power driven dnn dataflow optimization on fpga
play

Power-Driven DNN Dataflow Optimization on FPGA Qi Sun 1 , Tinghuan - PowerPoint PPT Presentation

Dec 01, 2023 •225 likes •452 views

Power-Driven DNN Dataflow Optimization on FPGA Qi Sun 1 , Tinghuan Chen 1 , Jin Miao 2 , Bei Yu 1 1 The Chinese University of Hong Kong 2 Cadence Design Systems, Inc. 1 / 22 FPGA Targeted DNN Accelerator Design Flow Dataflow Synthesis HLS Model

  1. Power-Driven DNN Dataflow Optimization on FPGA Qi Sun 1 , Tinghuan Chen 1 , Jin Miao 2 , Bei Yu 1 1 The Chinese University of Hong Kong 2 Cadence Design Systems, Inc. 1 / 22

  2. FPGA Targeted DNN Accelerator Design Flow Dataflow Synthesis HLS Model DNN Model Design Space & Description Exploration Deployment FPGA Architecture Design ◮ Design power efficient dataflow with limited latency loss ◮ Enabled by proposed power modeling and a hierarchical strategy 2 / 22

  3. <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> Dataflow Optimization Basic Techniques for Loop Nest Optimization for ��� in range(0, M): for row in range(0, H): ◮ Notations: for col in range(0, W): N : # input channel for ti in range(0, N): M : # output channel for k1 in range(0, K): for k2 in range(0, K): K : kernel size OUT to,row,col +=WT to,ti,k1,k2 H : height of feature × IN ti,row+k1,col+k2 <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> <latexit sha1_base64="S8fEYyO32q/PyIaGQOLFJBshd6s=">AB7XicbVDLSgNBEOz1GeMr6tHLYBA8hV0R9Bjw4jGCeUCyhNnJbDJmdmaZ6RXCkn/w4kERr/6PN/GSbIHTSxoKq6e6KUiks+v63t7a+sbm1Xdop7+7tHxWjo5bVmeG8SbTUptORC2XQvEmCpS8kxpOk0jydjS+nfntJ26s0OoBJykPEzpUIhaMopNaPRQJt/1K1a/5c5BVEhSkCgUa/cpXb6BZlnCFTFJru4GfYphTg4JPi3MstTysZ0yLuOKuqWhPn82ik5d8qAxNq4Ukjm6u+JnCbWTpLIdSYUR3bZm4n/ed0M45swFyrNkCu2WBRnkqAms9fJQBjOUE4cocwIdythI2oQxdQ2YUQL+8SlqXtcCvBfdX1bpfxFGCUziDCwjgGupwBw1oAoNHeIZXePO09+K9ex+L1jWvmDmBP/A+fwCwY8i</latexit> × W : width of feature A 6-level loop convolutional layer 3 / 22

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Power Optimization of FPGA Interconnect Via Circuit and CAD Techniques Safeen Huda and Jason

Power Optimization of FPGA Interconnect Via Circuit and CAD Techniques Safeen Huda and Jason

Power Optimization of FPGA Interconnect Via Circuit and CAD Techniques Safeen Huda and Jason Anderson International Symposium on Physical Design Santa Rosa, CA, April 6, 2016 1 Motivation * Google data centre FPGA power increasingly

717 views • 55 slides
Noise Driven In Package Decoupling Capacitor Optimization for Power Integrity Jun Chen and Lei

Noise Driven In Package Decoupling Capacitor Optimization for Power Integrity Jun Chen and Lei

Noise Driven In Package Decoupling Capacitor Optimization for Power Integrity Jun Chen and Lei He Design Automation Laboratory, UCLA Outline Introduction Electrical models Incremental impedance computation and noise computation

448 views • 27 slides
Status for BUFFER MANAGEMENT Firmware FPGA Centric Dataflow Erdem Motuk 17 October 2019

Status for BUFFER MANAGEMENT Firmware FPGA Centric Dataflow Erdem Motuk 17 October 2019

Status for BUFFER MANAGEMENT Firmware FPGA Centric Dataflow Erdem Motuk 17 October 2019 Current status Immediate plans Proposed changes 2 The buffer management block has been developed and debugged independently from the other

372 views • 12 slides
A UTOMATIC GENERATION OF DATAFLOW - BASED LOW - POWER E RECONFIGURABLE SYSTEMS O L A B CPS D

A UTOMATIC GENERATION OF DATAFLOW - BASED LOW - POWER E RECONFIGURABLE SYSTEMS O L A B CPS D

Tiziana Fanni University of Cagliari A UTOMATIC GENERATION OF DATAFLOW - BASED LOW - POWER E RECONFIGURABLE SYSTEMS O L A B CPS D ESIGN F ROM CONCEPT TO IMPLEMENTATION Cyber Physical Systems CGR approach for high run-time adaptivity Complex

