Age | Commit message (Collapse) | Author |
|
|
|
|
|
Issue: #212
|
|
Issue: #213
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
We could lose precision when multiplying by integral_max with floats.
For doubles, we can count on up to 2^56 - 1 to be expressible somewhat
exactly.
|
|
|
|
|
|
|
|
|
|
|
|
Float mantissa has 23 bits. 9 bits get wasted per value. Instead, use
uint16_t that doesn't lose much precision, but saves half the space.
|
|
|
|
|
|
Higher step value allows more precision when spline doesn't match the
control points.
|
|
|
|
Also, change value_type to more sensible value
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Issue: #118
Linux and OSX likely broken for now.
|
|
|
|
Issue: #204
|