A Methodology for Architecture Exploration of heterogeneous Signal Processing Systems Paul Lieverse Pieter van der Wolf Ed Deprettere Kees Vissers Outline Problem definition Related work Basic principles Methodology SPADE Case study Conclusion Problem Definition Modern signal processing systems have a heterogeneous architecture programmable components to offer various functions and support different standards for transmission dedicated hardware blocks for cost and power consideration New design Methodology SPADE for architecture exploration of heterogeneous signal processing systems Starts from a set of target applications Results in the definition of an architecture capable of executing the applications within predefined constraints Problem Definition cont d The design has to start with abstract yet executable models Cost of model construction and model evaluation Flexibility to explore alternative architectures Related Work Application modeling Models of computation Synchronous Dataflow SDF Dataflow Process Networks Kahn Process Networks Architecture modeling and performance analysis at system level Polis reactive systems Chinook embedded systems RASSP DSP systems Quantitative analysis of architectures by A C J Kienhuis limited to a specific class of dataflow architectures In contrast SPADE distinguishes between application models and architecture models and supports explicit mapping reusability Basic Principles The Y Chart a general scheme for the design of programmable architectures Basic Principles cont d Workload and Resources Computation communication workload Processing communication memory resources etc Applications and architectures are modeled separately Trace Driven Simulation performance analysis Applications network of concurrent communicating processes Trace workload SPADE Application Modeling Objective expose parallelism make communication explicit Kahn Process Networks The execution is deterministic It fits nicely with signal processing applications It allows programmers to easily combine communication primitives with control constructs Application Programmers Interface API of SPADE Read function Write function Execution function to handle symbolic instruction Trace entries are generated by them SPADE cont d Architecture Modeling Easy to construct No need to model the functional behavior Library of Generic building blocks which are parameterized Trace driven execution unit TDEU which interprets trace entries Interfaces which connects the I O ports of TDEU to communication resource Generic bus block SPADE cont d Mapping Each process is mapped onto a TDEU Can be many to one but not one to many Each process port is mapped one toone onto an I O port SPADE cont d Simulation Concurrently simulate the application model and the architecture model in a single memory space Performance Metrics The building blocks contain collectors for performance metrics Data is collected during simulation Case Study An MPEG 2 Decoder Starts with the C code of the MPEG 2 video decoder Step 1 partition the sequential program into a parallel Kahn Process Network using API functions Step 2 Collect statistics of the workload for different MPEG sequences by running the application in standalone mode Step 3 construct model for a realistic architecture TM2700 MPEG architecture using the blocks from the library Step 4 Define a mapping Step 5 Perform simulation Identify the bottlenecks Conclusion SPADE supports efficient exploration of heterogeneous signal processing architectures that must satisfy the workload demands of multiple target applications Kahn API functions can be used to structure applications A broad class of architectures can be modeled with the generic architecture blocks from the library Trace driven simulation is used for co simulation Simulation speed about 20 000 cycles per second for a relatively complex design A number of architectures and mappings can be explored
View Full Document
Unlocking...