LunarG

New LunarGOO Front End

Jens — Fri, 15 Apr 2011 17:36:41 +0000

The feedback on LunarGOO has been great. Getting shaders and comments from users is much more engaging than pondering what they might want to compile or say about our compiler. Our team continues to bring up new functionality at a rapid pace, and the web front has definitely been lagging…but that’s starting to change now that we have some dedicated help with that. Our first update to the front end just went up and it gives a much better idea of what our compiler is capable of…which is mixing and matching input languages, transforms and output languages. Some of the functionality that’s been requested if “greyed” out, and itsn’t support, yet…but we’re working to more and more online everyday. If there is functionality you need and it’s not listed on the page, let us know. We would love to hear your feedback.

The Next Online Service: 3D Drivers

Jens — Mon, 21 Mar 2011 13:31:08 +0000

1986: This is when I started developing 3D drivers. The internet only connected a very small number of folks. Sure, I could use it to send e-mail to my colleagues at HP, but I certainly never expected the entire planet to go online years later. In those days, we were told about big customers and were even allowed to attend a few meetings to hear about their 3D needs, but the idea of connecting directly with a large base of users (big and small), was well beyond our thinking.

1992: Fast forward a six years, and you see 3D driver developers starting to connect with one another online. Brian Paul starts the Mesa open source project, planting the seeds of the first online 3D driver development community. However, only a handful of 3D driver developers connected online at this time.

1998: Another six years go by before we see the 3Dfx Glide Driver ported to Mesa. This project provides the first open source driver with decent 3D HW acceleration. More 3D Driver Developers are starting to show up online, and a handful of brave users are starting to kick the tires on this new 3D driver. These users are brave, patient and capable of building and trouble shooting in a bare-bones, open-source environment.

2004: Another six years go by, and we start to see broad 3D driver support online. Open source and closed source 3D drivers are delivered via the web, and the users of these drivers now have high quality drivers with multiple support channels available to them via the web. However, the validation and support infrastructure required to achieve this broad support places large organizations of people and processes in between the 3D driver developers and users. Software as a service is starting emerge in other areas, but the common understanding is the something like a 3D driver is part of your operating system, and not something that can be served up over the web.

Today: We are starting to see the value of putting our first 3D driver components out as an online service. Most of our initial efforts are focused on the shader compiler, required by modern 3D hardware. The online capabilities can compliment the native shader compilers found in the OS of each device by providing deeper offline optimizations, language translation, and quick insights into platform compatibility. But even more valuable, is the direct feedback provided by the users, enabling 3D driver and shader compiler developers to make enhancement, fix bugs and update their service on daily basis.

Going forward, this direct connection between 3D driver developers, shader compiler developers and their users is going to continue to accelerate the rate of 3D innovation, dramatically decreasing the response time for fixes, and increasing the breadth of support for diverse new applications coming to the web via 3D standards like WebGL.

Look for it at a browser near you.

The Story Behind LunarGOO

Jens — Mon, 14 Mar 2011 23:08:42 +0000

LunarG has been developing LunarGLASS for about 4-5 months now, but less than a month ago we had a company retreat and came up with the idea for using this technology to create LunarGOO: LunarG’s Online Optimizing Shader Compiler. This is the story behind LunarGOO, and how we came to putting this free service online in less than a month.

In mid-February, about 4 months into developing LunarGLASS, we were working to identify which graphics platform we should use to demonstrate the benefits our new approach to developing shader compiler stacks.

The benefits we want to show are:

1) REDUCE the development burden of creating advanced shader compiler stacks

2) INCREASE the level of optimization achieved by real world applications

The challenge we faced is choosing an initial target that has:

A) A relatively fast GPU

B) Access to a thin, and efficient driver implementation that delivers good performance

C) Access to a good compiler back end

We felt, and still feel, that finding these characteristics in a graphics platform are necessary to provide an initial implementation where the high-level shader optimizations provided by LunarGLASS can really shine.

We were narrowing down the list to a hand full of potential platforms, but each one required a non-trivial development effort outside of LunarGLASS itself, either to optimize the 3D driver, or a specific platform’s compiler back end, or to create a translator to leverage an existing IR. We were willing to do the development necessary, but we were also concerned about the impact it would make on the great progress we were making on LunarGLASS proper. What we really wanted was a platform that allowed us to demonstrate the potential of our new approach while still keeping a majority of our development effort focused on LunarGLASS itself.

Then we found the solution. It was a combination of two ideas: the need for a GLSL optimizer, written up well on Aras Pranckevicius’ blog and our desire to engage more directly with 3D application developers and users by providing a web-based service. Frankly, we were more than a little shocked at how poor of a job some mobile platforms are doing with their native compiler optimizations, but the idea of addressing this issue with LunarGLASS felt very doable, and better yet, the missing components could be developed much quicker than a full driver level integration.

The first key component, a LunarGLASS GLSL Back End, has been developed in a little under a month, thanks to the modular LunarGLASS architecture. Our team has shifted from verifying LunarGLASS development by reading LunarGLASS Bottom IR to reading optimized GLSL. This is great stuff for internal development process, and it serves the purpose of being a GLSL optimizer for platforms that have to deal with inefficient shaders or limited optimization capabilities in the native shader compiler.

The second key component to be created was a delivery mechanism: a web front end so that anyone could run our compiler regardless of their development or target environment. We are not full time web developers, and we didn’t want to divert our compiler efforts to become web developers; so we put together a small side team of folks that could whip something up real quick and simple. The result is the LunarGOO tab you can now see on our web site.

Both the optimizing compiler and the web interface are very young, and have a long list of features that still need to be implemented. However, we have released LunarGOO today because we want to get this early version in your hands, so you can see the potential of this tool, and provide feedback to help guide our efforts to make this tool work well for you.

So, take our new shader compiler service for a spin and let us know what you would like to see.