Advanced shading language for production GI renderers
Welcome to Open Shading Language!
Open Shading Language (OSL) is a small but rich language for programmable shading in advanced renderers and other applications, ideal for describing materials, lights, displacement, and pattern generation.
OSL was developed by Sony Pictures Imageworks for use in its in-house renderer used for feature film animation and visual effects. The language specification was developed with input by other visual effects and animation studios who also wish to use it.
OSL is robust and production-proven, and was the exclusive shading system for big VFX films such as "Men in Black 3: and "The Amazing Spider-Man," animated features such as "Hotel Transylvania", and several other films currently in production.
OSL is distributed under the "New BSD" license (see the "LICENSE" file that comes with the distribution). In short, you are free to use it in your own applications, whether they are free or commercial, open or proprietary, as well as to modify the OSL code as you desire, provided that you retain the original copyright notices as described in the license.
OSL has syntax similar to C, as well as other shading languages. However, it is specifically designed for advanced rendering algorithms and has features such as radiance closures, BSDFs, and deferred ray tracing as first-class concepts.
OSL has several unique characteristics not found in other shading languages (certainly not all together). Here are some things you will find are different in OSL compared to other languages:
Surface and volume shaders compute radiance closures, not final colors.
OSL's surface and volume shaders compute an explicit symbolic description, called a "closure", of the way a surface or volume scatters light, in units of radiance. These radiance closures may be evaluated in particular directions, sampled to find important directions, or saved for later evaluation and re-evaluation. This new approach is ideal for a physically-based renderer that supports ray tracing and global illumination.
In contrast, other shading languages usually compute just a surface color as visible from a particular direction. These old shaders are "black boxes" that a renderer can do little with but execute to find this one piece of information (for example, there is no effective way to discover from them which directions are important to sample). Furthermore, the physical units of lights and surfaces are often underspecified, making it very difficult to ensure that shaders are behaving in a physically correct manner.
Surface and volume shaders do not loop over lights or shoot rays.
There are no "light loops" or explicitly traced illumination rays in OSL surface shaders. Instead, surface shaders compute a radiance closure describing how the surface scatters light, and a part of the renderer called an "integrator" evaluates the closures for a particular set of light sources and determines in which directions rays should be traced. Effects that would ordinarily require explicit ray tracing, such as reflection and refraction, are simply part of the radiance closure and look like any other BSDF.
Advantages of this approach include that integration and sampling may be batched or re-ordered to increase ray coherence; a "ray budget" can be allocated to optimally sample the BSDF; the closures may be used by for bidirectional ray tracing or Metropolis light transport; and the closures may be rapidly re-evaluated with new lighting without having to re-run the shaders.
Surface and light shaders are the same thing.
OSL does not have a separate kind of shader for light sources. Lights are simply surfaces that are emissive, and all lights are area lights.
Transparency is just another kind of illumination.
You don't need to explicitly set transparency/opacity variables in the shader. Transparency is just another way for light to interact with a surface, and is included in the main radiance closure computed by a surface shader.
Renderer outputs (AOV's) may be specified using "light path expressions."
Sometimes it is desirable to output images containing individual lighting components such as specular, diffuse, reflection, individual lights, etc. In other languages, this is usually accomplished by adding a plethora of "output variables" to the shaders that collect these individual quantities.
OSL shaders need not be cluttered with any code or output variables to accomplish this. Instead, there is a regular-expression-based notation for describing which light paths should contribute to which outputs. This is all done on the renderer side (though supported by the OSL implementation). If you desire a new output, there is no need to modify the shaders at all; you only need to tell the renderer the new light path expression.
Shaders are organized into networks.
OSL shaders are not monolithic, but rather can be organized into networks of shaders (sometimes called a shader group, graph, or DAG), with named outputs of some nodes being connected to named inputs of other nodes within the network. These connections may be done dynamically at render time, and do not affect compilation of individual shader nodes. Furthermore, the individual nodes are evaluated lazily, only when their outputs are "pulled" from the later nodes that depend on them (shader writers may remain blissfully unaware of these details, and write shaders as if everything is evaluated normally).
Arbitrary derivatives without grids or extra shading points.
In OSL, you can take derivatives of any computed quantity in a shader, and use arbitrary quantities as texture coordinates and expect correct filtering. This does not require that shaded points be arranged in a rectangular grid, or have any particular connectivity, or that any "extra points" be shaded. This is because derivatives are not computed by finite differences with neighboring points, but rather by "automatic differentiation", computing partial differentials for the variables that lead to derivatives, without any intervention required by the shader writer.
OSL optimizes aggressively at render time
OSL uses the LLVM compiler framework to translate shader networks into machine code on the fly (just in time, or "JIT"), and in the process heavily optimizes shaders and networks with full knowledge of the shader parameters and other runtime values that could not have been known when the shaders were compiled from source code. As a result, we are seeing our OSL shading networks execute 25% faster than the equivalent shaders hand-crafted in C! (That's how our old shaders worked in our renderer.)
