This work discusses shadowing methods, analyses them and describes implementation in DirectX 11 API. Theoretical part describes historical evolution of shadow usage in 3D applications and also analyzes shadowing algorithms. This work compares 2 variants of shadow mapping algorithm for omnidirectional lights, based on cube mapping and paraboloid projection, on demo application using quality, performance and implementation aspects.

With each computer game generation there is always a demand for more visually pleasing environments. This pushes game developers to create more powerful rendering techniques and game artists to create more detailed art. With a visually stunning backdrop also comes the need for high-resolution models. A common issue is that if all models in a scene are high-resolution it would not only require immensely powerful hardware, it would also be wasteful as only the models in the foreground are close enough that we would recognize the increased details. The common solution to this problem has been to load several versions of each model containing varying amounts of detail. However this solution has the drawback that it increases our memory footprints as more models are loaded into the memory. Tessellation offers a more dynamic solution to the problem as it only requires us to load a low-resolution model and higher resolution versions can be generated during run-time on the GPU. With the introduction of DirectX 11 tessellation is now supported in the hardware, however we are still a few years away from seeing DirectX 11 being used as the core of any 3D rendering engine. In a transitional period like this between hardware generations game developers has to tackle the dilemma that the current hardware generation has to be supported when creating games that will also utilize the next generation. This thesis focuses on comparing the performance of a tessellation scheme supported by the current hardware generation, DirectX 10, as opposed to a scheme developed for the next generation, DirectX 11. Two prototypes, one using the Geometry shader that was introduced in DirectX 10 and the other using the fixed function tessellator introduced in DirectX 11, were built to compare the performance of tessellated model rendering. Several different variants of each prototype were tested and the general conclusion is that the tessellator performed better than the Geometry shader.

It is a great challenge to develop a computer game. Today many games are developed in large game studios where lots of skilled people are working together. Everyone has to know what the final game should look like. Game designers are responsible for how the game should feel and look like. This also means that they decide if a programmer has to develop new techniques or not. Sometimes the game designers require lots of new techniques to be developed. Such a new technique may be rendering particle systems with a lot of particles in it. This is where this report will focus. To render particle systems it is necessary to know about the limitations there are in both hardware and software. Until today particle systems have been updated and calculated using the Central Processing Unit of the computer. With Microsoft Direct3D 10 there are new ways to render particles using Geometry Shaders. Geometry Shaders runs on the graphics card. This thesis focuses on testing rendering performance between using Geometry Shaders and not using Geometry Shaders. A questionnaire was sent to Swedish game developers to get more information about relevant topics for investigation. A general answer was that Geometry Shaders always increase particle rendering performance. This thesis investigates if and when the statement is true or not. The hypothesis was obtained from the answers to the questionnaire. Two test applications were used to investigate if the hypothesis was true or false. One test application has particle calculations on the CPU of the computer. The other test application has particle calculations on the GPU of the graphics card. Six different tests were done and the Geometry Shader approach went out to be the fastest in five of the tests. Since not all tests were faster than the CPU approach the hypothesis is not always true.