{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,8]],"date-time":"2026-09-08T15:03:50Z","timestamp":1788879830468,"version":"build-2803163510"},"reference-count":57,"publisher":"Association for Computing Machinery (ACM)","issue":"1","license":[{"start":{"date-parts":[[2021,12,6]],"date-time":"2021-12-06T00:00:00Z","timestamp":1638748800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/www.acm.org\/publications\/policies\/copyright_policy#Background"}],"funder":[{"name":"NSF","award":["1755876, 1909900, and 1822939"],"award-info":[{"award-number":["1755876, 1909900, and 1822939"]}]},{"name":"U.S. Department of Energy\u2019s National Nuclear Security Administration","award":["DE-NA-0003525"],"award-info":[{"award-number":["DE-NA-0003525"]}]}],"content-domain":{"domain":["dl.acm.org"],"crossmark-restriction":true},"short-container-title":["ACM Trans. Archit. Code Optim."],"published-print":{"date-parts":[[2022,3,31]]},"abstract":"<jats:p>A spatial accelerator\u2019s efficiency depends heavily on both its mapper and cost models to generate optimized mappings for various operators of DNN models. However, existing cost models lack a formal boundary over their input programs (operators) for accurate and tractable cost analysis of the mappings, and this results in adaptability challenges to the cost models for new operators. We consider the recently introduced Maestro Data-Centric (MDC) notation and its analytical cost model to address this challenge because any mapping expressed in the notation is precisely analyzable using the MDC\u2019s cost model.<\/jats:p>\n          <jats:p>\n            In this article, we characterize the set of input operators and their mappings expressed in the MDC notation by introducing a\n            <jats:italic>set of conformability rules<\/jats:italic>\n            . The outcome of these rules is that any loop nest that is perfectly nested with affine tensor subscripts and without conditionals is conformable to the MDC notation. A majority of the primitive operators in deep learning are such loop nests. In addition, our rules enable us to automatically translate a mapping expressed in the loop nest form to MDC notation and use the MDC\u2019s cost model to guide upstream mappers. Our conformability rules over the input operators result in a\n            <jats:italic>structured mapping space<\/jats:italic>\n            of the operators, which enables us to introduce a mapper based on our\n            <jats:italic>decoupled off-chip\/on-chip<\/jats:italic>\n            approach to accelerate mapping space exploration. Our mapper decomposes the original higher-dimensional mapping space of operators into two lower-dimensional off-chip and on-chip subspaces and then optimizes the off-chip subspace followed by the on-chip subspace. We implemented our overall approach in a tool called\n            <jats:italic>Marvel<\/jats:italic>\n            , and a benefit of our approach is that it applies to any operator conformable with the MDC notation. We evaluated Marvel over major DNN operators and compared it with past optimizers.\n          <\/jats:p>","DOI":"10.1145\/3485137","type":"journal-article","created":{"date-parts":[[2021,12,6]],"date-time":"2021-12-06T21:29:34Z","timestamp":1638826174000},"page":"1-26","update-policy":"https:\/\/doi.org\/10.1145\/crossmark-policy","source":"Crossref","is-referenced-by-count":29,"title":["Marvel: A Data-Centric Approach for Mapping Deep Learning Operators on Spatial Accelerators"],"prefix":"10.1145","volume":"19","author":[{"given":"Prasanth","family":"Chatarasi","sequence":"first","affiliation":[{"name":"Georgia Institute of