University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 6

• (TODO) YOUR NAME HERE
• Tested on: (TODO) Google Chrome 222.2 on Windows 22, i7-2222 @ 2.22GHz 22GB, GTX 222 222MB (Moore 2222 Lab)

DO NOT leave the README to the last minute! It is a crucial part of the project, and we will not be able to grade you without a good README.

This assignment has a considerable amount of performance analysis compared to implementation work. Complete the implementation early to leave time!

This is due at midnight on the evening of Tuesday, October 27.

Summary: In this project, you'll be introduced to the basics of deferred shading and WebGL. You'll use GLSL and WebGL to implement a deferred shading pipeline and various lighting and visual effects.

Recommendations: Take screenshots as you go. Use them to document your progress in your README!

Read (or at least skim) the full README before you begin, so that you know what to expect and what to prepare for.

If you have Python, you should be able to run server.py to start a server. Then, open http://localhost:10565/ in your browser.

This project requires a WebGL-capable web browser with support for WEBGL_draw_buffers. You can check for support on WebGL Report.

Google Chrome seems to work best on all platforms. If you have problems running the starter code, use Chrome or Chromium, and make sure you have updated your browser and video drivers.

In Moore 100C, both Chrome and Firefox work. See below for notes on profiling/debugging tools.

Use the screenshot button to save a screenshot.

Ask on the mailing list for any clarifications.

In this project, you are given code for:

• Loading OBJ files and color/normal map textures
• Camera control
• Partial implementation of deferred shading including many helper functions

Before doing performance analysis, you must disable debug mode by changing debugMode to false in framework.js. Keep it enabled when developing - it helps find WebGL errors much more easily.

You will need to perform the following tasks:

• Complete the deferred shading pipeline so that the Blinn-Phong and Post1 shaders recieve the correct input. Go through the Starter Code Tour before continuing!

Effects:

• Implement deferred Blinn-Phong shading (diffuse + specular) for point lights

  • With normal mapping (code provided)
  • For deferred shading, you want to use a lighting model for the point lights which has a limited radius - so that adding a scissor or proxy geometry will not cause parts of the lighting to disappear. It should look very similar both with and without scissor/proxy optimization. Here is a convenient lighting model, but you can also use others:
    • float attenuation = max(0.0, u_lightRad - dist_from_surface_to_light);

• Implement one of the following effects:

  • Bloom using post-process blur (box or Gaussian) [1]
  • Toon shading (with ramp shading + simple depth-edge detection for outlines)

Optimizations:

• Scissor test optimization: when accumulating shading from each point light source, only render in a rectangle around the light.

  • Show a debug view for this (showing scissor masks clearly), e.g. by modifying and using red.frag.glsl with additive blending and alpha = 0.1.
  • Code is provided to compute this rectangle for you, and there are comments at the relevant place in deferredRender.js with more guidance.
    • NOTE: The provided scissor function is not very accurate - it is a quick hack which results in some errors (as can be seen in the live demo).

• Optimized g-buffer format - reduce the number and size of g-buffers:

  • Ideas:
    • Pack values together into vec4s
    • Use 2-component normals
    • Quantize values by using smaller texture types instead of gl.FLOAT
    • Reduce number of properties passed via g-buffer, e.g. by:
      • Applying the normal map in the copy shader pass instead of copying both geometry normals and normal maps
      • Reconstructing world space position using camera matrices and X/Y/depth
  • For credit, you must show a good optimization effort and record the performance of each version you test, in a simple table.
    • It is expected that you won't need all 4 provided g-buffers for a basic pipeline - make sure you disable the unused ones.
  • See mainly: copy.frag.glsl, deferred/*.glsl, deferredSetup.js

You must do at least 10 points worth of extra features (effects or optimizations/analysis).

Effects:

• (3pts) The effect you didn't choose above (bloom or toon shading)

• (3pts) Screen-space motion blur (blur along velocity direction) [3]

• (2pts) Allow variability in additional material properties

  • Include other properties (e.g. specular coeff/exponent) in g-buffers
  • Use this to render objects with different material properties
  • These may be uniform across one model draw call, but you'll have to show multiple models

Optimizations/Analysis:

• (2pts) Improved screen-space AABB for scissor test (smaller/more accurate than provided - but beware of CPU/GPU tradeoffs)

• (3pts) Two-pass Gaussian blur using separable convolution (using a second postprocess render pass) to improve bloom or other 2D blur performance

• (4-6pts) Light proxies

  • (4pts) Instead of rendering a scissored full-screen quad for every light, render some proxy geometry which covers the part of the screen affected by the light (e.g. a sphere, for an attenuated point light).
    • A model called sphereModel is provided which can be drawn in the same way as the code in drawScene. (Must be drawn with a vertex shader which scales it to the light radius and translates it to the light position.)
  • (+2pts) To avoid lighting geometry far behind the light, render the proxy geometry (e.g. sphere) using an inverted depth test (gl.depthFunc(gl.GREATER)) with depth writing disabled (gl.depthMask). This test will pass only for parts of the screen for which the backside of the sphere appears behind parts of the scene.
    • Note that the copy pass's depth buffer must be bound to the FBO during this operation!
  • Show a debug view for this (showing light proxies)
  • Compare performance of this, naive, and scissoring.

