Hardware/software partitioning is a process that occurs frequently in embedded system design. It is the procedure of determining whether a part of a system should be implemented in software or hardware. This dissertation is a study of hardware/software partitioning and the use of scheduling algorithms to improve the performance of dynamically reconfigurable computing devices. Reconfigurable computing devices are devices that are adaptable at the logic level to solve specific problems [Tes05]. One example of a reconfigurable computing device is the field programmable gate array (FPGA). The emergence of dynamically reconfigurable FPGAs made it possible to configure FPGAs at runtime. Most current approaches use a simple on demand configuration scheduling algorithm for the FPGA configurations. The on demand configuration scheduling algorithm reconfigures the FPGA at runtime, whenever a configuration is needed and is found not to be configured. The problem with this approach of dynamic reconfiguration is the reconfiguration time overhead, which is the time it takes to reconfigure the FPGA with a new configuration at runtime. Configuration caches and partial configuration have been proposed as possible solutions to this problem, but these techniques suffer from various limitations. The emergence of dynamically reconfigurable FPGAs also made it possible to perform dynamic hardware/software partitioning (DHSP), which is the procedure of determining at runtime whether a computation should be performed using its software or hardware implementation. The drawback of performing DHSP using configurations that are generated at runtime is that the profiling and the dynamic generation of configurations require profiling tool and synthesis tool access at runtime. This study proposes that configuration scheduling algorithms, which perform DHSP using statically generated configurations, can be developed to combine the advantages and reduce the major disadvantages of current approaches. A case study is used to compare and evaluate the tradeoffs between the currently existing approach for dynamic reconfiguration and the DHSP configuration scheduling algorithm based approach proposed in the study. A simulation model is developed to examine the performance of the various configuration scheduling algorithms. First, the difference in the execution time between the different approaches is analyzed. Afterwards, other important design criteria such as power consumption, energy consumption, area requirements and unit cost are analyzed and estimated. Also, business and marketing considerations such as time to market and development cost are considered. The study illustrates how different types of DHSP configuration scheduling algorithms can be implemented and how their performance can be evaluated using a variety of software applications. It is also shown how to evaluate when which of the approaches would be more advantageous by determining the tradeoffs that exist between them. Also the underlying factors that affect when which design alternative is more advantageous are determined and analyzed. The study shows that configuration scheduling algorithms, which perform DHSP using statically generated configurations, can be developed to combine the advantages and reduce some major disadvantages of current approaches. It is shown that there are situations where DHSP configuration scheduling algorithms can be more advantageous than the other approaches.

Hardware/software partitioning is a process that occurs frequently in embedded system design. It is the procedure of determining whether a part of a system should be implemented in software or hardware. This dissertation is a study of hardware/software partitioning and the use of scheduling algorithms to improve the performance of dynamically reconfigurable computing devices. Reconfigurable computing devices are devices that are adaptable at the logic level to solve specific problems [Tes05]. One example of a reconfigurable computing device is the field programmable gate array (FPGA). The emergence of dynamically reconfigurable FPGAs made it possible to configure FPGAs at runtime. Most current approaches use a simple on demand configuration scheduling algorithm for the FPGA configurations. The on demand configuration scheduling algorithm reconfigures the FPGA at runtime, whenever a configuration is needed and is found not to be configured. The problem with this approach of dynamic reconfiguration is the reconfiguration time overhead, which is the time it takes to reconfigure the FPGA with a new configuration at runtime. Configuration caches and partial configuration have been proposed as possible solutions to this problem, but these techniques suffer from various limitations. The emergence of dynamically reconfigurable FPGAs also made it possible to perform dynamic hardware/software partitioning (DHSP), which is the procedure of determining at runtime whether a computation should be performed using its software or hardware implementation. The drawback of performing DHSP using configurations that are generated at runtime is that the profiling and the dynamic generation of configurations require profiling tool and synthesis tool access at runtime. This study proposes that configuration scheduling algorithms, which perform DHSP using statically generated configurations, can be developed to combine the advantages and reduce the major disadvantages of current approaches. A case study is used to compare and evaluate the tradeoffs between the currently existing approach for dynamic reconfiguration and the DHSP configuration scheduling algorithm based approach proposed in the study. A simulation model is developed to examine the performance of the various configuration scheduling algorithms. First, the difference in the execution time between the different approaches is analyzed. Afterwards, other important design criteria such as power consumption, energy consumption, area requirements and unit cost are analyzed and estimated. Also, business and marketing considerations such as time to market and development cost are considered. The study illustrates how different types of DHSP configuration scheduling algorithms can be implemented and how their performance can be evaluated using a variety of software applications. It is also shown how to evaluate when which of the approaches would be more advantageous by determining the tradeoffs that exist between them. Also the underlying factors that affect when which design alternative is more advantageous are determined and analyzed. The study shows that configuration scheduling algorithms, which perform DHSP using statically generated configurations, can be developed to combine the advantages and reduce some major disadvantages of current approaches. It is shown that there are situations where DHSP configuration scheduling algorithms can be more advantageous than the other approaches.

