Cache shared text shaping and glyph blobs independently of WebGPU

benchmarks/WebScene.NativeEngine.Benchmarks/Program.cs

@@ -8,6 +8,8 @@
     {
         switch (args[1].ToLowerInvariant())
         {
+            case "text-cache":
+                return TextCacheProbe.Run(probeArgs);
             case "variable-font":
                 return VariableFontProbe.Run(probeArgs);
             case "generated-realtime-chart":
@@ -36,7 +38,7 @@
     }
 
     Console.Error.WriteLine(
-        "Unknown probe. Use one of: variable-font, generated-realtime-chart, native-interop-race, " +
+        "Unknown probe. Use one of: text-cache, variable-font, generated-realtime-chart, native-interop-race, " +
         "native-runtime-work, native-dom-lookup, native-context-memory, native-lifecycle, " +
         "native-resize-cadence, native-inspector-disabled-performance, " +
         "native-retained-render, " +

benchmarks/WebScene.NativeEngine.Benchmarks/TextCacheProbe.cs

@@ -0,0 +1,69 @@
+using System.Diagnostics;
+using System.Text.Json;
+using SkiaSharp;
+using SkiaSharp.HarfBuzz;
+using WebScene.Backends.Avalonia.Native;
+
+namespace WebScene.NativeEngine.Benchmarks;
+
+// Warm repeated-label microbenchmark. Uses a CPU bitmap, not a GPU context.
+internal static class TextCacheProbe
+{
+    internal static int Run(string[] args)
+    {
+        if (args.Length != 0) throw new ArgumentException("text-cache takes no arguments");
+        const int iterations = 10000;
+        const string text = "Courtyard House 123 office a\u0301";
+        using var face = SKTypeface.FromFamilyName("sans-serif");
+        using var shaper = new SKShaper(face);
+        using var paint = new SKPaint { Typeface = face, TextSize = 18, Color = SKColors.Black };
+        using var bitmap = new SKBitmap(400, 64);
+        using var canvas = new SKCanvas(bitmap);
+        canvas.Clear(SKColors.White);
+        var cache = new NativeTextShaping.ShapedRunCache(2048, 4 * 1024 * 1024);
+        void DrawUncached()
+        {
+            var shaped = shaper.Shape(text, 0, 32, paint);
+            using var font = paint.ToFont();
+            NativeTextShaping.ApplyFontRasterizationProfile(font, 1,
+                NativeTextShaping.NativeFontRasterizationMode.Current);
+            using var builder = new SKTextBlobBuilder();
+            var run = builder.AllocatePositionedRun(font, shaped.Codepoints.Length);
+            for (var i = 0; i < shaped.Codepoints.Length; i++)
+            {
+                run.GetGlyphSpan()[i] = (ushort)shaped.Codepoints[i];
+                run.GetPositionSpan()[i] = shaped.Points[i];
+            }
+            using var blob = builder.Build();
+            canvas.DrawText(blob, 0, 0, paint);
+        }
+        void DrawCached() => NativeTextShaping.DrawShapedText(canvas, shaper, text, 0, 32,
+            paint, 0, rasterizationMode: NativeTextShaping.NativeFontRasterizationMode.Current);
+        Action[] operations = [() => shaper.Shape(text, 0, 0, paint),
+            () => cache.Shape(shaper, text, paint), DrawUncached, DrawCached];
+        string[] names = ["shape_uncached", "shape_cached", "draw_uncached", "draw_cached"];
+        foreach (var operation in operations)
+            for (var i = 0; i < 1000; i++) operation();
+        // Reverse each alternate round to reduce simple ordering bias.
+        for (var round = 0; round < 6; round++)
+        {
+            for (var step = 0; step < operations.Length; step++)
+            {
+                var index = round % 2 == 0 ? step : operations.Length - step - 1;
+                var allocated = GC.GetAllocatedBytesForCurrentThread();
+                var start = Stopwatch.GetTimestamp();
+                for (var i = 0; i < iterations; i++) operations[index]();
+                var elapsed = Stopwatch.GetTimestamp() - start;
+                var bytes = GC.GetAllocatedBytesForCurrentThread() - allocated;
+                Console.WriteLine(JsonSerializer.Serialize(new
+                {
+                    round, operation = names[index], iterations, family = face.FamilyName,
+                    microsecondsPerOperation = elapsed * 1_000_000.0 / Stopwatch.Frequency / iterations,
+                    managedBytesPerOperation = (double)bytes / iterations,
+                    cpuBitmap = true
+                }));
+            }
+        }
+        return 0;
+    }
+}

docs/validation/text-cache-macos.json

@@ -0,0 +1,218 @@
+[
+  {
+    "round": 0,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 5.0962290999999995,
+    "managedBytesPerOperation": 1784.0128,
+    "cpuBitmap": true
+  },
+  {
+    "round": 0,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.2909292,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 0,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 7.3138584,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 0,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.7095791,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 1,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.753475,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 1,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 7.1130832999999996,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 1,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.28925,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 1,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 4.9029958,
+    "managedBytesPerOperation": 1784.0056,
+    "cpuBitmap": true
+  },
+  {
+    "round": 2,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 4.0875667,
+    "managedBytesPerOperation": 1784,
+    "cpuBitmap": true
+  },
+  {
+    "round": 2,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.2092417,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 2,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 7.0206375,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 2,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.1541709,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 3,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.1295,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 3,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 6.9091583,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 3,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.08505,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 3,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 3.3211833,
+    "managedBytesPerOperation": 1784,
+    "cpuBitmap": true
+  },
+  {
+    "round": 4,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 3.3064625,
+    "managedBytesPerOperation": 1784,
+    "cpuBitmap": true
+  },
+  {
+    "round": 4,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.0806625,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 4,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 6.7598,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 4,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.109125,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 5,
+    "operation": "draw_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 2.1088958,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 5,
+    "operation": "draw_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 5.9011292,
+    "managedBytesPerOperation": 2104,
+    "cpuBitmap": true
+  },
+  {
+    "round": 5,
+    "operation": "shape_cached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 0.0755667,
+    "managedBytesPerOperation": 0,
+    "cpuBitmap": true
+  },
+  {
+    "round": 5,
+    "operation": "shape_uncached",
+    "iterations": 10000,
+    "family": "Helvetica",
+    "microsecondsPerOperation": 3.260175,
+    "managedBytesPerOperation": 1784,
+    "cpuBitmap": true
+  }
+]