Skip to content

Latest commit

 

History

History
422 lines (331 loc) · 30.6 KB

README.md

File metadata and controls

422 lines (331 loc) · 30.6 KB
Raw

p5.treegl

Shader development and space transformations WEBGL p5.js library.

A non-Euclidean geometry cube with faces showcasing teapot, bunny, and Buddha models.

In p5.treegl, all matrix operations are immutable. For example, invMatrix does not modify its parameter but returns a new matrix:

let matrix = new p5.Matrix()
// invMatrix doesn't modify its matrix param, it gives a new value
let iMatrix = invMatrix(matrix)
// iMatrix !== matrix

Note that functions in the Shaders and Matrix operations sections are available only to p5; those in the Matrix Queries, Space Transformations, Heads Up Display, Utilities, and Drawing Stuff sections are accessible to both p5 and p5.RendererGL instances; functions in the Frustum Queries section are available to p5, p5.RendererGL, and p5.Matrix instances.

Parameters for p5.treegl functions can be provided in any order, unless specified otherwise.

Shaders

p5.treegl simplifies the creation and application of shaders in WEBGL. It covers the essentials from setting up shaders with Setup, managing shader uniforms through a uniforms user interface, applying shaders using Apply shader, enhancing visuals with Post-effects, and setting common uniform variables using several Macros.

Have a look at the toon shading, blur with focal point, post-effects, and gpu-based photomosaic examples.

Setup

The readShader, makeShader, and parseShader functions take a fragment shader —specified in either GLSL ES 1.00 or GLSL ES 3.00— to create and return a p5.Shader object. They parse the fragment shader and use the matrices param to infer the corresponding vertex shader, which is then logged to the console if no vertex shader source is provided. These functions also create a uniformsUI user interface with p5.Elements from the fragment shader's uniform variables' comments, and if a key is provided, bind the shader to it, enabling its use as a Post-effect.

  1. readShader(fragFilename, [vertFilename], [matrices = Tree.NONE], [uniformsUIConfig], [key], [successCallback], [failureCallback]): Akin to loadShader, this function reads a fragment shader (and optionally a vertex shader) from a file, generates and logs a vertex shader if none is provided, and returns a p5.Shader instance. It builds a uniformsUI user interface using uniformsUIConfig and, if a key is given, it binds the shader to this key for potential use as a Post-effect. Note that readShader should be called within p5 preload. If both callbacks are provided, the first is used as successCallback and the second as failureCallback.
  2. makeShader(fragStr, [vertStr], [matrices = Tree.NONE], [uniformsUIConfig], [key]): Akin to createShader, this function takes a fragment shader source string (and optionally a vertex shader source string), generates and logs a vertex shader if none is provided, and returns a p5.Shader. It also sets up a uniformsUI user interface with uniformsUIConfig and, if a key is provided, binds the shader to this key for potential use as a Post-effect. Note that makeShader should be called within p5 setup.
  3. parseShader(fragSrc, [vertSrc], [matrices = Tree.NONE], [uniformsUIConfig], [key], [successCallback], [failureCallback]): A high-level dispatcher function that determines whether to call readShader or makeShader based on the input parameters and should be called within p5 preload or p5 setup, accordingly. If both callbacks are provided, the first is used as successCallback and the second as failureCallback.

Vertex shader generation observations

  • The matrices parameter uses the following mask bit fields Tree.vMatrix, Tree.pMatrix, Tree.mvMatrix, Tree.pmvMatrix, Tree.mMatrix and Tree.NONE which is the default, to determine how vertices are projected onto NDC, according to the following rules:
    Mask bit fields gl_Position
    Tree.NONE aPosition
    Tree.pmvMatrix uModelViewProjectionMatrix * aPosition
    Tree.mvMatrix | Tree.pMatrix uProjectionMatrix * uModelViewMatrix * aPosition
    Tree.pMatrix | Tree.vMatrix | Tree.mMatrix uProjectionMatrix * uViewMatrix * uModelMatrix * aPosition
    Tree.vMatrix | Tree.pMatrix uProjectionMatrix * uViewMatrix * aPosition
    Tree.pMatrix | Tree.mMatrix uProjectionMatrix * uModelMatrix * aPosition
    Tree.mvMatrix uModelViewMatrix * aPosition
    Tree.vMatrix | Tree.mMatrix uViewMatrix * uModelMatrix * aPosition
    Tree.pMatrix uProjectionMatrix * aPosition
    Tree.vMatrix uViewMatrix * aPosition
    Tree.mMatrix uModelMatrix * aPosition
  • The fragment shader's varyings variables are parsed to determine which and how vertex attributes should be interpolated from the vertex shader, following these naming conventions:
    type name space
    vec4 color4 color
    vec2 texcoords2 texture
    vec2 position2 local
    vec3 position3 local
    vec4 position4 eye
    vec3 normal3 eye

