Egon Elbre 24d6f3fb65 f32: optimize Affine2D ...

encode/decode seem to introduce significant overhead. Inline them
manually. It'll make code harder to read, however the performance wins
are significant.

name \ time/op                 before       after        delta
TransformOffset-32             2.64ns ± 0%  0.25ns ± 0%   ~     (p=0.100 n=3+3)
TransformScale-32              2.64ns ± 0%  0.25ns ± 1%   ~     (p=0.100 n=3+3)
TransformRotate-32             2.65ns ± 0%  0.24ns ± 3%   ~     (p=0.100 n=3+3)
TransformTranslateMultiply-32  42.5ns ± 0%  12.9ns ± 0%   ~     (p=0.100 n=3+3)
TransformScaleMultiply-32      42.6ns ± 0%  12.9ns ± 0%   ~     (p=0.100 n=3+3)
TransformMultiply-32           42.2ns ± 0%  12.9ns ± 2%   ~     (p=0.100 n=3+3)

Signed-off-by: Egon Elbre <egonelbre@gmail.com>

2020-10-13 15:14:33 +02:00

..

affine_test.go

f32: optimize Affine2D

2020-10-13 15:14:33 +02:00

affine.go

f32: optimize Affine2D

2020-10-13 15:14:33 +02:00

f32.go

layout,f32: add convenience functions for rectangles and points

2020-05-19 10:59:32 +02:00