{"title":"The MPRACE framework: An open source stack for communication with custom FPGA-based accelerators","authors":"G. Marcus, Wenxue Gao, A. Kugel, R. Manner","doi":"10.1109/SPL.2011.5782641","DOIUrl":null,"url":null,"abstract":"We present an open source stack for the development of custom FPGA boards, primarily but not limited to PCI Express interconnects. Supporting current Linux distributions, the stack consists of a PCI driver, an IP core for a DMA engine, a hardware abstraction library for IO operations, and a buffer management library for efficient handling of data transfers between an application and a FPGA design. The stack has been validated in diverse hardware and software platforms and provides several building blocks that facilitate the use of accelerators in applications. The DMA Engine IP provides high performance data transfers in PCIe 4-lane boards with Xilinx PCIe cores, with 380 MB/s read and 700 MB/s write maximum measured performance. The buffer management library allows the utilization of 80–95% of this bandwidth with reduced resource consumption and minimal effort.","PeriodicalId":6329,"journal":{"name":"2011 VII Southern Conference on Programmable Logic (SPL)","volume":"99 1","pages":"155-160"},"PeriodicalIF":0.0000,"publicationDate":"2011-04-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"21","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2011 VII Southern Conference on Programmable Logic (SPL)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/SPL.2011.5782641","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 21
Abstract
We present an open source stack for the development of custom FPGA boards, primarily but not limited to PCI Express interconnects. Supporting current Linux distributions, the stack consists of a PCI driver, an IP core for a DMA engine, a hardware abstraction library for IO operations, and a buffer management library for efficient handling of data transfers between an application and a FPGA design. The stack has been validated in diverse hardware and software platforms and provides several building blocks that facilitate the use of accelerators in applications. The DMA Engine IP provides high performance data transfers in PCIe 4-lane boards with Xilinx PCIe cores, with 380 MB/s read and 700 MB/s write maximum measured performance. The buffer management library allows the utilization of 80–95% of this bandwidth with reduced resource consumption and minimal effort.