Examples:

  • Example 1: parseShader(fragSrc) WEBGL2 (GLSL ES 3.00) fragSrc, with no varyings and highp precision:

    // inferred vertex shader
#version 300 es
precision highp float;
in vec3 aPosition;
void main() {
  gl_Position = vec4(aPosition, 1.0);
}

  • Example 2: Similar to Example 1 but with WEBGL (GLSL ES 1.00):

    // inferred vertex shader
precision highp float;
attribute vec3 aPosition;
void main() {
  gl_Position = vec4(aPosition, 1.0);
}

  • Example 3: parseShader(fragSrc, Tree.pmvMatrix) WEBGL2 fragSrc defining normal3 and position4 varyings, and mediump precision:

    // shader.frag excerpt
#version 300 es
precision mediump float;
in vec3 normal3;
in vec4 position4;
// ...

    infers the following vertex shader:

    // inferred vertex shader
#version 300 es
precision mediump float;
in vec3 aPosition;
in vec3 aNormal;
uniform mat3 uNormalMatrix;
uniform mat4 uModelViewMatrix;
uniform mat4 uModelViewProjectionMatrix;
out vec3 normal3;
out vec4 position4;
void main() {
  normal3 = normalize(uNormalMatrix * aNormal);
  position4 = uModelViewMatrix * vec4(aPosition, 1.0);
  gl_Position = uModelViewProjectionMatrix * vec4(aPosition, 1.0);
}

uniformsUI

By parsing comments within glsl shader code, a shader.uniformsUI object is built, mapping uniform variable names to p5.Element instances for interactively adjusting their values.

Supported elements include sliders for int and float types, color pickers for vec4 types, and checkboxes for bool types, as highlighted in the following examples:

  • Sliders: Create a slider by annotating a uniform float or int declaration in your shader code. The comment should specify the minimum value, maximum value, default value, and step value.

    Example:

    uniform float speed; // 1, 10, 5, 0.1

    This creates a slider for speed with a range from 1 to 10, a default value of 5, and a step of 0.1. The speed slider may be accessed as custom_shader.uniformsUI.speed.

  • Color Picker: To create a color picker, annotate a vec4 uniform. The comment can specify the default color using a CSS color name.

    Example:

    uniform vec4 color; // 'magenta'

    This creates a color picker for color with a default value of magenta.

  • Checkboxes: For bool uniforms, a checkbox is created. The comment can specify the default state as true or false.

    Example:

    uniform bool isActive; // true

    This creates a checkbox for isActive that is checked by default.

These functions manipulate the uniformsUI:

  1. parseUniformsUI(shader, [{ [x = 0], [y = 0], [offset = 0], [width = 120], [color] }]): Parses shader uniform variable comments into the shader.uniformsUI map. It automatically calls configUniformsUI with the provided uniformsUIConfig object. This function should be invoked on custom shaders created with loadShader or createShader, while readShader and makeShader already call it internally.
  2. configUniformsUI(shader, [{ [x = 0], [y = 0], [offset = 0], [width = 120], [color] }]): Configures the layout and appearance of the shader.uniformsUI elements based on the provided parameters:
    • x and y: Set the initial position of the first UI element.
    • offset: Determines the spacing between consecutive UI elements.
    • width: Sets the width of the sliders and color pickers.
    • color: Specifies the text color for the UI elements' labels.
  3. showUniformsUI(shader): Displays the shader.uniformsUI elements associated with the shader's uniforms. It attaches necessary event listeners to update the shader uniforms based on user interactions.
  4. hideUniformsUI(shader): Hides the shader.uniformsUI elements and removes the event listeners, stopping any further updates to the shader uniforms from ui interactions.
  5. resetUniformsUI(shader): Hides and resets the shader.uniformsUI which should be restored with a call to parseUniformsUI(shader, configUniformsUI).
  6. setUniformsUI(shader): Iterates over the uniformsUI map and sets the shader's uniforms based on the current values of the corresponding UI elements. This method should be called within the draw loop to ensure the shader uniforms are continuously updated. Note that applyShader automatically calls this method.

