Although the real-time GI technology has matured, but for the performance and quality of the different platforms to consider, more inclined to match some of the expected rendering technology to achieve, such as to provide the static object GI Lightmap, to the dynamic object to provide GI irradiance Volume, As well as to increase the GI specualar details of IBL and so on. Especially in the case of mobile platforms. In order to improve the speed of the calculation, reduce the art iteration time, but also to start to consider these technologies based on GPGPU, the latter two irradiance volume data is relatively small, the data organization is not affected by the scene, for some smaller scenarios before the operation of pre-calculation is possible, And the IBL one is each pre-filter envmap relatively independent, the range is relatively small, in addition envmap can also be used on multiple objects at the same time. Therefore, it is not difficult for GPGPU to realize the complexity and memory usage. And Lightmap, one is almost impossible to reuse on multiple models, in addition to the use of video memory will be very large, especially for that large-scale game, SQUARE Enix this PPT introduced them based on the Ray Bundle tracing, and use perpixel linked list to achieve GPGPU Lightmap baking design ideas and implementation methods. It is also a good reference for us to realize GPU baking. In addition to the basic methods, this article extends to support tessellation, support for the BRDF equation outside the Lambert, how to correspond to the super-large scene, and some of the scenes modified how to bake only a small part of the scene and so on the actual problem. Overall, it's worth seeing. of course, there should be several companies in the country have also achieved GPGPU baking. Do not know is not used this technology. because it is the conference used to explain the PPT, the details of the slide itself is not much, fortunately, the author has written the speech in the Remarks section, so this article has the meaning of translation, but the speech itself is a serious, but also bad for the author to make too many changes, but also most of the original translation came. However, after the translation I also proofread 1, 2 times, so the hair is also a bit late, as the author said, there is a graphical basis of the programmer should be able to read, if there are any questions to leave a message or contact me directly. In addition, the latest English version of the paper link I also put in the following, two although there are some intersection, but more can be done to complement it, if interested also must see, in addition, some reference data in the paper more valuable. About theSome of the earlier data from Ray Bundle Tracing,se are also available in the following SE's paper links.
PPT translation version link: http://pan.baidu.com/s/1tVEDC password: nx1p se latest ppt Adaptive ray-bundle tracing with Memory Usage Prediction:effici ENT Global illumination in Large sceneshttp://connect.jp.square-enix.com/wp-content/uploads/2014/10/ Adaptive-ray-bundle-tracing-with-memory-usage-prediction.pdf
Related URLs http://connect.jp.square-enix.com/?cat=7 In addition, individuals still want to do a GPGPU baking project to try, if anyone has achieved, or interested or questions are also welcome to discuss and guidance.
"Translation" CEDEC2012 SQUARE Enix Gpgpu method for realizing high-speed GI baking tools
