Project description
Project Title
Shader Enhancements
Brief project summary
Shaders in BRL-CAD are implemented using C language. Recently Sony released its shading language as open source, now called OSL. This is a C-like language that works like a framework to write shaders. It was specially developed to work with raytracers, which is also the application of BRL-CAD shaders. To this end, an idea is to implement BRL-CAD shaders using OSL. In this proposal I provide a brief description of BRL-CAD shader system and OSL specification. Then I describe a top-level approach to accomplish this task.
Detailed project description
Introduction
I've little (if none) experience with shaders, but I really want to learn about them and this project is an opportunity to both learn and contribute to open source. To this end, I'm declaring this project to have the greatest priority among the others I'm applying. In the past weeks, I've been researching BRL-CAD code and OSL to better understand how shaders sytems work. I can now devise how BRL-CAD could benefit in using OSL language to implement their shaders.
BRL-CAD Shader System
A shader is implemented at liboptical with the name sh_
BRL-CAD shaders are currently written in plain C. They're mainly used by the raytracer system, both in the stand-alone rt application or through mged application using librt. Shaders are compiled as shared libraries and are loaded dynamically at runtime.
OSL
Recently Sony opened the code of their shader language [3]. Chatting at IRC and from the project description, it's clear that this language is to be carefully analyzed in order to benefit BRL-CAD shader system, specially because it is developed for raytracers, which is the same use BRL-CAD current system has. OSL is a C-like language that can be used to write shaders.
One can write individual shaders using OSL and integrate them afterwards using shaders groups. These groups are basically a network (more specifically, a directed acyclic graph, also known as DAG) of individual shaders, called layers, and they're connected like UNIX pipes: the output of one is the input of the other. Using OSL, setting up this network is simple as coding the individual shaders and defining a set of pairs of (ouput shader, input shader). This is like the stacking of shaders present in BRL-CAD.
In general, the output of a shader is not a final color but instead it's a structure called closure. It stores parametrized formulas that would be used to evaluate the final color but the variables are maintained. This has several advantages such as:
We can recompute colors more quickly when there are changes only in the lights We can compute the final color for given only a direction. Given an input and output direction, it's possible to get the amount of light propagated from the input to the output. The last two items are quite interesting for raytracers.
Proposal Idea
The OSL represents a interesting framework to develop shaders. BRL-CAD shaders could benefit from it by improving performance with the concept of closures. Also, we could take advantage of already implemented shaders for OSL and when developing new shaders for BRL-CAD we would be contributing to OSL database, which is, according to [6] (may be outdated) quite sparse.
There are some issues to solve first. In the OSL specification document from January 2010, it's said that some components are assumed to exist in order to implement shaders using OSL. One of them is the integrator, which combines the color closure, lights and view-dependent variables. They say in the future they would provide a way to implement them through OSL, though I'm not sure what is the current state. If not, it would need to be included in the project.
Here is how I plan to execute this project: study the details of BRL-CAD and OSL shader. This includes implementing shaders for these systems as well a simple integration test, which can be done using the single ray shooter from BRL-CAD with no optics or shaders and plugging in the OSL shader. This will assess the feasibility of the project as well as provide an idea on which changes are necessary to adapt OSL shaders. So, we'll be able to design and implement these changes to the BRL-CAD shader system.
Reuse shaders that are already implemented with OSL and that fits the current shaders of BRL-CAD. Then, implement possibly missing shaders. After implementing/porting the first shader, we must check consistency of the new shader system, testing if it works well with the raytracers. Also, after each shader is added, we need to check if the result is similar to the replaced one. Extra tasks that can be done, if time allows it, is to implement new shaders with this language and/or write a small tutorial on how to implement shaders using OSL.
Timeline
- Community Bonding (4 weeks)
- Compile OSL [ok]
- Discover how testrender application works [ok]
- Implement and compile a shader using OSL [ok]
- Test the shader with testrender [ok]
- Week 1 (May 23th to May 30th)
- Implement a sample texture/material shader (polka-dot texture) for BRL-CAD [ok]
- Test it by assigning it to an object and rendering it. [ok]
- Week 2-6 (May 30th to Jul 4th)
- Implement a special shader for BRL-CAD that will be an interface for OSL system [ok]
- Integrate a shader implemented in OSL and this interface [ok]
- Week 7-8 (Jul 4th to Jul 18th)
- Implement some BRL-CAD shaders as OSL shaders
- Mirror
- Glass
- Cloud
- Checker
- Check if any additional changes in the system is necessary to render the new shaders
- Implement some BRL-CAD shaders as OSL shaders
- Week 9-12 (Jul 18th to Aug 15th)
- Resolve the multi-threading issue
- Extra: implement a mode in rt so the framebuffer updates the screen at every sample
- Extra: write a tutorial