CODEWITHSUNDEEP
webgl
Shukant Pal
Shukant Pal

Reputation: 730

WebGL rendering outside of browser paint time

We are building a WebGL application that has some high render-load objects. Is there a way we can render those object outside of browser-paint time, i.e. in the background? We don't want our FPS going down, and breaking up our rendering process is possible (to split between frames).

Upvotes: 1

Views: 593

Answers (1)

user128511
user128511

Reputation:

Three ideas come to mind.

  1. You can render to a texture via a framebuffer over many frames, when you're done you render that texture to the canvas.

const gl = document.querySelector('canvas').getContext('webgl');
const vs = `
attribute vec4 position;
attribute vec2 texcoord;
varying vec2 v_texcoord;
void main() {
  gl_Position = position;
  v_texcoord = texcoord;
}
`;
const fs = `
precision highp float;
uniform sampler2D tex;
varying vec2 v_texcoord;
void main() {
  gl_FragColor = texture2D(tex, v_texcoord);
}
`;

// compile shader, link program, look up locations
const programInfo = twgl.createProgramInfo(gl, [vs, fs]);

// gl.createBuffer, gl.bindBuffer, gl.bufferData
const bufferInfo = twgl.createBufferInfoFromArrays(gl, {
  position: {
    numComponents: 2,
    data: [
      -1, -1,
       1, -1,
      -1,  1,
      -1,  1,
       1, -1,
       1,  1,
    ],
  },
  texcoord: {
    numComponents: 2,
    data: [
       0,  0,
       1,  0,
       0,  1,
       0,  1,
       1,  0,
       1,  1,
    ],  
  },
});

// create a framebuffer with a texture and depth buffer
// same size as canvas
// gl.createTexture, gl.texImage2D, gl.createFramebuffer
// gl.framebufferTexture2D
const framebufferInfo = twgl.createFramebufferInfo(gl);

const infoElem = document.querySelector('#info');

const numDrawSteps = 16;
let drawStep = 0;
let time = 0;

// draw over several frames. Return true when ready
function draw() {
  // draw to texture
  // gl.bindFrambuffer, gl.viewport
  twgl.bindFramebufferInfo(gl, framebufferInfo);
  
  if (drawStep == 0) {
    // on the first step clear and record time
    gl.disable(gl.SCISSOR_TEST);
    gl.clearColor(0, 0, 0, 0);
    gl.clear(gl.COLOR_BUFFER_BIT  | gl.DEPTH_BUFFER_BIT);
    time = performance.now() * 0.001;
  }
  

  // this represents drawing something. 
  gl.enable(gl.SCISSOR_TEST);
  
  const halfWidth = framebufferInfo.width / 2;
  const halfHeight = framebufferInfo.height / 2;
  
  const a = time * 0.1 + drawStep
  const x = Math.cos(a      ) * halfWidth + halfWidth;
  const y = Math.sin(a * 1.3) * halfHeight + halfHeight;

  gl.scissor(x, y, 16, 16);
  gl.clearColor(
     drawStep / 16,
     drawStep / 6 % 1,
     drawStep / 3 % 1,
     1);
  gl.clear(gl.COLOR_BUFFER_BIT);
  
  drawStep = (drawStep + 1) % numDrawSteps;
  return drawStep === 0;
}

let frameCount = 0;
function render() {
  ++frameCount;
  infoElem.textContent = frameCount;
  
  if (draw()) {
    // draw to canvas
    // gl.bindFramebuffer, gl.viewport
    twgl.bindFramebufferInfo(gl, null);
    
    gl.disable(gl.DEPTH_TEST);
    gl.disable(gl.BLEND);
    gl.disable(gl.SCISSOR_TEST);
    
    gl.useProgram(programInfo.program);
    
    // gl.bindBuffer, gl.enableVertexAttribArray, gl.vertexAttribPointer
    twgl.setBuffersAndAttributes(gl, programInfo, bufferInfo);
    
    // gl.uniform...
    twgl.setUniformsAndBindTextures(programInfo, {
      tex: framebufferInfo.attachments[0],
    });
    
