this PR implements new parry crate (parry-i32f32) with fixed point numbers, as a reference fixed point number with 32 bits for integer part and 32 bits for decimal part was used (FixedI32F32) from simba crate.
This PR is separated into few commits with different meanings:
Add fixed point support
This commit adds real! macro for constructing f32, f64 and FixedI32F32 literals, this is used for all literals throughout the crate, this is needed as Rust does not have custom literals that can be used for fixed point numbers
Use feature flags properly, so rust can find functions for fixed point numbers
Tries to resolve few ambiguities, warnings and missing includes between implementations for FixedI32F32 and floats
Use conversion functions for converting reals
There were some as conversions between floats and integer numbers which do not work for FixedI32F32 and lowest common denominator needs to be used that works both for floats and FixedI32F32, so I used conversion functions
as operator uses narrowing conversion, I replaced it with functions that use checked conversions which panic if source cannot be converted to destination, is this ok? I have not found any problems with this yet
This is not pretty and I would like suggestions how to do this one better
There are still some tests that do not pass, as fixed point is not as accurate as floats, in parry I found undocumented improved_fixed_point_support cfg feature which when enabled fixes some of these tests, so I left it on, but these tests still fail:
test mass_properties::mass_properties::test::mass_properties_add_partial_zero
test mass_properties::mass_properties::test::mass_properties_add_sub
Could anyone point me in the right direction so I can fix these tests and what is the preferred solution for f32/f64/FixedI32F32 <-> Integer conversions?
Hello,
this PR implements new parry crate (
parry-i32f32
) with fixed point numbers, as a reference fixed point number with 32 bits for integer part and 32 bits for decimal part was used (FixedI32F32
) from simba crate.There was some missing functionality in simba, that was needed and is implemented in separate PR in simba repository here https://github.com/dimforge/simba/pull/48
This PR is separated into few commits with different meanings:
Add fixed point support
real!
macro for constructing f32, f64 and FixedI32F32 literals, this is used for all literals throughout the crate, this is needed as Rust does not have custom literals that can be used for fixed point numbersUse feature flags properly, so rust can find functions for fixed point numbers
Use conversion functions for converting reals
as
conversions between floats and integer numbers which do not work for FixedI32F32 and lowest common denominator needs to be used that works both for floats and FixedI32F32, so I used conversion functionsas
operator uses narrowing conversion, I replaced it with functions that use checked conversions which panic if source cannot be converted to destination, is this ok? I have not found any problems with this yetThere are still some tests that do not pass, as fixed point is not as accurate as floats, in parry I found undocumented
improved_fixed_point_support
cfg feature which when enabled fixes some of these tests, so I left it on, but these tests still fail:test mass_properties::mass_properties::test::mass_properties_add_partial_zero
test mass_properties::mass_properties::test::mass_properties_add_sub
Could anyone point me in the right direction so I can fix these tests and what is the preferred solution for f32/f64/FixedI32F32 <-> Integer conversions?