User Tools

Site Tools


melange:papers:fall2021

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Last revision Both sides next revision
melange:papers:fall2021 [2021/10/05 13:05]
corentin
melange:papers:fall2021 [2021/10/27 13:54]
corentin add Pouchet'15
Line 121: Line 121:
   loc          = {[2020] IEEE International Conference on Cluster Computing (CLUSTER)},   loc          = {[2020] IEEE International Conference on Cluster Computing (CLUSTER)},
   url          = {https://par.nsf.gov/servlets/purl/10224953}   url          = {https://par.nsf.gov/servlets/purl/10224953}
 +}
 +
 +@inproceedings{10.1145/2684746.2689065,
 +  author       = {Li, Peng and Zhang, Peng and Pouchet, Louis-Noel and Cong, Jason},
 +  title        = {Resource-Aware Throughput Optimization for High-Level Synthesis},
 +  year         = {2015},
 +  isbn         = {9781450333153},
 +  publisher    = {Association for Computing Machinery},
 +  address      = {New York, NY, USA},
 +  url          = {https://doi.org/10.1145/2684746.2689065},
 +  doi          = {10.1145/2684746.2689065},
 +  abstract     = {With the emergence of robust high-level synthesis tools to automatically transform codes written in high-level languages into RTL implementations, the programming productivity when synthesising accelerators improves significantly. However, although the state-of-the-art high-level synthesis tools can offer high-quality designs for simple nested loop kernels, there is still a significant performance gap between the synthesized and the optimal design for real world complex applications with multiple loops.In this work we first demonstrate that maximizing the throughput of each individual loop is not always the most efficient approach to achieving the maximum system-level throughput. More area efficient non-fully pipelined design variants may outperform the fully-pipelined version by enabling larger degrees of parallelism. We develop an algorithm to determine the optimal resource usage and initiation intervals for each loop in the applications to achieve maximum throughput within a given area budget. We report experimental results on eight applications, showing an average of 31% performance speedup over state-of-the-art HLS solutions.},
 +  booktitle    = {Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays},
 +  pages        = {200–209},
 +  numpages     = {10},
 +  keywords     = {resource sharing, area constraint, throughput optimization, high-level synthesis},
 +  location     = {Monterey, California, USA},
 +  series       = {FPGA '15},
 +  loc          = {Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays}
 } }
  
melange/papers/fall2021.txt · Last modified: 2021/10/27 13:55 by corentin