Commit 4136c5f8 authored by Dzmitry Malyshau's avatar Dzmitry Malyshau
Browse files

Bug 1748499 - Switch WR's scale_factors to operate in double precision r=nical 

This function computes squares of values, but the results are the same magnitude than the source components.
In fact, they can't be larger than the biggest component. Therefore, by doubling the precision of computation we can guarantee that the result will be finite if the source data is finite.

An alternative approach would be to figure out a different computation path that would preserve the precision as it goes. Looks to not be necessary here.

Differential Revision: https://phabricator.services.mozilla.com/D135140
parent 9148916b

gfx/wr/webrender/src/util.rs

+19 −3
Original line number Diff line number Diff line
@@ -1450,11 +1450,14 @@ impl<T: MallocSizeOf> MallocSizeOf for PrimaryArc<T> { 
/// modifications:

///

/// * Removed `xMajor` parameter.

/// * All arithmetics is done with double precision.

pub fn scale_factors<Src, Dst>(

    mat: &Transform3D<f32, Src, Dst>

) -> (f32, f32) {

    let m11 = mat.m11 as f64;

    let m12 = mat.m12 as f64;

    // Determinant is just of the 2D component.

    let det = mat.m11 * mat.m22 - mat.m12 * mat.m21;

    let det = m11 * mat.m22 as f64 - m12 * mat.m21 as f64;

    if det == 0.0 {

        return (0.0, 0.0);

    }
@@ -1462,10 +1465,23 @@ pub fn scale_factors<Src, Dst>( 
    // ignore mirroring

    let det = det.abs();



    let major = (mat.m11 * mat.m11 + mat.m12 * mat.m12).sqrt();

    let major = (m11 * m11 + m12 * m12).sqrt();

    let minor = if major != 0.0 { det / major } else { 0.0 };



    (major, minor)

    (major as f32, minor as f32)

}



#[test]

fn scale_factors_large() {

    // https://bugzilla.mozilla.org/show_bug.cgi?id=1748499

    let mat = Transform3D::<f32, (), ()>::new(

        1.6534229920333123e27, 3.673100922561787e27, 0.0, 0.0,

        -3.673100922561787e27, 1.6534229920333123e27, 0.0, 0.0,

        0.0, 0.0, 1.0, 0.0,

        -828140552192.0, -1771307401216.0, 0.0, 1.0,

    );

    let (major, minor) = scale_factors(&mat);

    assert!(major.is_normal() && minor.is_normal());

} 



/// Clamp scaling factor to a power of two.