The primary design goal of many embedded systems for multimedia applications is usually meeting the performance requirement at a minimum cost. In this thesis, we proposed two different ILP based approaches for hardware/software (HW/SW) partitioning and pipelined scheduling of embedded systems for multimedia applications. One ILP approach solves the HW/SW partitioning and pipelined scheduling problem simultaneously. Another ILP approach separates the HW/SW partitioning and pipelined scheduling problem into two phases. The first phase is focusing on the HW/SW partitioning and mapping problem. Second phase is used to solve the pipelined scheduling problem. The two ILP approaches not only partition and map each computation task of a particular multimedia application onto a component of the heterogeneous multiprocessor architecture, but also schedules and pipelines the execution of these computation tasks while considering communication time. For the first ILP model, the objective is to minimize the total component cost and the number of pipeline stages subject to the throughput constraint. In the second ILP approach, the objective of the first phase and second phase is to minimize the total component cost and the number of pipeline stages subject to the throughput constraint, respectively. Finally, experiments on three real multimedia applications (JPEG Encoder, MP3 Decoder, Wavelet Video Encoder) are used to demonstrate the effectiveness of the proposed approaches.

<p>This thesis describes the design of fast JPEG processing accelerators for a DSP processor. </p><p>Certain computation tasks are moved from the DSP processor to hardware accelerators. The accelerators are slave co processing machines and are controlled via a new instruction set. The clock cycle and power consumption is reduced by utilizing the custom built hardware. The hardware can perform the tasks in fewer clock cycles and several tasks can run in parallel. This will reduce the total number of clock cycles needed. </p><p>First a decoder and an encoder were implemented in DSP assembler. The cycle consumption of the parts was measured and from this the hardware/software partitioning was done. Behavioral models of the accelerators were then written in C++ and the assembly code was modified to work with the new hardware. Finally, the accelerators were implemented using Verilog. </p><p>Extension of the accelerator instructions was given following a custom design flow.</p>

<p>H.264/AVC is a new video compression standard designed for future broadband network. Compared with former video coding standards such as MPEG-2 and MPEG-4 part 2, it saves up to 40% in bit rate and provides important characteristics such as error resilience, stream switching etc. However, the improvement in performance also introduces increase in computational complexity, which requires more powerful hardware. At the same time, there are several image and video coding standards currently used such as JPEG and MPEG-4. Although ASIC design meets the performance requirement, it lacks flexibility for heterogeneous standards. Hence reconfigurable DSP processor is more suitable for media processing since it provides both real-time performance and flexibility. </p><p>Currently there are several single scalar DSP processors in the market. Compare to media processor, which is generally SIMD or VLIW, single scalar DSP is cheaper and has smaller area while its performance for video processing is limited. In this paper, a method to promote the performance of single scalar DSP by attaching hardware accelerators is proposed. And the bottleneck for performance promotion is investigated and the upper limit of acceleration of a certain single scalar DSP for H.264/AVC decoding is presented. </p><p>Behavioral model of H.264/AVC decoder is realized in pure software during the first step. Although real-time performance cannot be achieved with pure software implementation, computational complexity of different parts is investigated and the critical path in decoding was exposed by analyzing the first design of this software solution. Then both functional acceleration and addressing acceleration were investigated and designed to achieve the performance for real-time decoding using available clock frequency within 200MHz.</p>

<p>For demanding applications, for example image or video processing, there may be computations that aren’t very suitable for digital signal processors. While a DSP processor is appropriate for some tasks, the instruction set could be extended in order to achieve higher performance for the tasks that such a processor normally isn’t actually design for. The platform used in this project is flexible in the sense that new hardware can be designed to speed up certain computations.</p><p>This thesis analyzes the computational complex parts of JPEG2000. In order to achieve sufficient performance for JPEG2000, there may be a need for hardware acceleration.</p><p>First, a JPEG2000 decoder was implemented for a DSP processor in assembler. When the firmware had been written, the cycle consumption of the parts was measured and estimated. From this analysis, the bottlenecks of the system were identified. Furthermore, new processor instructions are proposed that could be implemented for this system. Finally the performance improvements are estimated.</p>