Technology, Atlanta, GA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hyoukjun","family":"Kwon","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology, Atlanta, GA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Angshuman","family":"Parashar","sequence":"additional","affiliation":[{"name":"NVIDIA, Westford, MA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Michael","family":"Pellauer","sequence":"additional","affiliation":[{"name":"NVIDIA, Westford, MA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tushar","family":"Krishna","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology, Atlanta, GA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Vivek","family":"Sarkar","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology, Atlanta, GA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"320","published-online":{"date-parts":[[2021,12,6]]},"reference":[{"key":"e_1_3_2_2_2","article-title":"Edge TPU: Google\u2019s Purpose-Built ASIC Designed to Run Inference at the Edge. Retrieved October 13, 2021 from","author":"Google Cloud.","unstructured":"Google Cloud. 2019. Edge TPU: Google\u2019s Purpose-Built ASIC Designed to Run Inference at the Edge. Retrieved October 13, 2021 fromhttps:\/\/cloud.google.com\/edge-tpu\/.","journal-title":"https:\/\/cloud.google.com\/edge-tpu\/"},{"key":"e_1_3_2_3_2","volume-title":"Optimizing Compilers for Modern Architectures: A Dependence-Based Approach","author":"Allen Randy","year":"2001","unstructured":"Randy Allen and Ken Kennedy. 2001. Optimizing Compilers for Modern Architectures: A Dependence-Based Approach. Morgan Kaufmann."},{"key":"e_1_3_2_4_2","doi-asserted-by":"publisher","DOI":"10.5555\/3314872.3314896"},{"key":"e_1_3_2_5_2","article-title":"End to end learning for self-driving cars","author":"Bojarski Mariusz","year":"2016","unstructured":"Mariusz Bojarski, Davide Del Testa, Daniel Dworakowski, Bernhard Firner, Beat Flepp, Prasoon Goyal, Lawrence D. Jackel, et al. 2016. End to end learning for self-driving cars. arXiv preprint arXiv:1604.07316 (2016)","journal-title":"arXiv preprint arXiv:1604.07316"},{"key":"e_1_3_2_6_2","doi-asserted-by":"publisher","DOI":"10.1145\/2654822.2541967"},{"key":"e_1_3_2_7_2","doi-asserted-by":"publisher","DOI":"10.5555\/3291168.3291211"},{"key":"e_1_3_2_8_2","doi-asserted-by":"publisher","DOI":"10.1145\/3007787.3001177"},{"key":"e_1_3_2_9_2","doi-asserted-by":"publisher","DOI":"10.1109\/JSSC.2016.2616357"},{"key":"e_1_3_2_10_2","doi-asserted-by":"publisher","DOI":"10.1109\/JSSC.2016.2616357"},{"key":"e_1_3_2_11_2","doi-asserted-by":"publisher","DOI":"10.1109\/JETCAS.2019.2910232"},{"key":"e_1_3_2_12_2","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2018.022071131"},{"key":"e_1_3_2_13_2","doi-asserted-by":"publisher","DOI":"10.1145\/3358198"},{"key":"e_1_3_2_14_2","doi-asserted-by":"publisher","DOI":"10.1145\/2872887.2750389"},{"key":"e_1_3_2_15_2","doi-asserted-by":"crossref","unstructured":"Bruce M. Fleischer et al. bibinfoyear2018. A scalable multi-TeraOPS deep learning processor core for AI trainina and inference. In Proceedings of the 2018 IEEE Symposium on VLSI Circuits . IEEE Los Alamitos CA 35\u201336. https:\/\/doi.org\/10.1109\/VLSIC.2018.8502276","DOI":"10.1109\/VLSIC.2018.8502276"},{"key":"e_1_3_2_16_2","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2012.231"},{"key":"e_1_3_2_17_2","doi-asserted-by":"publisher","DOI":"10.5555\/645669.665222"},{"key":"e_1_3_2_18_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"e_1_3_2_19_2","unstructured":"Xu Shell Hu Sergey Zagoruyko and Nikos Komodakis. 