Apply shader

The applyShader function applies a shader to a given scene and target, invoking setUniformsUI(shader) and enabling the passing of custom uniform values not specified in uniformsUI.

  1. applyShader(shader, [{ [target], [uniforms], [scene], [options] }]) applies shader to the specified target (which can be the current context, a p5.Framebuffer or a p5.Graphics), emits the shader uniformsUI (calling shader.setUniformsUI()) and the uniforms object (formatted as { uniform_1_name: value_1, ..., uniform_n_name: value_n }), renders geometry by executing scene(options) (defaults to an overlaying quad if not specified), and returns the target for method chaining.
  2. overlay(flip): A default rendering method used by applyShader, which covers the screen with a quad. It can also be called between beginHUD and endHUD to specify the scene geometry in screen space.

Post-effects

Post-effects1 play a key role in dynamic visual rendering, allowing for the interactive blending of various shader effects such as bloom, motion blur, ambient occlusion, and color grading, into a rendered scene. A user-space array of effects may be sequentially applied to a source with applyEffects(source, effects, [uniforms], [flip]). Example usage:

// noise_shader
uniform sampler2D blender; // <- shared source should be named 'blender'
uniform float time;
// bloom_shader
uniform sampler2D blender; // <- shared source should be named 'blender'
uniform sampler2D depth;
// p5 setup
let layer
let effects[] // user space array of shaders

function setup() {
  createCanvas(600, 400, WEBGL)
  layer = createFramebuffer()
  // instantiate shaders with keys for later
  // uniform settings and add them to effects
  effects.push(makeShader(noise_shader, 'noise'))
  effects.push(makeShader(bloom_shader, 'bloom'))
}
// p5 draw
function draw() {
  layer.begin()
  // render scene into layer
  layer.end()
  // render target by applying effects to layer
  let uniforms = { // emit uniforms to shaders (besides uniformsUI)
    bloom: { depth: layer.depth }, // <- use bloom key
    noise: { time: millis() / 1000 } // <- use noise key
  }
  const target = applyEffects(layer, effects, uniforms)
  // display target using screen space coords
  beginHUD()
  image(target, 0, 0)
  endHUD()
}
// p5 keyPressed
function keyPressed() {
  // swap effects
  [effects[0], effects[1]] = [effects[1], effects[0]]
}
  1. applyEffects(source, effects, [uniforms = {}], [flip = true]): Sequentially applies all effects (in the order they were added) to the source, which can be a p5.Framebuffer, p5.Graphics, p5.Image, or video p5.MediaElement. The uniforms param maps shader keys to their respective uniform values, formatted as { uniform_1_name: value_1, ..., uniform_n_name: value_n }, provided that a sampler2D uniform blender variable is declared in each shader effect as a common fbo layer. The flip boolean indicates whether the final image should be vertically flipped. This method processes each effect, applying its shader with the corresponding uniforms (using applyShader), and returns the final processed source, now modified by all effects.
  2. createBlender(effects, [options={}]): Creates and attaches an fbo layer with specified options to each shader in the effects array. If createBlender is not called, applyEffects automatically generates a blender layer for each shader, utilizing default options.
  3. removeBlender(effects): Removes the individual fbo layers associated with each shader in the effects array, freeing up resources by invoking remove.

Macros

Retrieve image offset, mouse position, pointer position and screen resolution which are common uniform vec2 variables

  1. texOffset(image) which is the same as: return [1 / image.width, 1 / image.height].
  2. mousePosition([flip = true]) which is the same as: return [this.pixelDensity() * this.mouseX, this.pixelDensity() * (flip ? this.height - this.mouseY : this.mouseY)].
  3. pointerPosition(pointerX, pointerY, [flip = true]) which is the same as: return [this.pixelDensity() * pointerX, this.pixelDensity() * (flip ? this.height - pointerY : pointerY)]. Available to both, the p5 object and p5.RendererGL instances. Note that pointerX should always be the first parameter and pointerY the second.
  4. resolution() which is the same as: return [this.pixelDensity() * this.width, this.pixelDensity() * this.height]. Available to both, the p5 object and p5.RendererGL instances.

