University of Patras
Βελτιστοποίηση και αυτοματοποίηση τεχνικών μεταγλώττισης μέσω μοντελοποίησης σε επαναπροσδιοριζόμενα συστήματα
Abstract
dc:descriptionThe research material that is presented in this PhD Thesis is related with the development ofcompiling optimization techniques for reconfigurable systems and application specific integrated circuits. The objective is the optimisation of the execution of applications in terms of speed, area and power consumption in these architectures. This is achieved by developing original compiling techniques and efficient architecture instances. Moreover, one of the fundamental objectives of this thesis is the automation of these techniques for the fast determination of most optimal solutions. The automation is achieved with a set of prototype tools for fast compiling and evaluation of the reconfigurable architectures. The largest part of the Thesis is devoted with development of compilation and evaluation techniques for coarse grain reconfigurable array architectures. These architectures consist of a big number of processing elements which work in parallel for exploiting the inherent parallelism existing in the algorithms making their execution faster and more efficient. These systems are used as coprocessors (accelerators) in embedded systems aiming at the acceleration of applications that execute on them. The higher level of control of these systems is implement most commonly by Instruction Set Processors. The mapping of applications on reconfigurable array architectures is characterized by high complexity. As result, the realisation without suitable tools makes it very time-consuming. Moreover, specialized compiling techniques are required in order to overcome the factors that constitute a bottleneck in the exploitation of the computational power of these architectures. For the optimised compilation in these architectures an original-automated-reconfigurable and formalized methodology was proposed. The automation was realized with a set of prototype compiling tools which target on a parametric template of coarse grain reconfigurable array architecture. This allowed fast exploration of the design space aiming at the determination of the most optimal architectures in terms of speed area for different categories of algorithms. Moreover, the incorporation of the reconfigurable array architecture in embedded systems on chip with Instruction Set Processor and Fine Grain reconfigurable hardware was also studied. For compiling applications in these systems a partitioning methodology was proposed for identifying the critical parts of applications that will be executed in the reconfigurable array architecture and ultimately will accelerate the execution of the overall application. The experiments showed important speedups in the execution of applications by the use of the reconfigurable array architecture as an accelerator. Also, an application specific processor was developed for the realization of the forward wavelet transform according to the JPEG2000 standard. The realization of the system includes the 1-dimensional convolutional digital filters that implement the forward wavelet transform and the architectural interconnection of these with memories. For the scheduling of calculations in the architecture a scheduling algorithm is proposed that aims the minimisation of memory requirements and the maximisation of speed. The description of architecture and the scheduling algorithm is parametric so that it is applied on pictures of different dimensions and for any of the convolutional filters that have been proposed in the JPEG 2000 for implementing the wavelet transform. Also, the optimization methodology that was introduced for the optimisation of the HIPERLAN/2 network standard and the determination of its memory architecture is also described. Also, this methodology can be applied to the optimization of other network protocols with characteristics similar to the ones of HIPERLAN/2. The methodology consists of three stages. In first, analysis partitions the application in layers according to their criticality in the optimization stage. The objective is to reduce the optimization time since the actual optimizable code is much less than the original application code. Hence, the designer’s scope reduces significantly. For the analysis a prototype compiler tool was realized. The second stage refers to the optimization of the protocol for which existing tools where used. The optimization targets the refinement of the application data types and the way their accessed for achieving more efficient memory operation. Finally, based on the new implementation of the application’s data types a power aware memory architecture is derived in the third stage that meets the timing constraints. The application of methodology resulted in improvements up to 37% in terms of the power consumption.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Dimitroulakos, Grigorios
- Δημητρουλάκος, Γρηγόριος
Subjects
dc:subject × 17- Μεταγλωττιστές
- Επαναπροσδιοριζόμενα συστήματα
- Αρχιτεκτονικές πίνακα
- Βελτιστοποίηση
- Εκμετάλευση παραλληλίας
- Παράλληλες αρχιτεκτονικές
- Αυτοματοποίηση τεχνικών μεταγλώττισης
- Compiles
- Reconfigurable systems
- Reconfigurable array architectures
- Optimization
- Parallel architectures
- Compilation techniques
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18981