299 views • 15 slides
Dataflow analysis Discovering Global Live Ranges of Variables cs4713 1 Optimization and

Dataflow analysis Discovering Global Live Ranges of Variables cs4713 1 Optimization and

Dataflow analysis Discovering Global Live Ranges of Variables cs4713 1 Optimization and analysis Requirement for optimizations Correctness (safety) must preserve the meaning of the input computation Profitability must

251 views • 10 slides
Scalable Distributed Training with Parameter Hub: a whirlwind tour TVM Stack Optimization

Scalable Distributed Training with Parameter Hub: a whirlwind tour TVM Stack Optimization

Scalable Distributed Training with Parameter Hub: a whirlwind tour TVM Stack Optimization High-Level Differentiable IR AutoTVM Tensor Expression IR LLVM, CUDA, Metal VTA AutoVTA Edge Cloud Hardware ASIC FPGA FPGA Fleet

1.19k views • 73 slides
FPGA What is a FPGA? How FPGAs work How do they work? Manufacturers

FPGA What is a FPGA? How FPGAs work How do they work? Manufacturers

2/21/2012 Content FPGA What is a FPGA? How FPGAs work How do they work? Manufacturers Distributed RAM History FPGA vs ASIC FPGA and Microprocessors Alternatives to FPGAs ETI135, Advanced Digital IC Design

123 views • 10 slides
Analyses, Hardware/Software Compilation, Code Optimization for Complex Dataflow HPC Applications

Analyses, Hardware/Software Compilation, Code Optimization for Complex Dataflow HPC Applications

Analyses, Hardware/Software Compilation, Code Optimization for Complex Dataflow HPC Applications CASH team proposal (Compilation and Analyses for Software and Hardware) Matthieu Moy and Christophe Alias and Laure Gonnord University of Lyon 1 /

264 views • 24 slides
A Low-Power MTJ-Based Nonvolatile FPGA Using Self-Terminated Logic-In-Memory Structure Daisuke

A Low-Power MTJ-Based Nonvolatile FPGA Using Self-Terminated Logic-In-Memory Structure Daisuke

26th International Conference on Field-Programmable Logic and Applications 29th August 2nd September, 2016, Lausanne, Switzerland Session S4b 31st August 10:30-12:35 A Low-Power MTJ-Based Nonvolatile FPGA Using Self-Terminated

108 views • 9 slides
The nextpnr FOSS FPGA place-and-route tool Clifford Wolf Symbiotic EDA FOSS FPGA PnR VPR

The nextpnr FOSS FPGA place-and-route tool Clifford Wolf Symbiotic EDA FOSS FPGA PnR VPR

The nextpnr FOSS FPGA place-and-route tool Clifford Wolf Symbiotic EDA FOSS FPGA PnR VPR (Versatile Place-and-Route) Over 20 years old (1997). Timing driven. Portable. Focuses more on architecture exploration via architecture XML

489 views • 20 slides
Naiad (Timely Dataflow) &amp; Streaming Systems CS 848: Models and Applications of Distributed

Naiad (Timely Dataflow) &amp; Streaming Systems CS 848: Models and Applications of Distributed

Naiad (Timely Dataflow) &amp; Streaming Systems CS 848: Models and Applications of Distributed Data Systems Mon, Nov 7th 2016 Amine Mhedhbi What is Timely Dataflow ?! What is its significance? Dataflow ?! Dataflow?! Dataflow?!

1.08k views • 70 slides
WWW.FPGA What is an FPGA? Field Programmable Gate Array Introduction to FPGA designs

WWW.FPGA What is an FPGA? Field Programmable Gate Array Introduction to FPGA designs

WWW.FPGA What is an FPGA? Field Programmable Gate Array Introduction to FPGA designs Generic logic cells + Programmable switches Why do we use it? High performance &amp; Flexible Shorter time to market Joachim Rodrigues

308 views • 8 slides
The Era of SoC FPGAs Nizar Abdallah, Ph.D. Workshop on FPGA Design for Scientific Instrumentation

The Era of SoC FPGAs Nizar Abdallah, Ph.D. Workshop on FPGA Design for Scientific Instrumentation

Power Matters The Era of SoC FPGAs Nizar Abdallah, Ph.D. Workshop on FPGA Design for Scientific Instrumentation and Computing International Centre for Theoretical Physics November 2017 Outline Introduction SoC FPGA Architectures: an

938 views • 78 slides
EVENT-DRIVEN AND DATA-DRIVEN CONTROL AND OPTIMIZATION IN CYBER-PHYSICAL SYSTEMS C. G. Cassandras

EVENT-DRIVEN AND DATA-DRIVEN CONTROL AND OPTIMIZATION IN CYBER-PHYSICAL SYSTEMS C. G. Cassandras

EVENT-DRIVEN AND DATA-DRIVEN CONTROL AND OPTIMIZATION IN CYBER-PHYSICAL SYSTEMS C. G. Cassandras Division of Systems Engineering Dept. of Electrical and Computer Engineering Center for Information and Systems Engineering Boston University

