mirror of
https://github.com/visgl/luma.gl.git
synced 2025-12-08 17:36:19 +00:00
16 lines
758 B
Markdown
16 lines
758 B
Markdown
# 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.
|