Space transformations

This section delves into matrix manipulations and queries which are essential for 3D rendering. It includes functions for matrix operations like creation, inversion, and multiplication in the Matrix operations subsection, and offers methods to retrieve transformation matrices and perform space conversions in Matrix queries, Frustum queries, and Coordinate Space conversions, facilitating detailed control over 3D scene transformations.

Have a look at the blur with focal point, post-effects, and visualizing perspective transformation to NDC examples.

Matrix operations

  1. iMatrix(): Returns the identity matrix.
  2. tMatrix(matrix): Returns the tranpose of matrix.
  3. invMatrix(matrix): Returns the inverse of matrix.
  4. axbMatrix(a, b): Returns the product of the a and b matrices.

Observation: All returned matrices are instances of p5.Matrix.

Matrix queries

  1. pMatrix(): Returns the current projection matrix.
  2. mvMatrix([{[vMatrix], [mMatrix]}]): Returns the modelview matrix.
  3. mMatrix(): Returns the model matrix. This matrix defines a local space transformation according to translate, rotate and scale commands. Refer also to push and pop.
  4. eMatrix(): Returns the current eye matrix (the inverse of vMatrix()). In addition to p5 and p5.RendererGL instances, this method is also available to p5.Camera objects.
  5. vMatrix(): Returns the view matrix (the inverse of eMatrix()). In addition to p5 and p5.RendererGL instances, this method is also available to p5.Camera objects.
  6. pvMatrix([{[pMatrix], [vMatrix]}]): Returns the projection times view matrix.
  7. pvInvMatrix([{[pMatrix], [vMatrix], [pvMatrix]}]): Returns the pvMatrix inverse.
  8. lMatrix([{[from = iMatrix()], [to = this.eMatrix()]}]): Returns the 4x4 matrix that transforms locations (points) from matrix from to matrix to.
  9. dMatrix([{[from = iMatrix()], [to = this.eMatrix()]}]): Returns the 3x3 matrix (only rotational part is needed) that transforms directions (vectors) from matrix from to matrix to. The nMatrix below is a special case of this one.
  10. nMatrix([{[vMatrix], [mMatrix], [mvMatrix]}]): Returns the normal matrix.

Observations

  1. All returned matrices are instances of p5.Matrix.
  2. The pMatrix, vMatrix, pvMatrix, eMatrix, mMatrix and mvMatrix default values are those defined by the renderer at the moment the query is issued.

Frustum queries

  1. lPlane(): Returns the left clipping plane.
  2. rPlane(): Returns the right clipping plane.
  3. bPlane(): Returns the bottom clipping plane.
  4. tPlane(): Returns the top clipping plane.
  5. nPlane(): Returns the near clipping plane.
  6. fPlane(): Returns the far clipping plane.
  7. fov(): Returns the vertical field-of-view (fov) in radians.
  8. hfov(): Returns the horizontal field-of-view (hfov) in radians.
  9. isOrtho(): Returns the camera projection type: true for orthographic and false for perspective.

Coordinate space conversions

  1. parsePosition(vector = Tree.ORIGIN, [{[from = Tree.EYE], [to = Tree.WORLD], [pMatrix], [vMatrix], [eMatrix], [pvMatrix], [pvInvMatrix]}]): transforms locations (points) from matrix from to matrix to.
  2. parseDirection(vector = Tree._k, [{[from = Tree.EYE], [to = Tree.WORLD], [vMatrix], [eMatrix], [pMatrix]}]): transforms directions (vectors) from matrix from to matrix to.