1.05k views • 50 slides
CO444H Dataflow Dataflow frameworks Ben Livshits Masters Projects Available 1. Crashes to

CO444H Dataflow Dataflow frameworks Ben Livshits Masters Projects Available 1. Crashes to

Static analysis CO444H Dataflow Dataflow frameworks Ben Livshits Masters Projects Available 1. Crashes to exploits 2. Pointer analysis for JavaScript 3. Private data management languages 4. Programming robots to assemble IKEA furniture

897 views • 57 slides
Characterizing Power Distribution Attacks in Multi-User FPGA Environments George Provelengios,

Characterizing Power Distribution Attacks in Multi-User FPGA Environments George Provelengios,

Characterizing Power Distribution Attacks in Multi-User FPGA Environments George Provelengios, Daniel Holcomb, and Russell Tessier Department of Electrical and Computer Engineering University of Massachusetts Amherst (Funded by a grand from

517 views • 16 slides
The Explosion in Neural Network Hardware USC Friday 19 th April Trevor Mudge Bredt Family

The Explosion in Neural Network Hardware USC Friday 19 th April Trevor Mudge Bredt Family

1 The Explosion in Neural Network Hardware USC Friday 19 th April Trevor Mudge Bredt Family Professor of Computer Science and Engineering The University of Michigan, Ann Arbor 1 What Just Happened? 2 For years the common wisdom was that

769 views • 42 slides
Towards General-Purpose Acceleration by Exploiting Common Data- Dependence Forms Vidushi Dadu ,

Towards General-Purpose Acceleration by Exploiting Common Data- Dependence Forms Vidushi Dadu ,

Towards General-Purpose Acceleration by Exploiting Common Data- Dependence Forms Vidushi Dadu , Jian Weng, Sihao Liu, Tony Nowatzki UCLA MICRO 2019 Challenging trade-off in domain-specific and domain-agnostic acceleration CPU Maximum

885 views • 55 slides
hendren@cs.mcgill.ca COMP 520 Winter 2016 Domain-Specific Languages - OncoTime (2) Designing

hendren@cs.mcgill.ca COMP 520 Winter 2016 Domain-Specific Languages - OncoTime (2) Designing

COMP 520 Winter 2016 Domain-Specific Languages - OncoTime (1) Domain Specific Languages COMP 520: Compiler Design (4 credits) Professor Laurie Hendren hendren@cs.mcgill.ca COMP 520 Winter 2016 Domain-Specific Languages - OncoTime (2)

589 views • 42 slides
Building Genomic Medicine Capability Challenges and opportunities of big data Andy Futreal MD

Building Genomic Medicine Capability Challenges and opportunities of big data Andy Futreal MD

Building Genomic Medicine Capability Challenges and opportunities of big data Andy Futreal MD Anderson Cancer Center Personalised/Stratified/Precision Medicine for Cancer Personalised medicine will enable the much needed paradigm shift in

133 views • 12 slides
Rewriting in Artin groups Sarah Rees University of Newcastle Paul Schupp Fest, Stevens Institute

Rewriting in Artin groups Sarah Rees University of Newcastle Paul Schupp Fest, Stevens Institute

Rewriting in Artin groups Sarah Rees University of Newcastle Paul Schupp Fest, Stevens Institute of Technology, June 2017 Introduction This is about the word problem in Artin groups, focusing on recent attempts to find a uniform approach.

1.5k views • 119 slides
Chapter 12: The Regression Line We already know that the regression line goes through the point

Chapter 12: The Regression Line We already know that the regression line goes through the point

Chapter 12: The Regression Line We already know that the regression line goes through the point of ____________ and has slope = _________. The equation for the line is Y = intercept + (slope) X where slope = and intercept = ave Y

433 views • 18 slides
SIMD Single Instruction Multiple Data Parallelism through simultaneous operations on different

SIMD Single Instruction Multiple Data Parallelism through simultaneous operations on different

SIMD Single Instruction Multiple Data Parallelism through simultaneous operations on different data Lecture 12: SIMD-machines &amp; data parallelism, Fine grain parallelism dependency analysis for automatic vectorizing and parallelizing of

502 views • 7 slides
Unit 3: Foundations for inference Lecture 3: Decision errors, significance levels, sample size,

Unit 3: Foundations for inference Lecture 3: Decision errors, significance levels, sample size,

Unit 3: Foundations for inference Lecture 3: Decision errors, significance levels, sample size, power, and bootstrapping Statistics 101 Thomas Leininger June 3, 2013 Decision errors Decision errors 1 Type 1 and Type 2 errors Error rates

735 views • 72 slides

More recommend