TS mirror of the shader polynomial (same coefficients, same clamp),
evaluated in float64. Use erfRef when accuracy matters; use this to
predict the GPU result to within accumulated float32 rounding
(measured ≤ ~7e-6 — see the coefficient docblock, and the ~5e-6
worst case observed against a real GPU).
TS mirror of the shader polynomial (same coefficients, same clamp), evaluated in float64. Use
erfRefwhen accuracy matters; use this to predict the GPU result to within accumulated float32 rounding (measured ≤ ~7e-6 — see the coefficient docblock, and the ~5e-6 worst case observed against a real GPU).