`fp64wgsl` is a WebGPU shader library for high-precision, 64-bit floating point math. This library has been verified against several high-precision floating point libraries and is tested to be correct within 0.5 ULP.

For support on this library, please reach out to the author via email at sienna at linux dot com. Pull requests not accepted.

It contains these supported functions:

* fp64_abs(a)
* fp64_acos(x)
* fp64_add(a, b)
* fp64_asin(x)
* fp64_atan(x)
* fp64_atan2(y, x)
* fp64_cos(x)
* fp64_div(a, b)
* fp64_eq(a, b)
* fp64_from_f32(v)
* fp64_from_i32(v)
* fp64_ge(a, b)
* fp64_gt(a, b)
* fp64_inf(sign)
* fp64_is_inf(v)
* fp64_is_nan(v)
* fp64_is_zero(v)
* fp64_le(a, b)
* fp64_lt(a, b)
* fp64_mantissa(v)
* fp64_mul(a, b)
* fp64_nan()
* fp64_neg(a)
* fp64_neg_zero()
* fp64_round_to_i32(v)
* fp64_sin(x)
* fp64_sqrt(a)
* fp64_sub(a, b)
* fp64_tan(x)
* fp64_zero()

Other functions are considered to be internal their usage is not advised or supported.

Performance is roughly 30% slower than native f32 for addition, subtraction, and multiplication. For division and square root, performance is about 13× slower than native f32. For trigonometric functions, performance is about 100–500× slower than native f32.

To reference this, use this reference:

@software{satterwhite_fp64wgsl_2026,
  author       = {Sienna Meridian Satterwhite},
  title        = {siennathesane/fp64wgsl: v1.0.0},
  month        = aug,
  year         = 2026,
  publisher    = {Zenodo},
  version      = {v1.0.0},
  doi          = {10.5281/zenodo.22018377},
  url          = {https://doi.org/10.5281/zenodo.22018377},
}