    // draw the quad
    gl.drawArrays(gl.TRIANGLES, 0, 6);
  }
  
  requestAnimationFrame(render);
}
requestAnimationFrame(render);
<canvas></canvas>
<div id="info"></div>
<script src="https://twgljs.org/dist/4.x/twgl.min.js"></script>

  1. You can make 2 canvases. A webgl canvas that is not in the DOM. You render to it over many frames and when you're done you draw it to a 2D canvas with ctx.drawImage(webglCanvas, ...) This is basically the same as #1 except you're letting the browser "render that texture to a canvas" part

const ctx = document.querySelector('canvas').getContext('2d');
const gl = document.createElement('canvas').getContext('webgl');
const vs = `
attribute vec4 position;
attribute vec2 texcoord;
varying vec2 v_texcoord;
void main() {
  gl_Position = position;
  v_texcoord = texcoord;
}
`;
const fs = `
precision highp float;
uniform sampler2D tex;
varying vec2 v_texcoord;
void main() {
  gl_FragColor = texture2D(tex, v_texcoord);
}
`;

// compile shader, link program, look up locations
const programInfo = twgl.createProgramInfo(gl, [vs, fs]);

const infoElem = document.querySelector('#info');

const numDrawSteps = 16;
let drawStep = 0;
let time = 0;

// draw over several frames. Return true when ready
function draw() {  
  if (drawStep == 0) {
    // on the first step clear and record time
    gl.disable(gl.SCISSOR_TEST);
    gl.clearColor(0, 0, 0, 0);
    gl.clear(gl.COLOR_BUFFER_BIT  | gl.DEPTH_BUFFER_BIT);
    time = performance.now() * 0.001;
  }
  

  // this represents drawing something. 
  gl.enable(gl.SCISSOR_TEST);
  
  const halfWidth = gl.canvas.width / 2;
  const halfHeight = gl.canvas.height / 2;
  
  const a = time * 0.1 + drawStep
  const x = Math.cos(a      ) * halfWidth + halfWidth;
  const y = Math.sin(a * 1.3) * halfHeight + halfHeight;

  gl.scissor(x, y, 16, 16);
  gl.clearColor(
     drawStep / 16,
     drawStep / 6 % 1,
     drawStep / 3 % 1,
     1);
  gl.clear(gl.COLOR_BUFFER_BIT);
  
  drawStep = (drawStep + 1) % numDrawSteps;
  return drawStep === 0;
}

let frameCount = 0;
function render() {
  ++frameCount;
  infoElem.textContent = frameCount;
  
  if (draw()) {
    // draw to canvas
    ctx.clearRect(0, 0, ctx.canvas.width, ctx.canvas.height);
    ctx.drawImage(gl.canvas, 0, 0);
  }
  
  requestAnimationFrame(render);
}
requestAnimationFrame(render);
<canvas></canvas>
<div id="info"></div>
<script src="https://twgljs.org/dist/4.x/twgl.min.js"></script>

  1. You can use OffscreenCanvas and render in a worker. This has only shipped in Chrome though.

Note that if you DOS the GPU (give the GPU too much work) you can still affect the responsiveness of the main thread because most GPUs do not support pre-emptive multitasking. So, if you have a lot of really heavy work then split it up into smaller tasks.

As an example if you took one of the heaviest shaders from shadertoy.com that runs at say 0.5 fps when rendered at 1920x1080, even offscreen it will force the entire machine to run at 0.5 fps. To fix you'd need to render smaller portions over several frames. If it's running at 0.5 fps that suggests you need to split it up into at least 120 smaller parts, maybe more, to keep the main thread responsive and at 120 smaller parts you'd only see the results every 2 seconds.

In fact trying it out shows some issues. Here's Iq's Happy Jumping Example drawn over 960 frames. It still can't keep 60fps on my late 2018 Macbook Air even though it's rendering only 2160 pixels a frame (2 columns of a 1920x1080 canvas). The issue is likely some parts of the scene have to recurse deeply and there is no way knowing before hand which parts of the scene that will be. One reason why shadertoy style shaders using signed distance fields are more of a toy (hence shaderTOY) and not actually a production style technique.

Anyway, the point of that is if you give the GPU too much work you'll still get an unresponsive machine.

Upvotes: 2

Related Questions