Pass matrix params when you cached those matrices (see the previous section), either to speedup computations, e.g.,

let pvInv

function draw() {
  // cache pvInv at the beginning of the rendering loop
  // note that this matrix rarely change within the iteration
  pvInv = pvInvMatrix()
  // ...
  // speedup parsePosition
  parsePosition(vector, { from: Tree.WORLD, to: Tree.SCREEN, pvInvMatrix: pvInv })
  parsePosition(vector, { from: Tree.WORLD, to: Tree.SCREEN, pvInvMatrix: pvInv })
  // ... many more parsePosition calls....
  // ... all the above parsePosition calls used the (only computed once) cached pvInv matrix
}

or to transform points (and vectors) between local spaces, e.g.,

let model

function draw() {
  // ...
  // save model matrix as it is set just before drawing your model
  model = mMatrix()
  drawModel()
  // continue drawing your tree...
  // let's draw a bulls eye at the model origin screen projection
  push()
  let screenProjection = parsePosition(Tree.ORIGIN, { from: model, to: Tree.SCREEN })
  // which is the same as:
  // let screenProjection = parsePosition(createVector(0, 0, 0), { from: model, to: Tree.SCREEN });
  // or,
  // let screenProjection = parsePosition([0, 0, 0], { from: model, to: Tree.SCREEN });
  // or, more simply:
  // let screenProjection = parsePosition({ from: model, to: Tree.SCREEN });
  bullsEye({ x: screenProjection.x, y: screenProjection.y })
  pop()
}

Observations

  1. Returned transformed vectors are instances of p5.Vector.
  2. from and to may also be specified as either: Tree.WORLD, Tree.EYE, Tree.SCREEN, Tree.NDC or Tree.MODEL.
  3. When no matrix params (eMatrix, pMatrix,...) are passed the renderer current values are used instead.
  4. The default parsePosition call (i.e., parsePosition(Tree.ORIGIN, {from: Tree.EYE, to: Tree.WORLD)) returns the camera world position.
  5. Note that the default parseDirection call (i.e., parseDirection(Tree._k, {from: Tree.EYE, to: Tree.WORLD)) returns the normalized camera viewing direction.
  6. Other useful vector constants, different than Tree.ORIGIN (i.e., [0, 0, 0]) and Tree._k (i.e., [0, 0, -1]), are: Tree.i (i.e., [1, 0, 0]), Tree.j (i.e., [0, 1, 0]), Tree.k (i.e., [0, 0, 1]), Tree._i (i.e., [-1, 0, 0]) and Tree._j (i.e., [0, -1, 0]).

Heads Up Display

  1. beginHUD(): Begins Heads Up Display, so that geometry specified between beginHUD() and endHUD() is defined in window space. Should always be used in conjunction with endHUD.
  2. endHUD(): Ends Heads Up Display, so that geometry specified between beginHUD() and endHUD() is defined in window space. Should always be used in conjunction with beginHUD.

Utilities

This section comprises a collection of handy functions designed to facilitate common tasks in 3D graphics, such as pixel ratio calculations, mouse picking and visibility determination.

  1. pixelRatio(location): Returns the world to pixel ratio units at given world location, i.e., a line of n * pixelRatio(location) world units will be projected with a length of n pixels on screen.
  2. mousePicking([{[mMatrix = this.mMatrix()], [x], [y], [size = 50], [shape = Tree.CIRCLE], [eMatrix], [pMatrix], [vMatrix], [pvMatrix]}]): same as return this.pointerPicking(this.mouseX, this.mouseY, { mMatrix: mMatrix, x: x, y: y, size: size, shape: shape, eMatrix: eMatrix, pMatrix: pMatrix, vMatrix: vMatrix, pvMatrix: pvMatrix }) (see below).
  3. pointerPicking(pointerX, pointerY, [{[mMatrix = this.mMatrix()], [x], [y], [size = 50], [shape = Tree.CIRCLE], [eMatrix], [pMatrix], [vMatrix], [pvMatrix]}]): Returns true if pointerX, pointerY lies within the screen space circle centered at (x, y) and having size diameter. Pass mMatrix to compute (x, y) as the screen space projection of the local space origin (defined by mMatrix), having size as its bounding sphere diameter. Use Tree.SQUARE to use a squared shape instead of a circled one. Note that pointerX should always be specified before pointerY.
  4. visibility([{[center], [radius], [corner1], [corner2], [bounds]}]): Returns object visibility, either as Tree.VISIBLE, Tree.INVISIBLE, or Tree.SEMIVISIBLE. Object may be either a point: visibility({ center, [bounds = this.bounds([{[eMatrix = this.eMatrix()], [vMatrix = this.vMatrix()]}])]}), a ball: visibility({ center, radius, [bounds = this.bounds()]}) or an axis-aligned box: visibility({ corner1, corner2, [bounds = this.bounds()]}).
  5. bounds([{[eMatrix], [vMatrix]}]): Returns the general form of the current frustum six plane equations, i.e., ax + by + cz + d = 0, formatted as an object literal having keys: Tree.LEFT, Tree.RIGHT, Tree.BOTTOM, Tree.TOP, Tree.NEAR and Tree.FAR, e.g., access the near plane coefficients as: 
    let bounds = bounds()
let near = bounds[Tree.NEAR] // near.a, near.b, near.c and near.d

