utilization of the GPU resources. Our results show that, across 13 benchmarks, SPAWN achieves 69% and 57% speedup over the flat (non-DP) implementation and baseline DP, respectively. |
Year | DOI | Venue |
|---|---|---|
2017 | 10.1109/HPCA.2017.14 | 2017 IEEE International Symposium on High Performance Computer Architecture (HPCA) |
Keywords | Field | DocType |
controlled kernel launch,dynamic parallelism,kernel on-demand spawning,GPU kernels,dynamically-generated kernels,SPAWN runtime framework,scheduler,GPU resource utilization | Kernel (linear algebra),Instruction set,Computer science,Latency (engineering),Parallel processing,Parallel computing,Real-time computing,Queueing theory,Benchmark (computing),Overhead (business),Speedup | Conference |
ISSN | ISBN | Citations |
1530-0897 | 978-1-5090-4986-8 | 10 |
PageRank | References | Authors |
0.50 | 23 | 9 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Xulong Tang | 1 | 128 | 7.49 |
| Ashutosh Pattnaik | 2 | 113 | 4.70 |
| Huaipan Jiang | 3 | 12 | 2.25 |
| Onur Kayıran | 4 | 356 | 13.47 |
| Adwait Jog | 5 | 568 | 23.32 |
| Sreepathi Pai | 6 | 177 | 9.58 |
| Mohamed Ibrahim | 7 | 453 | 35.03 |
| Mahmut T. Kandemir | 8 | 7371 | 568.54 |
| Chita R. Das | 9 | 1467 | 80.03 |