國立虎尾科技大學 |

Compiling Deep Learning Kernels to Locality-Aware Dataflow.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Compiling Deep Learning Kernels to Locality-Aware Dataflow./
Author:	Zhao, Tian.
Description:	1 online resource (110 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 84-12, Section: B.
Contained By:	Dissertations Abstracts International84-12B.
Subject:	Programming languages. -
Online resource:	click for full text (PQDT)
ISBN:	9798379651602

Compiling Deep Learning Kernels to Locality-Aware Dataflow.
Zhao, Tian.

Compiling Deep Learning Kernels to Locality-Aware Dataflow. - 1 online resource (110 pages)

Source: Dissertations Abstracts International, Volume: 84-12, Section: B.

Thesis (Ph.D.)--Stanford University, 2023.

Includes bibliographical references

Emerging deep learning applications require unprecedented computation and memory capacity. To accelerate these applications, novel processing systems such as dataflow accelerators strive to exploit multiple dimensions of parallelism within deep learning models, e.g., tensor and pipeline parallelism. Although these systems provide ultrahigh performance when fully utilized, compiling deep learning applications to harness their computation capability remains a challenging problem. With recent advances in domain-specific programming language, accelerator design, and machine learning, we now have the potential to better serve the needs of training and evaluating large deep learning applications on dataflow accelerators through algorithm, software, and hardware co-design.In this dissertation, I present the design and development of efficient deep learning optimizations and programming frameworks. I present two frameworks: SpatialRNN for accelerating recurrent neural network language models on spatial accelerators and Sigma for expressing and accelerating high-data-reuse deep learning kernels using reconfigurable dataflow accelerators. Our end-to-end evaluation using Sigma demonstrates a 5.4x speedup on kernels encompassing financial applications, traditional machine learning, language modeling and computer vision tasks over an Nvidia V100 GPU accelerator.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798379651602Subjects--Topical Terms:

1468403
Programming languages.
Index Terms--Genre/Form:

554714
Electronic books.

Compiling Deep Learning Kernels to Locality-Aware Dataflow.
LDR:02682ntm a22003737 4500 001 1146384
005 20240812064602.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798379651602
035 $a (MiAaPQ)AAI30462707
035 $a (MiAaPQ)STANFORDhd752ps4385
035 $a AAI30462707
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Zhao, Tian. $e author. $3 1265120
245 1 0 $a Compiling Deep Learning Kernels to Locality-Aware Dataflow.
264 0 $c 2023
300 $a 1 online resource (110 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 84-12, Section: B.
500 $a Advisor: Raina, Priyanka;Re, Christopher;Olukotun, Oyekunle.
502 $a Thesis (Ph.D.)--Stanford University, 2023.
504 $a Includes bibliographical references
520 $a Emerging deep learning applications require unprecedented computation and memory capacity. To accelerate these applications, novel processing systems such as dataflow accelerators strive to exploit multiple dimensions of parallelism within deep learning models, e.g., tensor and pipeline parallelism. Although these systems provide ultrahigh performance when fully utilized, compiling deep learning applications to harness their computation capability remains a challenging problem. With recent advances in domain-specific programming language, accelerator design, and machine learning, we now have the potential to better serve the needs of training and evaluating large deep learning applications on dataflow accelerators through algorithm, software, and hardware co-design.In this dissertation, I present the design and development of efficient deep learning optimizations and programming frameworks. I present two frameworks: SpatialRNN for accelerating recurrent neural network language models on spatial accelerators and Sigma for expressing and accelerating high-data-reuse deep learning kernels using reconfigurable dataflow accelerators. Our end-to-end evaluation using Sigma demonstrates a 5.4x speedup on kernels encompassing financial applications, traditional machine learning, language modeling and computer vision tasks over an Nvidia V100 GPU accelerator.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Programming languages. $3 1468403
650 4 $a Deep learning. $3 1127428
650 4 $a Bandwidths. $3 1372735
650 4 $a Optimization techniques. $3 1413598
650 4 $a Neural networks. $3 1011215
650 4 $a Design. $3 595500
650 4 $a Keyboards. $3 1372556
650 4 $a Linear algebra. $3 1468811
650 4 $a Computer science. $3 573171
650 4 $a Mathematics. $3 527692
655 7 $a Electronic books. $2 local $3 554714
690 $a 0389
690 $a 0729
690 $a 0800
690 $a 0984
690 $a 0405
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Stanford University. $3 1184533
773 0 $t Dissertations Abstracts International $g 84-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30462707 $z click for full text (PQDT)