• (8pts) Tile-based deferred shading with detailed performance comparison

  • On the CPU, check which lights overlap which tiles. Then, render each tile just once for all lights (instead of once for each light), applying only the overlapping lights.
    • The method is described very well in Yuqin & Sijie's README.
    • This feature requires allocating the global light list and tile light index lists as shown at this link. These can be implemented as textures.
  • Show a debug view for this (number of lights per tile)

• (6pts) Deferred shading without multiple render targets (i.e. without WEBGL_draw_buffers).

  • Render the scene once for each target g-buffer, each time into a different framebuffer object.
  • Include a detailed performance analysis, comparing with/without WEBGL_draw_buffers (like in the Mozilla blog article).

• (2-6pts) Compare performance to equivalently-lit forward-rendering:

  • (2pts) With no forward-rendering optimizations
  • (+2pts) Coarse, per-object back-to-front sorting of geometry for early-z
    • (Of course) must render many objects to test
  • (+2pts) Z-prepass for early-z

This extra feature list is not comprehensive. If you have a particular idea that you would like to implement, please contact us first (preferably on the mailing list).

Where possible, all features should be switchable using the GUI panel in ui.js.

Before doing performance analysis, you must disable debug mode by changing debugMode to false in framework.js. Keep it enabled when developing - it helps find WebGL errors much more easily.

Optimize your JavaScript and/or GLSL code. Chrome/Firefox's profiling tools (see Resources section) will be useful for this. For each change that improves performance, show the before and after render times.

For each new effect feature (required or extra), please provide the following analysis:

• Concise overview write-up of the feature.
• Performance change due to adding the feature.
  • If applicable, how do parameters (such as number of lights, etc.) affect performance? Show data with simple graphs.
    • Show timing in milliseconds, not FPS.
• If you did something to accelerate the feature, what did you do and why?
• How might this feature be optimized beyond your current implementation?

For each performance feature (required or extra), please provide:

• Concise overview write-up of the feature.
• Detailed performance improvement analysis of adding the feature
  • What is the best case scenario for your performance improvement? What is the worst? Explain briefly.
  • Are there tradeoffs to this performance feature? Explain briefly.
  • How do parameters (such as number of lights, tile size, etc.) affect performance? Show data with graphs.
    • Show timing in milliseconds, not FPS.
  • Show debug views when possible.
    • If the debug view correlates with performance, explain how.

You'll be working mainly in deferredRender.js using raw WebGL. Three.js is included in the project for various reasons. You won't use it for much, but its matrix/vector types may come in handy.

It's highly recommended that you use the browser debugger to inspect variables to get familiar with the code. At any point, you can also console.log(some_var); to show it in the console and inspect it.

The setup in deferredSetup is already done for you, for many of the features. If you want to add uniforms (textures or values), you'll change them here. Therefore, it is recommended that you review the comments to understand the process, BEFORE starting work in deferredRender.

In deferredRender, start at the START HERE! comment. Work through the appropriate TODOs as you go - most of them are very small. Test incrementally (after implementing each part, instead of testing all at once).

• (The first thing you should be doing is implementing the fullscreen quad!)
• See the note in the Debugging section on how to test the first part of the pipeline incrementally.

Your next first goal should be to get the debug views working. Add code in debug.frag.glsl to examine your g-buffers before trying to render them. (Set the debugView in the UI to show them.)

For editing JavaScript, you can use a simple editor with syntax highlighting such as Sublime, Vim, Emacs, etc., or the editor built into Chrome.

• js/: JavaScript files for this project.
  • main.js: Handles initialization of other parts of the program.
  • framework.js: Loads the scene, camera, etc., and calls your setup/render functions. Hopefully, you won't need to change anything here.
  • deferredSetup.js: Deferred shading pipeline setup code.
    • createAndBind(Depth/Color)TargetTexture: Creates empty textures for binding to frame buffer objects as render targets.
  • deferredRender.js: Your deferred shading pipeline execution code.
    • renderFullScreenQuad: Renders a full-screen quad with the given shader program.
  • ui.js: Defines the UI using dat.GUI.
    • The global variable cfg can be accessed anywhere in the code to read configuration values.
  • utils.js: Utilities for JavaScript and WebGL.
    • abort: Aborts the program and shows an error.
    • loadTexture: Loads a texture from a URL into WebGL.
    • loadShaderProgram: Loads shaders from URLs into a WebGL shader program.
    • loadModel: Loads a model into WebGL buffers.
    • readyModelForDraw: Configures the WebGL state to draw a model.
    • drawReadyModel: Draws a model which has been readied.
    • getScissorForLight: Computes an approximate scissor rectangle for a light in world space.
• glsl/: GLSL code for each part of the pipeline:
  • clear.*.glsl: Clears each of the NUM_GBUFFERS g-buffers.
  • copy.*.glsl: Performs standard rendering without any fragment shading, storing all of the resulting values into the NUM_GBUFFERS g-buffers.
  • quad.vert.glsl: Minimal vertex shader for rendering a single quad.
  • deferred.frag.glsl: Deferred shading pass (for lighting calculations). Reads from each of the NUM_GBUFFERS g-buffers.
  • post1.frag.glsl: First post-processing pass.
• lib/: JavaScript libraries.
• models/: OBJ models for testing. Sponza is the default.
• index.html: Main HTML page.
• server.bat (Windows) or server.py (OS X/Linux): Runs a web server at localhost:10565.