2018. Exploring weight symmetry in deep neural networks. arxiv:1812.11027 [cs.LG]"},{"key":"e_1_3_2_20_2","doi-asserted-by":"publisher","DOI":"10.1109\/PACT52795.2021.00010"},{"key":"e_1_3_2_21_2","doi-asserted-by":"publisher","DOI":"10.1145\/3140659.3080246"},{"key":"e_1_3_2_22_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2015.7298932"},{"key":"e_1_3_2_23_2","doi-asserted-by":"publisher","DOI":"10.5555\/2999134.2999257"},{"key":"e_1_3_2_24_2","article-title":"MAESTRO: An Open-Source Infrastructure for Modeling Dataflows Within Deep Learning Accelerators","author":"Kwon Hyoukjun","year":"2020","unstructured":"Hyoukjun Kwon. 2020. MAESTRO: An Open-Source Infrastructure for Modeling Dataflows Within Deep Learning Accelerators. Retrieved October 13, 2021 from https:\/\/github.com\/maestro-project\/maestro\/tree\/master\/data\/mapping.","journal-title":"h"},{"key":"e_1_3_2_25_2","doi-asserted-by":"publisher","DOI":"10.1145\/3352460.3358252"},{"key":"e_1_3_2_26_2","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2020.2985963"},{"key":"e_1_3_2_27_2","doi-asserted-by":"publisher","DOI":"10.1145\/3296957.3173176"},{"key":"e_1_3_2_28_2","unstructured":"Chris Lattner Mehdi Amini Uday Bondhugula Albert Cohen Andy Davis Jacques Pienaar River Riddle Tatiana Shpeisman Nicolas Vasilache and Oleksandr Zinenko. 2020. MLIR: A compiler infrastructure for the end of Moore\u2019s law. arxiv:2002.11054 [cs.PL]"},{"key":"e_1_3_2_29_2","doi-asserted-by":"publisher","DOI":"10.5555\/3014904.3014977"},{"key":"e_1_3_2_30_2","doi-asserted-by":"publisher","DOI":"10.1145\/3020078.3021736"},{"key":"e_1_3_2_31_2","doi-asserted-by":"publisher","DOI":"10.1145\/3229762.3229766"},{"key":"e_1_3_2_32_2","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2019.2928962"},{"key":"e_1_3_2_33_2","doi-asserted-by":"publisher","DOI":"10.1109\/ASPDAC.2016.7428073"},{"key":"e_1_3_2_34_2","unstructured":"NVIDIA. 2018. NVIDIA Deep Learning Accelerator (NVDLA). Retrieved October 13 2021 from https:\/\/nvidia.org."},{"key":"e_1_3_2_35_2","doi-asserted-by":"crossref","unstructured":"Angshuman Parashar Priyanka Raina Yakun Sophia Shao Yu-Hsin Chen Victor A. Ying Anurag Mukkara Rangarajan Venkatesan Brucek Khailany Stephen W. Keckler and Joel Emer. 2019. Timeloop: A systematic approach to DNN accelerator evaluation. In Proceedings of the 2019 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS\u201919) .","DOI":"10.1109\/ISPASS.2019.00042"},{"key":"e_1_3_2_36_2","doi-asserted-by":"publisher","DOI":"10.1145\/3140659.3080254"},{"key":"e_1_3_2_37_2","unstructured":"https:\/\/github.com\/plaidml\/plaidml 2020 PlaidML"},{"key":"e_1_3_2_38_2","doi-asserted-by":"crossref","unstructured":"Eric Qin Ananda Samajdar Hyoukjun Kwon Vineet Nadella Sudarshan Srinivasan Dipankar Das Bharat Kaul and Tushar Krishna. 2020. SIGMA: A sparse and irregular GEMM accelerator with flexible interconnects for DNN training. In Proceedings of the 2020 IEEE International Symposium on High Performance Computer Architecture (HPCA\u201920) .","DOI":"10.1109\/HPCA47549.2020.00015"},{"key":"e_1_3_2_39_2","unstructured":"Vijay Janapa Reddi Christine Cheng David Kanter Peter Mattson Guenther Schmuelling Carole-Jean Wu Brian Anderson et al. 2019. MLPerf inference benchmark. arxiv:1911.02549 [cs.LG]"},{"key":"e_1_3_2_40_2","doi-asserted-by":"publisher","DOI":"10.1109\/SC.2004.3"},{"key":"e_1_3_2_41_2","doi-asserted-by":"publisher","DOI":"10.5555\/1413370.1413426"},{"key":"e_1_3_2_42_2","doi-asserted-by":"crossref","unstructured":"O. Ronneberger P. Fischer and T. Brox. 2015. U-Net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention . Lecture Notes in Computer Science Vol. 9351. Springer 