incremental-delaunay article

articles/hand-opt-simplex-2d/page.mmd

@@ -35,7 +35,7 @@ Benchmark Iterations    Min(ns)    Max(ns)   Variance   Mean(ns)
 full(0)          100  119354512  731397135 3705581696928414  125714847
 ```
 
-Mean difference is `3ms 522µs 206ns`, min difference is `5ms 493µs 883ns`.
+Mean difference is `3ms 522µs 206ns (-2.7%)`, min difference is `5ms 493µs 883ns (-4.4%)`
 
 This suggests that given driver is suboptimal in its optimizing capabilities,
 and I imagine there might be GLSL compilers a lot worse than this.

articles/incremental-delaunay/page.mmd

@@ -0,0 +1,35 @@
+Title:  Optimized Incremental Delaunay
+Brief:  Classic triangulation algorithm to use for one by one insertion of points, with SIMD and caching.
+Date:   1694711563
+Tags:   Programming, Zig, Generation
+CSS:    /style.css
+
+![](/articles/incremental-delaunay/web.png)
+
+Based on [this paper](https://www.cs.umd.edu/class/spring2020/cmsc754/Lects/lect13-delaun-alg.pdf)
+
+Full usable isolated code is [here](/articles/incremental-delaunay/incremental-delaunay.zig.txt).
+
+### Usage example ###
+```zig
+var gpa = std.heap.GeneralPurposeAllocator(.{}){};
+defer std.debug.assert(gpa.deinit() == .ok);
+var triangulator = try Delaunay.Builder.init(gpa.allocator(), Delaunay.Area{-1, -1, 1, 1});
+
+const point_count = 128;
+
+var prng = std.rand.DefaultPrng.init(123123);
+const rng = prng.random();
+for (0..point_count) |_| {
+    const x = rng.float(f32) * 2 - 1;
+    const y = rng.float(f32) * 2 - 1;
+    try triangulator.insertAtRandom(Delaunay.Vertex{ x, y }, rng);
+}
+
+var triangles: [point_count * 2 + 2]gfx.triangle.ScreenspaceTriangle = undefined;
+for (&triangles, triangulator.triangles.items) |*out, in| {
+    out.a = triangulator.vertices.items[in.points[0]];
+    out.b = triangulator.vertices.items[in.points[1]];
+    out.c = triangulator.vertices.items[in.points[2]];
+}
+```