{"title":"深度学习编译器的一次性调谐器","authors":"Jaehun Ryu, Eunhyeok Park, Hyojin Sung","doi":"10.1145/3497776.3517774","DOIUrl":null,"url":null,"abstract":"Auto-tuning DL compilers are gaining ground as an optimizing back-end for DL frameworks. While existing work can generate deep learning models that exceed the performance of hand-tuned libraries, they still suffer from prohibitively long auto-tuning time due to repeated hardware measurements in large search spaces. In this paper, we take a neural-predictor inspired approach to reduce the auto-tuning overhead and show that a performance predictor model trained prior to compilation can produce optimized tensor operation codes without repeated search and hardware measurements. To generate a sample-efficient training dataset, we extend input representation to include task-specific information and to guide data sampling methods to focus on learning high-performing codes. We evaluated the resulting predictor model, One-Shot Tuner, against AutoTVM and other prior work, and the results show that One-Shot Tuner speeds up compilation by 2.81x to 67.7x compared to prior work while providing comparable or improved inference time for CNN and Transformer models.","PeriodicalId":333281,"journal":{"name":"Proceedings of the 31st ACM SIGPLAN International Conference on Compiler Construction","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":"{\"title\":\"One-shot tuner for deep learning compilers\",\"authors\":\"Jaehun Ryu, Eunhyeok Park, Hyojin Sung\",\"doi\":\"10.1145/3497776.3517774\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Auto-tuning DL compilers are gaining ground as an optimizing back-end for DL frameworks. While existing work can generate deep learning models that exceed the performance of hand-tuned libraries, they still suffer from prohibitively long auto-tuning time due to repeated hardware measurements in large search spaces. In this paper, we take a neural-predictor inspired approach to reduce the auto-tuning overhead and show that a performance predictor model trained prior to compilation can produce optimized tensor operation codes without repeated search and hardware measurements. To generate a sample-efficient training dataset, we extend input representation to include task-specific information and to guide data sampling methods to focus on learning high-performing codes. We evaluated the resulting predictor model, One-Shot Tuner, against AutoTVM and other prior work, and the results show that One-Shot Tuner speeds up compilation by 2.81x to 67.7x compared to prior work while providing comparable or improved inference time for CNN and Transformer models.\",\"PeriodicalId\":333281,\"journal\":{\"name\":\"Proceedings of the 31st ACM SIGPLAN International Conference on Compiler Construction\",\"volume\":\"1 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-03-18\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"5\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Proceedings of the 31st ACM SIGPLAN International Conference on Compiler Construction\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3497776.3517774\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 31st ACM SIGPLAN International Conference on Compiler Construction","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3497776.3517774","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Auto-tuning DL compilers are gaining ground as an optimizing back-end for DL frameworks. While existing work can generate deep learning models that exceed the performance of hand-tuned libraries, they still suffer from prohibitively long auto-tuning time due to repeated hardware measurements in large search spaces. In this paper, we take a neural-predictor inspired approach to reduce the auto-tuning overhead and show that a performance predictor model trained prior to compilation can produce optimized tensor operation codes without repeated search and hardware measurements. To generate a sample-efficient training dataset, we extend input representation to include task-specific information and to guide data sampling methods to focus on learning high-performing codes. We evaluated the resulting predictor model, One-Shot Tuner, against AutoTVM and other prior work, and the results show that One-Shot Tuner speeds up compilation by 2.81x to 67.7x compared to prior work while providing comparable or improved inference time for CNN and Transformer models.