Windows7, part I

<p ALIGN="left"> Win7 as Microsoft officially launch next-generation operating system, Microsoft researchers are inspired to celebrate this moment, because in this highly anticipated new products, many researchers have devoted many years of global .innovation and research. .Incubation at Microsoft Research Asia is one very important DirectX11 a technical invention. .</ P> <p ALIGN="right"> <wbr> </ P> <p ALIGN="right"> of: Gong Minmin Network Graphics Group </ P> <p> <wbr> </ P> <p> .Windows7 toward faster and faster as the majority of users, the new features it brings more and more known. .Among them, it has a little-known features - Windows 7 comes with DirectX11, support named DirectCompute of GPGPU (GPGPU) programming interface. .In other words, Windows7 became the first native operating system support for GPGPU. .With the support of GPGPU, Windows7 to the large number of developers and users to open the door to parallel computing. .</ P> <p> <wbr> </ P> <p> progressive refinement is perfect </ P> <p> <wbr> </ P> <p> in DirectX11 in the development process, the relevant product group .people are thinking, how can the power of GPGPU, how can other developers to demonstrate how to use the new GPGPU interface? .Microsoft Research Asia Network Graphics group (IG group) in the GPGPU field has many years of research, has accumulated a lot of skills and experience. .So the headquarters is a natural product group of colleagues to find us, want to see our group there any good idea. .</ P> <p> <wbr> </ P> <p> made for the product group this demand, and soon, we proposed to try to move the triangle subdivision on the GPU. .Subdivision scheme is widely used in the film industry. .From the early "Toy Story" to the most recent "Ice Age 3", rounded fine on the big screen models that are ultimately subdivision credit. .The analysis of subdivision algorithms can be found, the algorithm has traditionally done only by the CPU, because it involves many irregular, you need a serial to a step by step implementation. .If you insist on using GPU to achieve, it is necessary to involve a large number of external IO, reducing the efficiency of the subdivision. .DirectCompute bring new features enabling developers to use more flexible and efficient GPU, parallel those of irregular operations, without any extra external IO. .As a result, GPU to run at full speed in parallel segments. .To test how well the presentation of ideas, not too long, was born out of our first prototype demo. .We give a name from this algorithm, called iTess, meaning unlimited subdivision (infinitytessellation). .It can be rough model of the input adaptive subdivision, with more sophisticated models. .Later, the demo successfully become a DirectX11 SDK for an example. .The following diagram shows the subdivision two input and output. .</ P> <p> <wbr> <wbr> </ P> <p> you can see the left, enter the rough mesh, the lack of details. .Right, after the breakdown of the model after, fine a lot more lifelike. .</ P> <p ALIGN="center"> 7, there is a force I "TITLE =" Windows <wbr> 7, there is a force I "/> <wbr> 7, there is a force I" TITLE = ."Windows <wbr> 7, there is a force I" /> </ P> <p> <wbr> </ P> <p> image compression show their skills </ P> <p> <wbr> </ P> .<p> and product groups with the first successful co-operation, they seek new projects in our help. .In computer graphics, traditional color component is represented by 0-255, which can only show 256 different brightness changes, the scope is limited, as LDR (Low Dynamic Range) images. .The new HDR (high dynamic range) images can show the range of tens of thousands of bright, largely to make up for the shortcomings of traditional methods. .However, the attendant is a huge memory footprint. .Since the "Far Cry" for the first time since the introduction of HDR, it is in the game such as the increasing use of software, how to compress HDR data has become essential. .To this end, DirectX11 introduced a new texture compression format that is used to compress HDR images BC6H format. .</ P> <p> 7, there is a force I "TITLE =" Windows <wbr> 7, there is a force I "/> <wbr> 7, there is a force I" TITLE = "Windows <wbr> .7, there I am a force "/> </ P> <p> <wbr> <wbr> </ P> <p> Similarly, we can see in the picture to the left, LDR images, the contrast is small, the window part of the .appear gray. .Right, HDR images can show greater brightness changes, the details of more abundant after adjustments. .</ P> <p> <wbr> </ P> <p> compression BC6H computation required is very striking. .The original implementation in the product group, this process relies entirely on the CPU to calculate, compressed image often requires 10-20 minutes, or even longer. .This is obviously for the game makers is not acceptable waiting time. .To solve this problem, we still thought to accelerate the compression using DirectCompute ways to bring the compression algorithm used for analysis, improvement, making a parallel algorithm, and onto the GPU. .CPU GPU's parallel program like the program as a mature development environment and a variety of aids, and because of the characteristics of the parallel itself, often while in the debugging data to observe tens of thousands. .Results accidentally a problem, you have to spend ten times or hundred times the time and energy to deal with it. .For example, we once encountered such a situation: </ P> <p ALIGN="center"> 7, there is a force I "TITLE =" Windows <wbr> 7, there is a force I "/> < ./ P> <p ALIGN="center"> <wbr> </ P> <p> compressed image, in some places the right, in some places is not correct. in some places is a mistake, in some places is another .a mistake. such as the edge of the red point where the green has become blue. while staring at the final result and not a solution, we can only variables in a shader in a lead out of view. on the road in the dark after .hard to explore, and finally to solve the problem, and has accumulated experience in dealing with such problems. In the end, without changing the compression quality of the situation, 10 seconds can handle a picture about 100 times faster, become .a practical tool. product group is very interested in what we do, the compression process has become DirectX11SDK an example, and may enter in future releases D3DX library. Since then, HDR image is no longer a large memory consumption, it .faster access to common game, for the benefit of all gamers. </ P> <p> <wbr> </ P> <p> With the increasing popularity of Windows7, with the continuous development of software and hardware, GPGPU .everywhere parallel computing will become indispensable to each software part. Although the arrival of that day need to take time, but we are through our efforts, this goal has taken a solid step forward. </ P> <p> .<wbr> </ P> <p> About the Author: </ P> <p> <wbr> </ P> <p> Gongmin Min, in 2006 graduated from the China Science and Technology Master of Computer Science and Software Engineering University, joined the same year .Network Graphics Group Microsoft Research Asia, in the graphics related research and technology transfer work. areas of interest include real-time rendering, natural phenomena simulation, GPU general computing. </ P> <p> <wbr> </ P> < .p> Related reading: "CTCP stationed Windows story" </ P> <p> <wbr> </ P> <p> Come microblogging Microsoft Research Asia, the siege of it: http://t.sina.com. .cn / msra </ P> <p> <wbr> </ P>.