glsl-rgba-to-float

Installation

npm install --save glsl-rgba-to-float

Signature

float rgbaToFloat(vec4 texelRGBA, bool littleEndian);

Usage

This package exports a GLSL function, rgbaToFloat, that decodes 32-bit floating point values from the RGBA channels of a WebGL texture. This is a workaround for the lack of support in baseline WebGL 1.0 for floating-point textures. WebGL 2.0 allows floating-point textures, and WebGL 1.0 allows them when the extension OES_texture_float is available, but some users may be using a browser or hardware that don't have these capabilities. This package seeks to fill that gap.

In baseline WebGL 1.0, a maximum of 32 bits are allotted for each texel (pixel in a texture). It's therefore possible to create textures in which each texel represents a single 32-bit float. A straightforward way to do that is to create a Float32Array with the desired texel values, then reinterpret the underlying binary data of the typed array's ArrayBuffer as a Uint8Array. This latter array can then be passed to the WebGL system as an RGBA texture of type UNSIGNED_BYTE. This "tricks" WebGL into treating a single 32-bit float as a vector of four bytes.

// JavaScript control code

// construct the pixel data
const floatArray = new Float32Array([ /* pixels */ ]);
const uintArray = new Uint8Array(floatArray.buffer);

// bind the pixel data to a WebGL texture
gl.texImage2D(
  gl.TEXTURE_2D,    // target
  0,                // mipmap level
  gl.RGBA,          // internal format: 4 color channels
  width,            // width of texture in pixels
  height,           // height of texture in pixels
  0,                // border: must be 0
  gl.RGBA,          // format: must be the same as internal format
  gl.UNSIGNED_BYTE, // type: 8 bits per channel
  uintArray,        // pixels
);

rgbaToFloat is designed to be consumed as a GLSLify module (though other methods may be used to copy the source code into your shader). To decode a texel in your shader code, pass the vec4 representing the texel to the function rgbaToFloat. Note that because the function has to decode the float from its constituent bits, it needs to know whether the machine on which it's running uses big-endian or little-endian representation.

// fragment shader

#pragma glslify: rgbaToFloat = require('glsl-rgba-to-float')

// does the machine use little-endian representation?
uniform bool littleEndian;

// the texture
uniform sampler2D texture;

// texture coordinates, passed from the vertex shader
varying vec2 textureCoords;

void main() {
  vec4 texelRGBA = texture2D(texture, textureCoords);
  float texelFloat = rgbaToFloat(texelRGBA, littleEndian);
  // use the decoded float for something wonderful...
}

Here's a simple way to compute endianness in JavaScript. Pass the result as a uniform variable to your shader.

const littleEndian = (function machineIsLittleEndian() {
  const uint8Array = new Uint8Array([0xAA, 0xBB]);
  const uint16array = new Uint16Array(uint8Array.buffer);
  return uint16array[0] === 0xBBAA;
})();