Read/Write bytes of float in JS

问题:

Is there any way I can read bytes of a float value in JS? What I need is to write a raw FLOAT or DOUBLE value into some binary format I need to make, so is there any way to get a byte-by-byte IEEE 754 representation? And same question for writing of course.

回答1:

Would this snippet help?

@Kevin Gadd:

var parser = new BinaryParser
  ,forty = parser.encodeFloat(40.0,2,8) 
  ,twenty = parser.encodeFloat(20.0,2,8);  
console.log(parser.decodeFloat(forty,2,8).toFixed(1));   //=> 40.0
console.log(parser.decodeFloat(twenty,2,8).toFixed(1));  //=> 20.0

回答2:

You can do it with typed arrays:

var buffer = new ArrayBuffer(4);
var intView = new Int32Array(buffer);
var floatView = new Float32Array(buffer);

floatView[0] = Math.PI
console.log(intView[0].toString(2)); //bits of the 32 bit float

Or another way:

var view = new DataView(new ArrayBuffer(4));
view.setFloat32(0, Math.PI);
console.log(view.getInt32(0).toString(2)); //bits of the 32 bit float

Not sure what browser support is like though

回答3:

I've created an expansion of Milos' solution that should be a bit faster, assuming TypedArrays are not an option of course (in my case I'm working with an environment where they're not available):

function Bytes2Float32(bytes) {
    var sign = (bytes & 0x80000000) ? -1 : 1;
    var exponent = ((bytes >> 23) & 0xFF) - 127;
    var significand = (bytes & ~(-1 << 23));

    if (exponent == 128) 
        return sign * ((significand) ? Number.NaN : Number.POSITIVE_INFINITY);

    if (exponent == -127) {
        if (significand == 0) return sign * 0.0;
        exponent = -126;
        significand /= (1 << 22);
    } else significand = (significand | (1 << 23)) / (1 << 23);

    return sign * significand * Math.pow(2, exponent);
}

Given an integer containing 4 bytes holding an IEEE-754 32-bit single precision float, this will produce the (roughly) correct Javascript number value without using any loops.

回答4:

Koolinc's snippet is good if you need a solution that powerful, but if you need it for limited use you are better off writing your own code. I wrote the following function for converting a string hex representation of bytes to a float:

function decodeFloat(data) {
    var binary = parseInt(data, 16).toString(2);
    if (binary.length < 32) 
        binary = ('00000000000000000000000000000000'+binary).substr(binary.length);
    var sign = (binary.charAt(0) == '1')?-1:1;
    var exponent = parseInt(binary.substr(1, 8), 2) - 127;
    var significandBase = binary.substr(9);
    var significandBin = '1'+significandBase;
    var i = 0;
    var val = 1;
    var significand = 0;

    if (exponent == -127) {
        if (significandBase.indexOf('1') == -1)
            return 0;
        else {
            exponent = -126;
            significandBin = '0'+significandBase;
        }
    }

    while (i < significandBin.length) {
        significand += val * parseInt(significandBin.charAt(i));
        val = val / 2;
        i++;
    }

    return sign * significand * Math.pow(2, exponent);
}

There are detailed explanations of algorithms used to convert in both directions for all formats of floating points on wikipedia, and it is easy to use those to write your own code. Converting from a number to bytes should be more difficult because you need to normalize the number first.

回答5:

I had a similar problem, I wanted to convert any javascript number to a Buffer and then parse it back without stringifying it.

function numberToBuffer(num) {
  const buf = new Buffer(8)
  buf.writeDoubleLE(num, 0)
  return buf
}

Use example:

// convert a number to buffer
const buf = numberToBuffer(3.14)
// and then from a Buffer
buf.readDoubleLE(0) === 3.14

This works on current Node LTS (4.3.1) and up. didn't test in lower versions.

回答6:

I expect you could figure it out (blech), but I assume you're asking if there's something built-in. Not as far as I've ever heard; see sections 8.5 and 15.7 of the spec.