Computer graphics are traditionally computation intensive, it stared at high-end system such as work stations for scientific purpose and then it showed up at the desktop computer. For now, portable devices are going to have ability to perform 3D applications. As we can see, the hardware resources for the computer graphics are getting less and less. If we can use lesser hardware to accelerated larger portions of application, it will be beneficial for the small system. In this thesis, we propose a method to analyze the application and propose a run-time scheduling for that reconfigurable system.
1.1 Computer Graphics
Since the mid-1970s the developmental motivation of computer graphics from the viewpoint of its practitioners has been photorealism or the pursuit of techniques that make a graphics image of an object or scene in distinguishable from a TV image or photograph. A more recent strand of the application of these techniques is to display information in, for example, medicine, science and engineering.
The calculation of light-object interaction is the foundation of photo-realism and this split neatly into two fields – the development of local reflection models and the development of global models. Local or direct reflection models only consider the interaction of an object with a light source as if the object and light were floating in dark space. That is, only the first reflection of light from the object is considered. Global reflection models consider how light reflects from one object and travels onto another.
In other words the light impinging on a point on the surface can come either from a light source (direct light) or indirect light that has first hit another object. Although two partial solutions for global interaction, ray tracing and radiosity, are implementer, global interaction is still for the most part an unsolved problem.
Much modern scientific research comes from computer graphics research and early
major advances are created and consolidated into a practical technology. Later significant advances seem to be more difficult to achieve. We can say that most images are produced using the Phong local reflection model (first reported in 1975), fewer using ray tracing (first popularized in 1980) and fewer still using radiosity (first reported in 1984). Although there is still much research being carried out in light-scene interaction methodologies much of the current research in computer graphics is concerned more with applications, for example, with such general applications as animation, visualization and virtual reality. In the most important computer graphics publication (the annual SIGGRAPH conference proceedings) there was in 1985 a total of 22 papers concerned with the production techniques of images (rendering, modeling and hardware) compared with 13 on what could loosely be called applications. A decade later in 1995 there were 37 papers on applications and 19 on image production techniques. [1]
1.2 Run-time Reconfiguration
Frequently, the areas of a program that can be accelerated by using the reconfigurable hardware are too numerous or complex to be loaded simultaneously onto the available hardware. For these cases, if we can swap different configurations in and out of the reconfigurable hardware as they are needed during program execution, it will be beneficial (Figure 1.1). This concept is called run-time reconfiguration (RTR).
Figure 1-1 The concept of Run-time Reconfiguration [2]
Run-time reconfiguration is based upon the concept of virtual hardware, which is similar to virtual memory. Here, the physical hardware is much smaller than the sum of the resources required by each of the configurations. Therefore, instead of reducing the
number of configurations that are mapped, we instead swap them in and out of the actual hardware as they are needed. Because run-time reconfiguration allows more sections of an application to be mapped into hardware than can be fit in a non-run-time reconfigurable system, a greater portion of the program can be accelerated. This provides potential for an overall improvement in performance.
Configurations are swapped in and out of the reconfigurable hardware during a single program’s execution. Some of these configurations will likely require access to the results of other configurations. Configurations that are active at different periods in time therefore must be provided with a method to communicate with one another. Primarily, this can be done through the use of registers [7] [8] [9] [10], the contents of which can remain intact between reconfigurations. This allows one configuration to store a value, and a later configuration to read back that value for use in further computations. An alternative for reconfigurable systems that do not include state-holding devices is to write the result back to registers or memory external to the reconfigurable array, which is then read back by successive configurations [11].
There are a few different configuration memory styles that can be used with reconfigurable systems. A single context device is a serially programmed chip that requires a complete reconfiguration in order to change any of the programming bits. A multicontext device has multiple layers of programming bits, each of which can be active at a different point in time. Devices that can be selectively programmed without a complete reconfiguration are called partially reconfigurable. These different types of configuration memory are described in more detail later. An advantage of the multicontext FPGA over a single context architecture is that it allows for an extremely fast context switch (on the order of nanoseconds), whereas the single context may take milliseconds or more to reprogram. The partially reconfigurable architecture is also more suited to run-time reconfiguration than the single context, because small areas of the array can be modified without requiring that the entire logic array be reprogrammed.
For all of these run-time reconfigurable architectures, there are also some compilation issues which are not encountered in systems that only configure at the beginning of an application. For example, run-time reconfigurable systems are able to optimize based on values that are known only at run-time. Furthermore, compilers must consider the run-time reconfigurability when generating the different circuit mappings, not only to be aware of the increase in time-multiplexed capacity, but also to schedule reconfigurations so as to minimize the overhead that they incur. These software issues, as well as an
overview of methods to perform fast configuration, will be explored in the sections that follow.
1.3 Motivation and Objective
Now graphics accelerators are usually made by dedicated hardware, it can achieve very high performance but it loses the flexibility. Reconfigurable computing provides another way to accelerate the computer graphics applications which retain flexibility of a software solution. Once if we use reconfigurable hardware to accelerate such a application, scheduling is a critical issue that execution time of each stage in rendering process is not constant.
Here our objective is to analyze operations, parallelism and computation flow of rendering process to design a Run-time reconfiguration scheduling.
1.4 Organization of This Thesis
The organization of this thesis is as follows: In Chapter 2, the background is presented.
In Chapter 3, analysis of the rendering process and the design of Run-time
reconfiguration are described. In Chapter 4, we analyze our simulation result and show our simulation environment. Finally, conclusion and future work are presented in Chapter 5.