Commercially available behavioral synthesis tools do not adequately support FPGA vendor-specific external memory interfaces making it extremely difficult to exploit pipelined memory access modes as well as application specific memory operations scheduling
Synthesis of Pipelined Memory Access Controllers for Streamed Data Applications on FPGA-based
Computing Engines
Joonseok Park and Pedro C. Diniz
University of Southern California / Information Sciences Institute
4676 Admiralty Way, Suite 1001 Marina del Rey, California 90292
ABSTRACT
{joonseok, pedro}@isi.edu
Commercially available behavioral synthesis tools do not adequately support FPGA vendor-specific external memory interfaces making it extremely difficult to exploit pipelined memory access modes as well as application specific memory operations scheduling critical for high-performance solutions. This lack of support substantially increases the complexity and the burden on designers in the mapping of applications to FPGA-based computing engines. In this paper we address the problem of external memory interfacing and aggressive scheduling of memory operations by proposing a decoupled architecture with two components - one component captures the specific target architecture timing while the other component uses application specific memory access pattern information. Our results support the claim that it is possible to exploit application specific information and integrate that knowledge into custom schedulers that mix pipelined and non-pipelined access modes aimed at reducing the overhead associated with external memory accesses. The results also reveal that the additional design complexity of the scheduler, and its impact in the overall design is minimal.
Keywords: FPGA-based configurable computing; Scheduling
of Memory Accesses; Hardware Interfaces and Customizable Memory Controllers.
1. INTRODUCTION
Commercially available synthesis tools for FPGAs have mostly ignored system level issues when dealing with external memories. While some tools now incorporate internal RAM modules and the mapping of array variables to them, they have avoided all external memory interfacing issues. Even for internal memories, the current tools require specific syntax and synthesis expertise precluding the interfacing with third party vendors for
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
ISSS’01, October 1-3, 2001, Montréal, Québec, Canada, Copyright, 2001 ACM 1-58113-418-5/01/00010…$5.00
which timing information is either unavailable or cannot fit into the timing model of the tool. In addition most, if not all, behavioral synthesis tools do not support pipelined modes further restricting an important source of performance improvement. As a result designers must engage in laborious and error-prone matching and synthesis of interface signals.
Configurable architectures offer a unique opportunity to address the memory access and interfacing issues via customization. Improvements can be achieved by creating specialized hardware components for generating addresses and packing and unpacking data items. Compiler analysis can provide application knowledge as to the memory access patterns for pipelining of data references corresponding to array references across multiple iterations of loops. Furthermore the compiler can derive information about the relative rate among various array references and embed that knowledge into the scheduling of memory operations.
In this paper we focus on the design, implementation and validation of external memory interfacing modules that are generated by a compiler and behavioral synthesis tool that translate high-level computations directly to FPGA hardware. For maximum generality, we have separated the memory interface into two components. The first component is target-dependent and captures the specific target architecture timing requirements for accessing memory. The second component is architecture-independent and provides a set of channels and memory access modes abstractions for the application to store and retrieve data from memory. In particular the designs we have implemented allow compiler-generated designs to exploit pipelined access modes and mix both pipelined and non-pipelined memory accesses modes.
We used the interfaces described in this paper to develop application specific scheduling strategies for a set of four digital image-processing applications. The experimental results reveal that while the designs do not exploit all of the available external memory bandwidth, they achieve respectable performance at a reasonable implementation and generality cost.
热门文档