nearly

Approximate float equality, comparing computed values without exact-match fragility.

Two floats that should be equal rarely are once they’ve been through any arithmetic, so a == b is a coin flip on the GPU. nearly checks that they land within a small tolerance instead. It’s the comparison you almost always want.

bool nearly(in float a, in float b)             { return abs(a - b) < 0.001; }
bool nearly(in float a, in float b, in float k) { return abs(a - b) < k; }

The two-argument form uses a sensible default epsilon; the three-argument form lets you set the tolerance k yourself.

Usage

float cell = floor(uv.x * 10.0);
if (nearly(cell, 3.0)) {            // this column, within 0.001
    col = highlight;
}

if (nearly(d, 0.0, 0.01)) {        // "close enough to the surface"
    col += glow;
}

Notes

Pick k relative to the scale of what you’re comparing: 0.001 is fine for small indices, too tight for large coordinates.
It’s still a branch: on values that vary per pixel, prefer a smooth falloff (smoothstep, _pulse) when you can, and save nearly for genuinely discrete checks like cell indices or state flags.