Drawing stuff

This section includes a range of functions designed for visualizing various graphical elements in 3D space, such as axes, grids, bullseyes, and view frustums. These tools are essential for debugging, illustrating spatial relationships, and enhancing the visual comprehension of 3D scenes. Have a look at the toon shading, blur with focal point, post-effects, and visualizing perspective transformation to NDC examples.

  1. parseGeometry(fn, ...args): Captures geometry by running fn, which should be passed as first parameter, with args, then returns a p5.Geometry object. Applies colors from args if specified, or clears them, and computes normals.
  2. axes([{ [size = 100], [colors = ['Red', 'Lime', 'DodgerBlue']], [bits = Tree.LABELS | Tree.X | Tree.Y | Tree.Z] }]): Draws axes with given size in world units, colors, and bitwise mask that may be composed of Tree.X, Tree._X, Tree.Y, Tree._Y, Tree.Z, Tree._Z and Tree.LABELS bits.
  3. grid([{ [size = 100], [subdivisions = 10], [style = Tree.DOTS] }]): Draws grid with given size in world units, subdivisions and dotted (Tree.DOTS) or solid (Tree.SOLID) lines.
  4. cross([{ [mMatrix = this.mMatrix()], [x], [y], [size = 50], [eMatrix], [pMatrix], [vMatrix], [pvMatrix] }]): Draws a cross at x, y screen coordinates with given size in pixels. Pass mMatrix to compute (x, y) as the screen space projection of the local space origin (defined by mMatrix).
  5. bullsEye([{ [mMatrix = this.mMatrix()], [x], [y], [size = 50], [shape = Tree.CIRCLE], [eMatrix], [pMatrix], [vMatrix], [pvMatrix] }]): Draws a circled bullseye (use Tree.SQUARE to draw it as a square) at x, y screen coordinates with given size in pixels. Pass mMatrix to compute (x, y) as the screen space projection of the local space origin (defined by mMatrix).
  6. viewFrustum([{ [pg], [bits = Tree.NEAR | Tree.FAR], [viewer = () => this.axes({ size: 50, bits: Tree.X | Tree._X | Tree.Y | Tree._Y | Tree.Z | Tree._Z })], [eMatrix = pg?.eMatrix()], [pMatrix = pg?.pMatrix()], [vMatrix = this.vMatrix()] }]): Draws a view frustum based on the specified bitwise mask bits Tree.NEAR, Tree.FAR, Tree.APEX, Tree.BODY, and viewer callback visual representation. The function determines the view frustum's position, orientation, and viewing volume either from a given pg, or directly through eMatrix and pMatrix parameters.

Installation

Link the p5.treegl.js library into your HTML file, after you have linked in p5.js. For example:

<!doctype html>
<html>
<head>
  <script src="p5.js"></script>
  <script src="p5.sound.js"></script>
  <script src=https://cdn.jsdelivr.net/gh/VisualComputing/p5.treegl/p5.treegl.js></script>
  <script src="sketch.js"></script>
</head>
<body>
</body>
</html>

to include its minified version use:

<script src=https://cdn.jsdelivr.net/gh/VisualComputing/p5.treegl/p5.treegl.min.js></script>

instead.

vs-code & vs-codium & gitpod hacking instructions

Clone the repo (git clone https://github.com/VisualComputing/p5.treegl) and open it with your favorite editor.

Don't forget to check these p5.js references:

  1. Library creation.
  2. Software architecture.
  3. Webgl mode.

Footnotes

  1. For an in-depth review, please refer to the post-effects study conducted by Diego Bulla. ↩