The OSL open source distribution consists of the following components:
oslc, a standalone compiler that translates OSL source code into an assembly-like intermediate code (in the form of .oso files).
liboslc, a library that implements the OSLCompiler class, which contains the guts of the shader compiler, in case anybody needs to embed it into other applications and does not desire for the compiler to be a separate executable.
liboslquery, a library that implements the OSLQuery class, which allows applications to query information about compiled shaders, including a full list of its parameters, their types, and any metadata associated with them.
oslinfo, a command-line program that uses liboslquery to print to the console all the relevant information about a shader and its parameters.
liboslexec, a library that implements the ShadingSystem class, which allows compiled shaders to be executed within an application. Currently, it uses LLVM to JIT compile the shader bytecode to x86 instructions.
testshade, a program that lets you execute a shader (or connected shader network) on a rectangular array of points, and save any of its outputs as images. This allows for verification of shaders (and the shading system) without needing to be integrated into a fully functional renderer, and is the basis for most of our testsuite verification. Along with testrender, testshade is a good example of how to call the OSL libraries.
testrender, a tiny ray-tracing renderer that uses OSL for shading. Features are very minimal (only spheres are permitted at this time) and there has been no attention to performance, but it demonstrates how the OSL libraries may be integrated into a working renderer, what interfaces the renderer needs to supply, and how the BSDFs/radiance closures should be evaluated and integrated (including with multiple importance sampling).
A few sample shaders.
Documentation -- at this point consisting of the OSL language specification (useful for shader writers), but in the future will have detailed documentation about how to integrate the OSL libraries into renderers.
Please see the "INSTALL" file in the OSL distribution for instructions for building the OSL source code.
At Sony Pictures Imageworks, we are exclusively using OSL in our proprietary renderer, "Arnold." Completed productions that used 100% OSL for their shading include Men in Black 3, The Amazing Spider-Man, and Hotel Transylvania, and other unreleased shows that are still in production. Our shader-writing team works entirely in OSL, all productions use OSL, and we've even removed all the code from the renderer that allows people to write the old-style "C" shaders. At the time we removed the old shader facility, the OSL shaders were consistently outperforming their equivalent old compiled C shaders in the old system.
In the longer term, there are a number of projects we hope to get to leading to a 2.x or 3.x cut of the language and library. Among our long-term goals:
More documentation, in particular the "Integration Guide" that documents all the public APIs of the OSL libraries that you use when integrating into a renderer. Currently, the source code to "testrender" is the best/only example of how to integrate OSL into a renderer.
Our set of sample shaders is quite anemic. We will eventually have a more extensive set of useful, production-quality shaders and utility functions you can call from your shaders.
Currently "closure primitives" are implemented in C++ in the OSL library or in the renderer, but we would like a future spec of the language to allow new closure primitives to be implemented in OSL itself.
Similarly, integrators are now implemented in the renderer, but we want a future OSL release to allow new integrators to be implemented in OSL itself.
We would like to implement alternate "back ends" that would allow translation of OSL shaders (and shader networks) into code that can run on GPUs or other exotic hardware (at least for the biggest subset of OSL that can be expressed on such hardware). This would, for example, allow you to view close approximations to your OSL shaders in realtime preview windows in a modeling system or lighting tool.
We (the renderer development team at Sony Pictures Imageworks) probably can't do these all right away (in fact, probably can't do ALL of them in any time range). But we hope that as an open source project, other users and developers will step up to help us explore more future development avenues for OSL than we would be able to do alone.
Read or subscribe to the OSL development mail list
Email the lead architect: lg AT imageworks DOT com
Most recent PDF of the OSL language specification
Sony Pictures Imageworks main open source page
If you want to contribute code back to the project, you'll need to sign a Contributor License Agreement.
The main developers of OSL are (in order of joining the project):
Larry Gritz
Cliff Stein
Chris Kulla
Alejandro Conty
Jay Reynolds
Solomon Boulos
Adam Martinez
We cannot possibly express sufficient gratitude to the managers at Sony Pictures Imageworks who allowed this project to proceed, supported it wholeheartedly, and permitted us to release the source, especially Rob Bredow, Brian Keeney, Barbara Ford, and Rene Limberger.
Huge thanks also go to the crack shading team at SPI, and the brave lookdev TDs and CG supes willing to use OSL on their shows. They served as our guinea pigs, inspiration, testers, and a fantastic source of feedback. Thank you, and we hope we've been responsive to your needs.
OSL was not developed in isolation. We owe a debt to the individuals and studios who patiently read early drafts of the language specification and gave us very helpful feedback and additional ideas. (I hope to mention them by name after we get permission of the people and studios involved.)
The OSL implementation incorporates or depends upon several other open source packages:
OpenImageIO (c) Larry Gritz, et al