A New Compilation Flow for Software-Defined Radio Applications on Heterogeneous MPSoCs

Abstract

International audienceThe advent of portable software-defined radio (SDR) technology is tightly linked to the resolution of a difficult problem: efficient compilation of signal processing applications on embedded computing devices. Modern wireless communication protocols use packet processing rather than infinite stream processing and also introduce dependencies between data value and computation behavior leading to dynamic dataflow behavior. Recently, parametric dataflow has been proposed to support dynamicity while maintaining the high level of analyzability needed for efficient real-life implementations of signal processing computations. This article presents a new compilation flow that is able to compile parametric dataflow graphs. Built on the LLVM compiler infrastructure, the compiler offers an actor-based C++ programming model to describe parametric graphs, a compilation front end for graph analysis, and a back end that currently matches the Magali platform: a prototype heterogeneous MPSoC dedicated to LTE-Advanced. We also introduce an innovative scheduling technique, called microscheduling, allowing one to adapt the mapping of parametric dataflow programs to the specificities of the different possible MPSoCs targeted. A specific focus on FIFO sizing on the target architecture is presented. The experimental results show compilation of 3GPP LTE-Advanced demodulation on Magali with tight memory size constraints. The compiled programs achieve performance similar to handwritten code. This work was sponsored by Région Rhône Alpes ADR 11 01302401. Part of this work was published as M. Dardaillon, K. Marquet, T. Risset, J. Martin, and H.-P. Charles. 2014. A compilation flow for parametric dataflow: Programming model, scheduling, and application to heterogeneous MPSoC. " In Proceedings of the International Conference on Compilers, Architecture, and Synthesis for Embedded Systems (CASES'14). This work extends the previous publication by providing a thorough presentation of a new front end technique, introducing a new Promela model for buffer size verification, and updated results reflecting the subtantial improvements provided by the new Promela model

Dataflow Modeling for Reconfigurable Signal Processing Systems

“…Examples of SPDF and BPDF graphs of multimedia, signal processing, and software defined radio applications can be found in [6,16,22]. Compilation techniques for deploying SPDF graphs onto multi and many-core architecture will be presented in Section 3.…”

“…The mode selected for the next firing of an actor a depends on the parameter values, called a configuration, of a set of parameters P aram(a) specified at graph-level. The set of all valid configurations for an actor a is noted DOMAIN (a) Figure 7 presents an example of PSM-CFDF graph modeling part of the Orthogonal Frequency-Division Multiplexing (OFDM) demodulation of an LTE receiver, inspired by [16,50]. Parameter M takes value in {1, 2} and is used to switch the application behavior between a low-power mode M = 1, where a 5MHz bandwidth is received with QPSK modulation, and a highthroughput mode M = 2, where a 10MHz bandwidth is received with 16QAM modulation.…”

“…A compilation framework for deploying reconfigurable applications specified with the SPDF MoC onto heterogeneous MPSoCs is presented in [16]. This development flow for SPDF graphs differs from the flow based on the runtime manager presented in the previous section in that it shifts most of the deployment decisions to the compilation framework.…”

“…13 Overview of the compilation flow for SPDF graphs [16] In the SPDF compilation flow, the specifications of both the SPDF graph and the internal behavior of actors are based on a hierarchy of specialized C++ classes called Parametric Dataflow Format (padaf). Figure 14 illustrates the syntax used to specify part of an LTE application with padaf [16]. The SPDF graph presented in Figure 14a contains a parameterizable number N b of FFT actors, each processing 1024 samples received from an antenna.…”

“…-Analysis: In the third step of the front-end, executed in parallel with the second, the LLVM IR corresponding to the internal behavior of actors is analyzed in order to detect and annotate the instructions responsible for pushing and popping data into the Fifos connected to each actor. -Link : In the last step of the front-end, a full padaf IR of the application is obtained by linking the complete SPDF graph with annotated actor code, as detailed in [16].…”