Due to the rapid advancement of VLSI technology, functions of multimedia systems (e.g. MP3, image processing etc,) become more complex nowadays. Therefore, more complicated system architecture and powerful computing ability are required to attain real-time process. In order to fulfill the requirement of cost and efficiency, multimedia systems usually consist of processors, ASICs and other various components. Diverse real-time functions of multimedia system can be implemented via co-operation of these components. However, these components have the differences in area, efficiency, and cost. Accordingly, it is necessary to establish a useful method and tool to decide better system architecture. ¡@¡@The main objective of the thesis is to develop a decision tool of system architecture. Given the system specification and constraints of a multimedia system, it can quickly decide a good system architecture to satisfy the constraints in accordance with the system specification . Except for partitioning various functions and mapping functions to hardware components, the decision tool must implement scheduling and pipelining to fulfill resource constraints and reach the real-time requirement. A great deal of time and cost for implementing the multimedia system can be reduced by careful scheduling and pipelining. Besides, owing to the pressure of time to market in system development, we propose a series of design flow to speed up the design process. All decision tasks in the flow are finished by automatic or semi-automatic method to reduce the exhaustion of manpower and time. Finally, we make experiments with the proposed tool on several multimedia systems. The results show that the automatic process can conform to constraints which set up by system specifications and ensure the accuracy of the process.

In the last years, with more and more transistors can fit into a chip, the growth the IC design complexity is fast and original design flow can¡¦t cater for designers. Therefore, so many people promote to integrate the system into a single chip gradually with the last technology using the concept of hardware/software co-design. In this thesis, we use the hardware/software co-design concept to build a simple video codec from system level and implement it on the ARM¡¦s SOC platform. We focus on the hardware/software co-ordination. Because we use the platform-based design method, the build hardware/software modules can be used in the similar architecture on the ARM platform In our Video codec system, discrete wavelet transform(DWT) and RGBtoYCbCr are the most timing-consuming parts. Since DWT has inherent scalability and excellent features of energy compaction, it has been applied widely in the various image compression systems. We adopt the 5-3 filter lifting-based DWT in the hardware part of our system and design three different lifting-based DWT architectures by using the high level synthesis concept to optimize the hardware utilization and speed. In the premise of not increasing memory access times and additional processes of software, we overcome the boundary extension of DWT and verify it by means of FPGA after combining it with the RGBtoYCbCr hardware architecture. Finally, the hardware part is integrated with the other part implemented by software, we build a completely video encode system on the ARM SOC platform using the hardware/software co-design.

MP3 compression format is essential categorized one of the MPEG (Moving Picture Experts Group) standards for digital audio compression nowadays. For its superiority and convenient,MP3 has been widely used in multimedia player and storage application. In this thesis, we use software/hardware co-design methodology to design the MP3 player system. In addition, system level scheduling is adopted to arrange the execute time of SW and HW and significantly reduce the hardware cost under the construct of real-time processing. We can obtain fewer extra hardware cost while attaining the goal of real- time playing system. In order to perform software/hardware partitioning, simulate and analyze the MP3 application program to find out the critical parts with high time complexity and regular computation. These parts with high time complexity, e.g. IMDCT and Poly Phase synthesis filter bank, then are implemented by hardware to achieve better system performance. We use high level synthesis concept to optimize the hardware part and integrate software and hardware¡Asuch that communication between software and hardware can be performed smoothly. Finally, MP3 player system is using verified by hardware¡Bsoftware co- verified methodology on an SoC development platform. In order to build a complete verification environment, we attach extra input and output interfaces to the SoC development platform, e.g. the network card and sound card. Write some driver to drive related peripheral device. Since OS is conducive to the operations between software and hardware, Linux OS is ported to the SoC platform to manage software and hardware resources and drive the peripheral devices.

For demanding applications, for example image or video processing, there may be computations that aren’t very suitable for digital signal processors. While a DSP processor is appropriate for some tasks, the instruction set could be extended in order to achieve higher performance for the tasks that such a processor normally isn’t actually design for. The platform used in this project is flexible in the sense that new hardware can be designed to speed up certain computations. This thesis analyzes the computational complex parts of JPEG2000. In order to achieve sufficient performance for JPEG2000, there may be a need for hardware acceleration. First, a JPEG2000 decoder was implemented for a DSP processor in assembler. When the firmware had been written, the cycle consumption of the parts was measured and estimated. From this analysis, the bottlenecks of the system were identified. Furthermore, new processor instructions are proposed that could be implemented for this system. Finally the performance improvements are estimated.