Rapid, High-Level Performance Estimation for DSE using Calibrated Weight Tables
ثبت نشده
چکیده
Automated Design Space Exploration (DSE) is a critical part of system-level design. It relies on performance estimation to make rapid decisions. However, since a plethora of design alternatives need to be compared, the run-time of performance estimation itself may pose a bottleneck. In DSE, fastest performance estimation is of essence while some accuracy may be sacrificed. Fast estimation can be realised through capturing application demand, as well as Processing Element (PE) characteristics (later on called weight table) in a matrix each. Then, performance estimation (retargeting) is reduced to a matrix multiplication. However, defining the weight table from a data sheet is in-feasible due to the multitude of (micro-) architecture aspects. This paper introduces a novel methodology for automatically generating Weight Tables in context of C source-level estimation using dynamic profiling and Linear Programming. (LP). LP solving is based on the measured performance of training benchmarks on an actual PE. We validated the proposed approach using a synthetic processor and benchmark model, and also analyse the impact of non-observable features on estimation accuracy. We evaluate the efficiency of the proposed methodology using 49 benchmarks on 2 different processors with varying configurations (multiple memory configurations and software optimizations). On a 3.1 GHz i5-3450 Intel host, 25 million estimations / second can be obtained regardless of the application size and PE complexity. The accuracy is sufficient for early DSE with a 24% average error.
منابع مشابه
System - Level Design Space Exploration of Reconfigurable Architectures
R econfigurable architectures are becoming increasingly popular as they bear a promise of combining the flexibility of software with the performance of hardware. Nevertheless, such architectures are subject to numerous constraints, such as performance, memory requirements, chip area, and power consumption. To create an efficient design, performing Design Space Exploration (DSE) at various stage...
متن کاملA Model-Based Methodology for Application Specific Energy Efficient Data Path Design Using FPGAs
We present a methodology to design energy-efficient data paths using FPGAs. Our methodology integrates domain specific modeling, coarse-grained performance evaluation, design space exploration, and low level simulation to understand the tradeoffs between energy, latency, and area. The domain specific modeling technique defines a high-level model by identifying various components and parameters ...
متن کاملA fast MPSoC virtual prototyping for intensive signal processing applications
Due to the growing computation rates of intensive signal processing applications, using Multiprocessor System on Chip (MPSoC) becomes an incontrovertible solution to meet the functional requirements. Today, Electronic System Level (ESL) design is considered a vital premise to overcome the design complexity intrinsic in the heterogeneity of these devices. However, the development of tools at the...
متن کاملRapid Estimation of Water Flooding Performance and Optimization in EOR by Using Capacitance Resistive Model
متن کامل
Generation of Abstract IP/XACT Platform Descriptions from UML/MARTE for System-Level Performance Estimation Towards a MARTE to IP/XACT Generation Framework of HW platform descriptions for a DSE Multi- level Performance Estimation Framework
UML/MARTE is enabling the development of methodologies for the specification of a whole real-time system, and of holistic MDA methodologies where the UML/MARTE description is taken as the source for different design activities, such as system-level performance estimation and implementation refinement. A crucial issue to make these methodologies working and efficient is the development of tools,...
متن کامل