See the comments in deferredSetup.js/deferredRender.js for low-level guidance.

In order to enable and disable effects using the GUI, upload a vec4 uniform where each component is an enable/disable flag. In JavaScript, the state of the UI is accessible anywhere as cfg.enableEffect0, etc.

Pass 1: Renders the scene geometry and its properties to the g-buffers.

• copy.vert.glsl, copy.frag.glsl
• The framebuffer object pass_copy.fbo must be bound during this pass.
• Renders into pass_copy.depthTex and pass_copy.gbufs[i], which need to be attached to the framebuffer.

Pass 2: Performs lighting and shading into the color buffer.

• quad.vert.glsl, deferred/blinnphong-pointlight.frag.glsl
• Takes the g-buffers pass_copy.gbufs/depthTex as texture inputs to the fragment shader, on uniforms u_gbufs and u_depth.
• pass_deferred.fbo must be bound.
• Renders into pass_deferred.colorTex.

Pass 3: Performs post-processing.

• quad.vert.glsl, post/one.frag.glsl
• Takes pass_BlinnPhong_PointLight.colorTex as a texture input u_color.
• Renders directly to the screen if there are no additional passes.

More passes may be added for additional effects (e.g. combining bloom with motion blur) or optimizations (e.g. two-pass Gaussian blur for bloom)

If there is a WebGL error, it will be displayed on the developer console and the renderer will be aborted. To find out where the error came from, look at the backtrace of the error (you may need to click the triangle to expand the message). The line right below wrapper @ webgl-debug.js will point to the WebGL call that failed.

When working in the early pipeline (before you have a lit render), it can be useful to render WITHOUT post-processing. To do this, you have to make sure that there is NO framebuffer bound while rendering to the screen (that is, bind null) so that the output will display to the screen instead of saving into a texture. Writing to gl_FragData[0] is the same as writing to gl_FragColor, so you'll see whatever you were storing into the first g-buffer.

Note that the g-buffers are just vec4s - you can put any values you want into them. However, if you want to change the total number of g-buffers (add more for additional effects or remove some for performance), you will need to make changes in a number of places:

• deferredSetup.js/deferredRender.js: search for NUM_GBUFFERS
• copy.frag.glsl
• deferred.frag.glsl
• clear.frag.glsl

• [1] Bloom: GPU Gems, Ch. 21
• [2] Screen-Space Ambient Occlusion: Floored Article
• [3] Post-Process Motion Blur: GPU Gems 3, Ch. 27

Also see: The articles linked in the course schedule.

Built into Firefox:

• Canvas inspector
• Shader Editor
• JavaScript debugger and profiler

Built into Chrome:

• JavaScript debugger and profiler

Plug-ins:

• Web Tracing Framework Does not currently work with multiple render targets, which are used in the starter code.
• (Chrome) Shader Editor

Libraries:

• Stats.js (already included)

Firefox can also be useful - it has a canvas inspector, WebGL profiling and a shader editor built in.

Replace the contents of this README.md in a clear manner with the following:

• A brief description of the project and the specific features you implemented.
• At least one screenshot of your project running.
• A 30+ second video of your project running showing all features. Open Broadcaster Software is recommended. (Even though your demo can be seen online, using multiple render targets means it won't run on many computers. A video will work everywhere.)
• A performance analysis (described below).

See above.

Since this assignment is in WebGL, you can make your project easily viewable by taking advantage of GitHub's project pages feature.

Once you are done with the assignment, create a new branch:

git branch gh-pages

Push the branch to GitHub:

git push origin gh-pages

Now, you can go to <user_name>.github.io/<project_name> to see your renderer online from anywhere. Add this link to your README.

1. Open a GitHub pull request so that we can see that you have finished. The title should be "Submission: YOUR NAME".
   • ADDITIONALLY: In the body of the pull request, include a link to your repository.
2. Send an email to the TA (gmail: kainino1+cis565@) with:
   • Subject: in the form of [CIS565] Project N: PENNKEY.
   • Direct link to your pull request on GitHub.
   • Estimate the amount of time you spent on the project.
   • If there were any outstanding problems, briefly explain.
   • List the extra features you did.
   • Feedback on the project itself, if any.

• Use of any third-party code must be approved by asking on our mailing list.
• If it is approved, all students are welcome to use it. Generally, we approve use of third-party code that is not a core part of the project. For example, for the path tracer, we would approve using a third-party library for loading models, but would not approve copying and pasting a CUDA function for doing refraction.
• Third-party code MUST be credited in README.md.
• Using third-party code without its approval, including using another student's code, is an academic integrity violation, and will, at minimum, result in you receiving an F for the semester.