DRAW

Investigating benefits of adaptive fetch group size on GPU

Myung Kuk Yoon, Yunho Oh, Sangpil Lee, Seung Hun Kim, Deokho Kim, Won Woo Ro

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Citations (Scopus)

Abstract

Previously, hiding operation stalls is one of the important issues to suppress performance degradation of Graphics Processing Units (GPUs). In this paper, we first conduct a detailed study of factors affecting the operation stalls in terms of the fetch group size on the warp scheduler. Throughout this paper, we find that the size of fetch group is highly involved in hiding various types of operation stalls. The short latency stalls can be hidden by issuing other available warps from the same fetch group. Therefore, the short latency stalls may not be hidden well under small sized fetch group since the group has the limited number of issuable warps to hide stalls. On the contrary, the long latency stalls can be hidden by dividing warps into multiple fetch groups. The scheduler switches the fetch groups when the warps in each fetch group reach the long latency memory operation point. Therefore, the stalls may not be hidden well at the large sized fetch group. Increasing the size of fetch group reduces the number of fetch groups to hide the stalls. In addition, the load/store unit stalls are caused by the limited hardware resources to handle the memory operations. To hide all these stalls effectively, we propose a Dynamic Resizing on Active Warps (DRAW) scheduler which adjusts the size of active fetch group. From the evaluation results, DRAW scheduler reduces an average of 16.3% of stall cycles and improves an average performance of 11.3% compared to the conventional two-level warp scheduler.

Original languageEnglish
Title of host publicationISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages183-192
Number of pages10
ISBN (Electronic)9781479919567
DOIs
Publication statusPublished - 2015 Apr 27
Event2015 15th IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2015 - Philidelphia, United States
Duration: 2015 Mar 292015 Mar 31

Publication series

NameISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software

Other

Other2015 15th IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2015
CountryUnited States
CityPhilidelphia
Period15/3/2915/3/31

Fingerprint

Data storage equipment
Computer hardware
Switches
Degradation
Graphics processing unit

All Science Journal Classification (ASJC) codes

  • Software

Cite this

Yoon, M. K., Oh, Y., Lee, S., Kim, S. H., Kim, D., & Ro, W. W. (2015). DRAW: Investigating benefits of adaptive fetch group size on GPU. In ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software (pp. 183-192). [7095804] (ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISPASS.2015.7095804
Yoon, Myung Kuk ; Oh, Yunho ; Lee, Sangpil ; Kim, Seung Hun ; Kim, Deokho ; Ro, Won Woo. / DRAW : Investigating benefits of adaptive fetch group size on GPU. ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software. Institute of Electrical and Electronics Engineers Inc., 2015. pp. 183-192 (ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software).
@inproceedings{7e40a70278944b3a8ea74a5df1baef1f,
title = "DRAW: Investigating benefits of adaptive fetch group size on GPU",
abstract = "Previously, hiding operation stalls is one of the important issues to suppress performance degradation of Graphics Processing Units (GPUs). In this paper, we first conduct a detailed study of factors affecting the operation stalls in terms of the fetch group size on the warp scheduler. Throughout this paper, we find that the size of fetch group is highly involved in hiding various types of operation stalls. The short latency stalls can be hidden by issuing other available warps from the same fetch group. Therefore, the short latency stalls may not be hidden well under small sized fetch group since the group has the limited number of issuable warps to hide stalls. On the contrary, the long latency stalls can be hidden by dividing warps into multiple fetch groups. The scheduler switches the fetch groups when the warps in each fetch group reach the long latency memory operation point. Therefore, the stalls may not be hidden well at the large sized fetch group. Increasing the size of fetch group reduces the number of fetch groups to hide the stalls. In addition, the load/store unit stalls are caused by the limited hardware resources to handle the memory operations. To hide all these stalls effectively, we propose a Dynamic Resizing on Active Warps (DRAW) scheduler which adjusts the size of active fetch group. From the evaluation results, DRAW scheduler reduces an average of 16.3{\%} of stall cycles and improves an average performance of 11.3{\%} compared to the conventional two-level warp scheduler.",
author = "Yoon, {Myung Kuk} and Yunho Oh and Sangpil Lee and Kim, {Seung Hun} and Deokho Kim and Ro, {Won Woo}",
year = "2015",
month = "4",
day = "27",
doi = "10.1109/ISPASS.2015.7095804",
language = "English",
series = "ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "183--192",
booktitle = "ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software",
address = "United States",

