High Performance Zero-Memory Overhead Direct Convolutions

Jiyuan Zhang, Franz Franchetti, Tze Meng Low
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:5776-5785, 2018.

Abstract

The computation of convolution layers in deep neural networks typically rely on high performance routines that trade space for time by using additional memory (either for packing purposes or required as part of the algorithm) to improve performance. The problems with such an approach are two-fold. First, these routines incur additional memory overhead which reduces the overall size of the network that can fit on embedded devices with limited memory capacity. Second, these high performance routines were not optimized for performing convolution, which means that the performance obtained is usually less than conventionally expected. In this paper, we demonstrate that direct convolution, when implemented correctly, eliminates all memory overhead, and yields performance that is between 10% to 400% times better than existing high performance implementations of convolution layers on conventional and embedded CPU architectures. We also show that a high performance direct convolution exhibits better scaling performance, i.e. suffers less performance drop, when increasing the number of threads.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-zhang18d, title = {High Performance Zero-Memory Overhead Direct Convolutions}, author = {Zhang, Jiyuan and Franchetti, Franz and Low, Tze Meng}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {5776--5785}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/zhang18d/zhang18d.pdf}, url = {http://proceedings.mlr.press/v80/zhang18d.html}, abstract = {The computation of convolution layers in deep neural networks typically rely on high performance routines that trade space for time by using additional memory (either for packing purposes or required as part of the algorithm) to improve performance. The problems with such an approach are two-fold. First, these routines incur additional memory overhead which reduces the overall size of the network that can fit on embedded devices with limited memory capacity. Second, these high performance routines were not optimized for performing convolution, which means that the performance obtained is usually less than conventionally expected. In this paper, we demonstrate that direct convolution, when implemented correctly, eliminates all memory overhead, and yields performance that is between 10% to 400% times better than existing high performance implementations of convolution layers on conventional and embedded CPU architectures. We also show that a high performance direct convolution exhibits better scaling performance, i.e. suffers less performance drop, when increasing the number of threads.} }
Endnote
%0 Conference Paper %T High Performance Zero-Memory Overhead Direct Convolutions %A Jiyuan Zhang %A Franz Franchetti %A Tze Meng Low %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-zhang18d %I PMLR %P 5776--5785 %U http://proceedings.mlr.press/v80/zhang18d.html %V 80 %X The computation of convolution layers in deep neural networks typically rely on high performance routines that trade space for time by using additional memory (either for packing purposes or required as part of the algorithm) to improve performance. The problems with such an approach are two-fold. First, these routines incur additional memory overhead which reduces the overall size of the network that can fit on embedded devices with limited memory capacity. Second, these high performance routines were not optimized for performing convolution, which means that the performance obtained is usually less than conventionally expected. In this paper, we demonstrate that direct convolution, when implemented correctly, eliminates all memory overhead, and yields performance that is between 10% to 400% times better than existing high performance implementations of convolution layers on conventional and embedded CPU architectures. We also show that a high performance direct convolution exhibits better scaling performance, i.e. suffers less performance drop, when increasing the number of threads.
APA
Zhang, J., Franchetti, F. & Low, T.M.. (2018). High Performance Zero-Memory Overhead Direct Convolutions. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:5776-5785 Available from http://proceedings.mlr.press/v80/zhang18d.html.

Related Material