Skip to content

benjamin-cates/astro_nalgebra

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

astro_float for nalgebra

This package implements the traits required to use the linear algebra library nalgebra on the type BigFloat in astro_float. This library exports it's own type [BigFloat<CTX>] where CTX is a computational context that stores precision and rounding mode.

Examples

Compile-time const precision

Using this library with a precision and rounding mode that is known at compile-time is straight forward using the [ConstCtx] computational context. It is recommended to store BigFloat<ConstCtx<P,RM>> in a type alias so the code is more readable.

use astro_nalgebra::{BigFloat, ConstCtx, RoundingMode};
use nalgebra::Vector2;

// This defines a type that has a precision upper bound of
// 1024 bits in the mantissa and no explicit rounding mode 
type BF1024 = BigFloat<ConstCtx<1024>>;

// See the documentation on ConstCtx for how to specify a rounding mode

fn main() {
    let two: BF1024 = "2".parse().unwrap();
    let six: BF1024 = "6".parse().unwrap();
    let vec: Vector2<BF1024> = Vector2::new(two,six);
    let seven: BF1024 = "7".parse().unwrap();
    // Prints [2/7, 6/7] as decimals until 1024 bits
    println!("{}", vec / seven);
}

While it is completely allowed to name the type something like f1024, it does technically break the floating point naming scheme because the type BigFloat<ConstCtx<64>> has 64 bits in the mantissa, while types like f64 only have 52 bits in the mantissa with 12 bits reserved for sign and exponent. So f64 and BigFloat<ConstCtx<64>> are not the same.

Run-time dynamic precision

Dynamic precision is more tricky to implement because some methods outlined in nalgebra::RealField do not have any arguments, so the precision has to be stored in the type. However run-time determined variables cannot be stored in a const generic, so there has to be a dummy type with the methods get_prec and get_rm. There is a macro to quickly define this dummy type which references a global, thread-safe OnceLock that has to be set at runtime.

Example:

use astro_nalgebra::{BigFloat, make_dyn_ctx, RoundingMode};

// This macro takes in the name of a dynamic context (the new type to be made)
// And the name of a global OnceLock to store the precision and rounding mode.
// The name of this global variable is not important, it just has to be unique
// within the scope of the macro call.
make_dyn_ctx!(DynCtx, DYN_CTX_CELL);

type DynFloat = BigFloat<DynCtx>;
fn main() {
    let precision = 88;
    let rounding_mode = RoundingMode::None;
    // Sets the precision and rounding mode of the DynCtx context.
    // This method can only be called once or it will panic.
    DynCtx::set(precision, rounding_mode);
    
    let num: DynFloat = "120".parse().unwrap();
}

Why they have to be implemented with generics

There are two possible ways this library could have been implemented:

  1. A very simple wrapper around astro_float::BigFloat that stored precision in the struct itself.
  2. A wrapper that stores precision as a generic in the type.

The only way to guarantee proper accuracy is to use the latter technique. Here is an example of where the former technique would break:

use nalgebra::RealField;
fn mul_by_pi<T: RealField>(val: T) -> T {
    val * T::pi()
}

If the desired precision of the calculation was stored in val, then the call to BigFloat::pi() would not be able to access the precision because RealField::pi() does not take any arguments. Therefore, the desired precision has to be stored in the type itself.