Bug 215994 – Optimize the implementation of TransformationMatrix::rotate(double)

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[Wenson Hsieh]

TransformationMatrix::rotate(angle) currently just calls TransformationMatrix::rotate3d(0, 0, 1, angle), which has a fast path for the case where we're rotating about the z-axis. However, we should be able to make this *slightly* faster by omitting the hypotenuse computation, several floating point comparisons, and several assignments.

[Wenson Hsieh]

Created attachment 407580 [details]
For EWS

[Tim Horton]

Comment on attachment 407580 [details]
For EWS

View in context: https://bugs.webkit.org/attachment.cgi?id=407580&action=review

> Source/WebCore/platform/graphics/transforms/TransformationMatrix.cpp:905
> +    multiply({ cosZ, sinZ, -sinZ, cosZ, 0, 0 });

I assume something extends this with the 1s in the diagonals to preserve the other columns?

[Wenson Hsieh]

Comment on attachment 407580 [details]
For EWS

View in context: https://bugs.webkit.org/attachment.cgi?id=407580&action=review

>> Source/WebCore/platform/graphics/transforms/TransformationMatrix.cpp:905
>> +    multiply({ cosZ, sinZ, -sinZ, cosZ, 0, 0 });
> 
> I assume something extends this with the 1s in the diagonals to preserve the other columns?

Indeed — this uses the a, b, c, d, e, f constructor for TransformationMatrix, which ensures that the last two values along the diagonal are 1:

    constexpr TransformationMatrix(double a, double b, double c, double d, double e, double f)
        : m_matrix {
            { a, b, 0, 0 },
            { c, d, 0, 0 },
            { 0, 0, 1, 0 },
            { e, f, 0, 1 },
        }
    {
    }

[EWS]

Committed r266353: <https://trac.webkit.org/changeset/266353>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 407580 [details].

[Radar WebKit Bug Importer]

<rdar://problem/68076663>