float16

IEEE 754-2008 half-precision (Float16) and BFloat16 arithmetic library for Go.

Part of the Zerfoo ML ecosystem.

Features

• Full IEEE 754-2008 compliance for 16-bit floating-point arithmetic
• BFloat16 support — Google Brain format for ML training and inference
• Special value handling — ±0, ±Inf, NaN (with payload), normalized and subnormal numbers
• Multiple rounding modes — nearest-even, toward zero, toward ±Inf, nearest-away
• Vectorized operations — batch add, multiply, and dot product
• Fast math mode — optional lookup-table acceleration for performance-critical paths
• Zero dependencies — pure Go, no CGo

Installation

go get github.com/zerfoo/float16

Requires Go 1.26+.

Quick Start

package main

import (
    "fmt"
    "github.com/zerfoo/float16"
)

func main() {
    a := float16.FromFloat32(3.14159)
    b := float16.FromFloat32(2.71828)

    sum := a.Add(b)
    product := a.Mul(b)

    fmt.Printf("Sum: %f\n", sum.ToFloat32())
    fmt.Printf("Product: %f\n", product.ToFloat32())

    // Special values
    inf := float16.Inf(1)
    fmt.Printf("Inf: %v, IsInf: %v\n", inf, inf.IsInf(0))
}

Conversion

// From float32/float64
f16 := float16.FromFloat32(3.14)
f16 := float16.FromFloat64(2.718)

// From bit representation
f16 := float16.FromBits(0x4200) // 3.0

// Back to native types
f32 := f16.ToFloat32()
f64 := f16.ToFloat64()

Rounding Modes

config := float16.GetConfig()
config.DefaultRoundingMode = float16.RoundTowardZero
float16.Configure(config)

// RoundNearestEven (default), RoundTowardZero, RoundTowardPositive,
// RoundTowardNegative, RoundNearestAway

Range and Precision

Property	Value
Range	±65,504
Precision	~3-4 decimal digits
Smallest normal	~6.10 × 10⁻⁵
Smallest subnormal	~5.96 × 10⁻⁸
Machine epsilon	~9.77 × 10⁻⁴

Used By

• ztensor — GPU-accelerated tensor library

License

Apache 2.0