234\u2013241. http:\/\/lmb.informatik.uni-freiburg.de\/Publications\/2015\/RFB15a.","DOI":"10.1007\/978-3-319-24574-4_28"},{"key":"e_1_3_2_43_2","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-015-0816-y"},{"key":"e_1_3_2_44_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00474"},{"key":"e_1_3_2_45_2","doi-asserted-by":"publisher","DOI":"10.1147\/rd.413.0233"},{"key":"e_1_3_2_46_2","doi-asserted-by":"publisher","DOI":"10.5555\/1153923.1154542"},{"key":"e_1_3_2_47_2","doi-asserted-by":"publisher","DOI":"10.5555\/3195638.3195659"},{"key":"e_1_3_2_48_2","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-642-28652-0_6"},{"key":"e_1_3_2_49_2","volume-title":"Proceedings of the International Conference on Learning Representations (ICLR\u201915)","author":"Simonyan Karen","year":"2015","unstructured":"Karen Simonyan and Andrew Zisserman. 2015. Very deep convolutional networks for large-scale image recognition. In Proceedings of the International Conference on Learning Representations (ICLR\u201915)."},{"key":"e_1_3_2_50_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2014.214"},{"key":"e_1_3_2_51_2","article-title":"Tensor comprehensions: Framework-agnostic high-performance machine learning abstractions","author":"Vasilache Nicolas","year":"2018","unstructured":"Nicolas Vasilache, Oleksandr Zinenko, Theodoros Theodoridis, Priya Goyal, Zachary DeVito, William S. Moses, Sven Verdoolaege, Andrew Adams, and Albert Cohen. 2018. Tensor comprehensions: Framework-agnostic high-performance machine learning abstractions. arXiv preprint arXiv:1802.04730 (2018)","journal-title":"arXiv preprint arXiv:1802.04730"},{"key":"e_1_3_2_52_2","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2019.2931584"},{"key":"e_1_3_2_53_2","article-title":"Google\u2019s neural machine translation system: Bridging the gap between human and machine translation","author":"Wu Yonghui","year":"2016","unstructured":"Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, et al. 2016. Google\u2019s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 (2016)","journal-title":"arXiv preprint arXiv:1609.08144"},{"key":"e_1_3_2_54_2","unstructured":"https:\/\/www.xilinx.com\/support\/documentation\/white_papers\/wp504-accel-dnns.pdf Accelerating DNNs with Xilinx Alveo Accelerator Cards"},{"key":"e_1_3_2_55_2","doi-asserted-by":"publisher","DOI":"10.1145\/3373376.3378514"},{"key":"e_1_3_2_56_2","unstructured":"Tim Zerrell and Jeremy Bruestle. 2019. Stripe: Tensor compilation via the nested polyhedral model. arxiv:1903.06498 [cs.DC]"},{"key":"e_1_3_2_57_2","doi-asserted-by":"publisher","DOI":"10.1145\/2684746.2689060"},{"key":"e_1_3_2_58_2","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS.2019.00040"}],"container-title":["ACM Transactions on Architecture and Code Optimization"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3485137","content-type":"unspecified","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3485137","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3485137","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,6,17]],"date-time":"2025-06-17T20:18:35Z","timestamp":1750191515000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3485137"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021,12,6]]},"references-count":57,"journal-issue":{"issue":"1","published-print":{"date-parts":[[2022,3,31]]}},"alternative-id":["10.1145\/3485137"],"URL":"https:\/\/doi.org\/10.1145\/3485137","relation":{},"ISSN":["1544-3566","1544-3973"],"issn-type":[{"value":"1544-3566","type":"print"},{"value":"1544-3973","type":"electronic"}],"subject":[],"published":{"date-parts":[[2021,12,6]]},"assertion":[{"value":"2021-04-01","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2021-09-01","order":1,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2021-12-06","order":2,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}