Clipper1 performance vs Clipper2

[ErikSom]

I was working on a simple benchmark to compare the WASM port of Clipper1 with the WASM port of Clipper2 and currently Clipper1 is outperforming Clipper2 slightly, here is my setup:
https://github.com/ErikSom/Clipper2-WASM/tree/main/clipper2-wasm/examples/benchmark

These are my results:

clipper1 Intersection EvenOdd 5000 ops over a circle of 5000 points: 2847.0999999940395 ms
clipper2 Intersection EvenOdd 5000 ops over a circle of 5000 points: 3056 ms

clipper1 Union EvenOdd 5000 ops over a circle of 5000 points: 3722.800000011921 ms
clipper2 Union EvenOdd 5000 ops over a circle of 5000 points: 4470 ms

clipper1 Difference EvenOdd 5000 ops over a circle of 5000 points: 3259.7000000178814 ms
clipper2 Difference EvenOdd 5000 ops over a circle of 5000 points: 3751.399999976158 ms

clipper1 Xor EvenOdd 5000 ops over a circle of 5000 points: 4066 ms
clipper2 Xor EvenOdd 5000 ops over a circle of 5000 points: 4982.799999982119 ms

Here clipper2 is slightly slower - and I'm still figuring out whether this is due compile differences, implementation differences or a difference in the actual lib.

[AngusJohnson]

Test1: Intersect 2 random self-intersecting polygons

Clipper1:
3,583 msecs

Clipper2:
557 msecs


Test2: Intersect 2 overlapping ellipses

Clipper1:
1,058 microsecs

Clipper2:
761 microsecs

Clipper1_vs_Clipper2_performance.zip

I suspect the performance issues you're seeing with Clipper2 is related to your translation.

Edit: I suggest that your 'benchmark' intersecting/unioning etc 2 overlapping circles really isn't a good test because despite the number of edges in each circle, there will only be 2 points of intersections.

[ErikSom]

Thanks! I will investigate further, will report back when i got it sorted.

[ErikSom]

Regarding the test - i took it from js-angus wasm build for Clipper1 to have a fair comparison with that:
https://github.com/xaviergonz/js-angusj-clipper/blob/380839193ce6c0d196cdfa7c7d44844bf3c2eaa9/__tests__/tests.spec.ts#L249

[ErikSom]

I've gotten it a lot closer to Clipper1 in terms of performance, but still haven't seen the improvements I would like, i've moved the discussion here for now:
emscripten-core/emscripten#20985

[ErikSom]

@AngusJohnson I have figured it out and am finally seeing the improvements :) I was comparing IntersectD with an intersect64, there is the overhead of transforming the doubles to ints. The new results here:

clipper1 Intersection EvenOdd 5000 ops 5000 points: 2898.5 ms
clipper2 Intersection EvenOdd 5000 ops 5000 points: 2061.5 ms
28.88% improvement

clipper1 Union EvenOdd 5000 ops 5000 points: 3812.699999988079 ms
clipper2 Union EvenOdd 5000 ops 5000 points: 3094.800000011921 ms
18.83% improvement

clipper1 Difference EvenOdd 5000 ops 5000 points: 3291 ms
clipper2 Difference EvenOdd 5000 ops 5000 points: 2554.699999988079 ms
22.37% improvement

clipper1 Xor EvenOdd 5000 ops 5000 points: 4045.400000035763 ms
clipper2 Xor EvenOdd 5000 ops 5000 points: 3576 ms
11.60% improvement

clipper1 Intersection EvenOdd 10000 ops 100 points: 185 ms
clipper2 Intersection EvenOdd 10000 ops 100 points: 106.40000003576279 ms
42.49% improvement

clipper1 Union EvenOdd 10000 ops 100 points: 236.39999997615814 ms
clipper2 Union EvenOdd 10000 ops 100 points: 141.89999997615814 ms
39.97% improvement

clipper1 Difference EvenOdd 10000 ops 100 points: 209.19999998807907 ms
clipper2 Difference EvenOdd 10000 ops 100 points: 123.40000003576279 ms
41.01% improvement

clipper1 Xor EvenOdd 10000 ops 100 points: 256.60000002384186 ms
clipper2 Xor EvenOdd 10000 ops 100 points: 156.4000000357628 ms
39.05% improvement