Operation and data mapping for CGRAS with multi-bank memory

Yongjoo Kim, Jongeun Lee, Aviral Shrivastava, Yunheung Paek

Research output: Contribution to journalArticle

9 Citations (Scopus)

Abstract

Coarse Grain Reconfigurable Architectures (CGRAs) promise high performance at high power efficiency. They fulfil this promise by keeping the hardware extremely simple, and moving the complexity to application mapping. One major challenge comes in the form of data mapping. For reasons of power-efficiency and complexity, CGRAs use multi-bank local memory, and a row of PEs share memory access. In order for each row of the PEs to access any memory bank, there is a hardware arbiter between the memory requests generated by the PEs and the banks of the local memory. However, a fundamental restriction remains that a bank cannot be accessed by two different PEs at the same time. We propose to meet this challenge by mapping application operations onto PEs and data into memory banks in a way that avoids such conflicts. Our experimental results on kernels from multimedia benchmarks demonstrate that our local memory-aware compilation approach can generate mappings that are up to 40% better in performance (17.3% on average) compared to a memory-unaware scheduler.

Original languageEnglish (US)
Pages (from-to)17-25
Number of pages9
JournalACM SIGPLAN Notices
Volume45
Issue number4
StatePublished - Apr 2010

Fingerprint

Data storage equipment
Reconfigurable architectures
Computer hardware
Hardware

Keywords

  • Arbiter
  • Bank conflict
  • Coarse-grained reconfigurable architecture
  • Compilation
  • Multi-bank memory

ASJC Scopus subject areas

  • Computer Science(all)

Cite this

Operation and data mapping for CGRAS with multi-bank memory. / Kim, Yongjoo; Lee, Jongeun; Shrivastava, Aviral; Paek, Yunheung.

In: ACM SIGPLAN Notices, Vol. 45, No. 4, 04.2010, p. 17-25.

Research output: Contribution to journalArticle

Kim, Yongjoo ; Lee, Jongeun ; Shrivastava, Aviral ; Paek, Yunheung. / Operation and data mapping for CGRAS with multi-bank memory. In: ACM SIGPLAN Notices. 2010 ; Vol. 45, No. 4. pp. 17-25.
@article{4bb918aa7323415391d328aa61bd6f8e,
title = "Operation and data mapping for CGRAS with multi-bank memory",
abstract = "Coarse Grain Reconfigurable Architectures (CGRAs) promise high performance at high power efficiency. They fulfil this promise by keeping the hardware extremely simple, and moving the complexity to application mapping. One major challenge comes in the form of data mapping. For reasons of power-efficiency and complexity, CGRAs use multi-bank local memory, and a row of PEs share memory access. In order for each row of the PEs to access any memory bank, there is a hardware arbiter between the memory requests generated by the PEs and the banks of the local memory. However, a fundamental restriction remains that a bank cannot be accessed by two different PEs at the same time. We propose to meet this challenge by mapping application operations onto PEs and data into memory banks in a way that avoids such conflicts. Our experimental results on kernels from multimedia benchmarks demonstrate that our local memory-aware compilation approach can generate mappings that are up to 40{\%} better in performance (17.3{\%} on average) compared to a memory-unaware scheduler.",
keywords = "Arbiter, Bank conflict, Coarse-grained reconfigurable architecture, Compilation, Multi-bank memory",
author = "Yongjoo Kim and Jongeun Lee and Aviral Shrivastava and Yunheung Paek",
year = "2010",
month = "4",
language = "English (US)",
volume = "45",
pages = "17--25",
journal = "SIGPLAN Notices (ACM Special Interest Group on Programming Languages)",
issn = "1523-2867",
publisher = "Association for Computing Machinery (ACM)",
number = "4",

}

TY - JOUR

T1 - Operation and data mapping for CGRAS with multi-bank memory

AU - Kim, Yongjoo

AU - Lee, Jongeun

AU - Shrivastava, Aviral

AU - Paek, Yunheung

PY - 2010/4

Y1 - 2010/4

N2 - Coarse Grain Reconfigurable Architectures (CGRAs) promise high performance at high power efficiency. They fulfil this promise by keeping the hardware extremely simple, and moving the complexity to application mapping. One major challenge comes in the form of data mapping. For reasons of power-efficiency and complexity, CGRAs use multi-bank local memory, and a row of PEs share memory access. In order for each row of the PEs to access any memory bank, there is a hardware arbiter between the memory requests generated by the PEs and the banks of the local memory. However, a fundamental restriction remains that a bank cannot be accessed by two different PEs at the same time. We propose to meet this challenge by mapping application operations onto PEs and data into memory banks in a way that avoids such conflicts. Our experimental results on kernels from multimedia benchmarks demonstrate that our local memory-aware compilation approach can generate mappings that are up to 40% better in performance (17.3% on average) compared to a memory-unaware scheduler.

AB - Coarse Grain Reconfigurable Architectures (CGRAs) promise high performance at high power efficiency. They fulfil this promise by keeping the hardware extremely simple, and moving the complexity to application mapping. One major challenge comes in the form of data mapping. For reasons of power-efficiency and complexity, CGRAs use multi-bank local memory, and a row of PEs share memory access. In order for each row of the PEs to access any memory bank, there is a hardware arbiter between the memory requests generated by the PEs and the banks of the local memory. However, a fundamental restriction remains that a bank cannot be accessed by two different PEs at the same time. We propose to meet this challenge by mapping application operations onto PEs and data into memory banks in a way that avoids such conflicts. Our experimental results on kernels from multimedia benchmarks demonstrate that our local memory-aware compilation approach can generate mappings that are up to 40% better in performance (17.3% on average) compared to a memory-unaware scheduler.

KW - Arbiter

KW - Bank conflict

KW - Coarse-grained reconfigurable architecture

KW - Compilation

KW - Multi-bank memory

UR - http://www.scopus.com/inward/record.url?scp=77951219640&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77951219640&partnerID=8YFLogxK

M3 - Article

VL - 45

SP - 17

EP - 25

JO - SIGPLAN Notices (ACM Special Interest Group on Programming Languages)

JF - SIGPLAN Notices (ACM Special Interest Group on Programming Languages)

SN - 1523-2867

IS - 4

ER -