luma.gl/docs/api-reference/shadertools/modules/fp32.md

32-bit Floating Point (Shader Module)

Provides "improved" 32-bit math support to GPU shaders on certain platforms, mainly compensating for non-IEEE compliant "fast-math" functions that GPU vendors use to trade accuracy for speed.

Function	Description
vec2 tan_fp32(vec2 a)	Improved accuracy tan - falls back to tan on non-Intel platforms

Performance Implications

Since 32-bit floating point trigonometric is using more accurate Taylor approximations, they cost more GPU cycles than built-in versions. If you need the precision, it is typically worth the cost.