}

Yoon, MK, Oh, Y, Lee, S, Kim, SH, Kim, D & Ro, WW 2015, DRAW: Investigating benefits of adaptive fetch group size on GPU. in ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software., 7095804, ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software, Institute of Electrical and Electronics Engineers Inc., pp. 183-192, 2015 15th IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2015, Philidelphia, United States, 15/3/29. https://doi.org/10.1109/ISPASS.2015.7095804

DRAW : Investigating benefits of adaptive fetch group size on GPU. / Yoon, Myung Kuk; Oh, Yunho; Lee, Sangpil; Kim, Seung Hun; Kim, Deokho; Ro, Won Woo.

ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software. Institute of Electrical and Electronics Engineers Inc., 2015. p. 183-192 7095804 (ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - DRAW

T2 - Investigating benefits of adaptive fetch group size on GPU

AU - Yoon, Myung Kuk

AU - Oh, Yunho

AU - Lee, Sangpil

AU - Kim, Seung Hun

AU - Kim, Deokho

AU - Ro, Won Woo

PY - 2015/4/27

Y1 - 2015/4/27

N2 - Previously, hiding operation stalls is one of the important issues to suppress performance degradation of Graphics Processing Units (GPUs). In this paper, we first conduct a detailed study of factors affecting the operation stalls in terms of the fetch group size on the warp scheduler. Throughout this paper, we find that the size of fetch group is highly involved in hiding various types of operation stalls. The short latency stalls can be hidden by issuing other available warps from the same fetch group. Therefore, the short latency stalls may not be hidden well under small sized fetch group since the group has the limited number of issuable warps to hide stalls. On the contrary, the long latency stalls can be hidden by dividing warps into multiple fetch groups. The scheduler switches the fetch groups when the warps in each fetch group reach the long latency memory operation point. Therefore, the stalls may not be hidden well at the large sized fetch group. Increasing the size of fetch group reduces the number of fetch groups to hide the stalls. In addition, the load/store unit stalls are caused by the limited hardware resources to handle the memory operations. To hide all these stalls effectively, we propose a Dynamic Resizing on Active Warps (DRAW) scheduler which adjusts the size of active fetch group. From the evaluation results, DRAW scheduler reduces an average of 16.3% of stall cycles and improves an average performance of 11.3% compared to the conventional two-level warp scheduler.

AB - Previously, hiding operation stalls is one of the important issues to suppress performance degradation of Graphics Processing Units (GPUs). In this paper, we first conduct a detailed study of factors affecting the operation stalls in terms of the fetch group size on the warp scheduler. Throughout this paper, we find that the size of fetch group is highly involved in hiding various types of operation stalls. The short latency stalls can be hidden by issuing other available warps from the same fetch group. Therefore, the short latency stalls may not be hidden well under small sized fetch group since the group has the limited number of issuable warps to hide stalls. On the contrary, the long latency stalls can be hidden by dividing warps into multiple fetch groups. The scheduler switches the fetch groups when the warps in each fetch group reach the long latency memory operation point. Therefore, the stalls may not be hidden well at the large sized fetch group. Increasing the size of fetch group reduces the number of fetch groups to hide the stalls. In addition, the load/store unit stalls are caused by the limited hardware resources to handle the memory operations. To hide all these stalls effectively, we propose a Dynamic Resizing on Active Warps (DRAW) scheduler which adjusts the size of active fetch group. From the evaluation results, DRAW scheduler reduces an average of 16.3% of stall cycles and improves an average performance of 11.3% compared to the conventional two-level warp scheduler.

UR - http://www.scopus.com/inward/record.url?scp=84937460070&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84937460070&partnerID=8YFLogxK

U2 - 10.1109/ISPASS.2015.7095804

DO - 10.1109/ISPASS.2015.7095804

M3 - Conference contribution

T3 - ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software

SP - 183

EP - 192

BT - ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software

PB - Institute of Electrical and Electronics Engineers Inc.

ER -

Yoon MK, Oh Y, Lee S, Kim SH, Kim D, Ro WW. DRAW: Investigating benefits of adaptive fetch group size on GPU. In ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software. Institute of Electrical and Electronics Engineers Inc. 2015. p. 183-192. 7095804. (ISPASS 2015 - IEEE International Symposium on Performance Analysis of Systems and Software). https://doi.org/10.1109/ISPASS.2015.7095804