tmp - progress capture only

Author: JimBobSquarePants

ImageSharp.Drawing.sln

@@ -346,6 +346,8 @@ Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "WebGPUWindowDemo", "samples
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ImageSharp.Drawing.WebGPU.ShaderGen", "src\ImageSharp.Drawing.WebGPU.ShaderGen\ImageSharp.Drawing.WebGPU.ShaderGen.csproj", "{C7606104-5D58-4670-912C-3F336606B02D}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ImageSharp", "..\ImageSharp\src\ImageSharp\ImageSharp.csproj", "{FE71764E-E7BD-B378-8A4C-86929A28AD25}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Any CPU = Debug|Any CPU
@@ -452,6 +454,18 @@ Global
 		{C7606104-5D58-4670-912C-3F336606B02D}.Release|x64.Build.0 = Release|Any CPU
 		{C7606104-5D58-4670-912C-3F336606B02D}.Release|x86.ActiveCfg = Release|Any CPU
 		{C7606104-5D58-4670-912C-3F336606B02D}.Release|x86.Build.0 = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|x64.ActiveCfg = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|x64.Build.0 = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|x86.ActiveCfg = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Debug|x86.Build.0 = Debug|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|Any CPU.Build.0 = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|x64.ActiveCfg = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|x64.Build.0 = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|x86.ActiveCfg = Release|Any CPU
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25}.Release|x86.Build.0 = Release|Any CPU
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE
@@ -483,6 +497,7 @@ Global
 		{061582C2-658F-40AE-A978-7D74A4EB2C0A} = {815C0625-CD3D-440F-9F80-2D83856AB7AE}
 		{2541FDCD-78AC-40DB-B5E3-6A715DC132BA} = {528610AC-7C0C-46E8-9A2D-D46FD92FEE29}
 		{C7606104-5D58-4670-912C-3F336606B02D} = {815C0625-CD3D-440F-9F80-2D83856AB7AE}
+		{FE71764E-E7BD-B378-8A4C-86929A28AD25} = {815C0625-CD3D-440F-9F80-2D83856AB7AE}
 	EndGlobalSection
 	GlobalSection(ExtensibilityGlobals) = postSolution
 		SolutionGuid = {5F8B9D1F-CD8B-4CC5-8216-D531E25BD795}
@@ -491,6 +506,7 @@ Global
 		shared-infrastructure\src\SharedInfrastructure\SharedInfrastructure.projitems*{061582c2-658f-40ae-a978-7d74a4eb2c0a}*SharedItemsImports = 5
 		shared-infrastructure\src\SharedInfrastructure\SharedInfrastructure.projitems*{2e33181e-6e28-4662-a801-e2e7dc206029}*SharedItemsImports = 5
 		shared-infrastructure\src\SharedInfrastructure\SharedInfrastructure.projitems*{68a8cc40-6aed-4e96-b524-31b1158fdeea}*SharedItemsImports = 13
+		..\ImageSharp\shared-infrastructure\src\SharedInfrastructure\SharedInfrastructure.projitems*{fe71764e-e7bd-b378-8a4c-86929a28ad25}*SharedItemsImports = 5
 	EndGlobalSection
 	GlobalSection(Performance) = preSolution
 		HasPerformanceSessions = true

samples/DrawingBackendBenchmark/SkiaSharpBenchmarkBackend.cs

@@ -24,6 +24,13 @@ public SkiaSharpBenchmarkBackend(GRContext? context = null)
 
     public bool IsGpu => this.context is not null;
 
+    /// <summary>
+    /// Renders the benchmark scene through Skia and optionally captures a readback preview.
+    /// </summary>
+    /// <remarks>
+    /// The returned duration measures submission through <c>Flush()</c>. Preview readback happens afterward, so the
+    /// reported GPU timing is a submission timing rather than a fully synchronized render-complete timing.
+    /// </remarks>
     public BenchmarkRenderResult Render(ReadOnlySpan<VisualLine> lines, int width, int height, bool capturePreview)
     {
         this.EnsureSurface(width, height);

samples/DrawingBackendBenchmark/VisualLine.cs

@@ -1,4 +1,4 @@
-// Copyright (c) Six Labors.
+// Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
 using SixLabors.ImageSharp.Drawing.Processing;
@@ -23,6 +23,11 @@ internal readonly record struct VisualLine(PointF Start, PointF End, Color Color
     /// </summary>
     public SKColor SkiaColor { get; } = ToSkiaColor(Color);
 
+    /// <summary>
+    /// Gets the pre-created ImageSharp pen for this line, avoiding one per-frame allocation in the benchmark hot path.
+    /// </summary>
+    public SolidPen Pen { get; } = new(Color, Width);
+
     private static SKColor ToSkiaColor(Color color)
     {
         Rgba32 rgba = color.ToPixel<Rgba32>();
@@ -38,7 +43,7 @@ public static void RenderLinesToCanvas(DrawingCanvas<Bgra32> canvas, ReadOnlySpa
         canvas.Fill(BackgroundBrush);
         foreach (VisualLine visualLine in lines)
         {
-            canvas.DrawLine(new SolidPen(visualLine.Color, visualLine.Width), [visualLine.Start, visualLine.End]);
+            canvas.DrawLine(visualLine.Pen, visualLine.Start, visualLine.End);
         }
     }
 }

samples/DrawingBackendBenchmark/WebGpuBenchmarkBackend.cs

@@ -63,6 +63,11 @@ public static bool TryCreate([NotNullWhen(true)] out WebGpuBenchmarkBackend? res
     /// <summary>
     /// Renders the benchmark scene through the WebGPU backend and optionally captures a readback preview.
     /// </summary>
+    /// <remarks>
+    /// The returned duration measures scene submission through <see cref="DrawingCanvas{TPixel}.Flush()"/>.
+    /// Readback for preview capture happens afterward, so the reported GPU timing is a submission timing rather than
+    /// a fully synchronized render-complete timing.
+    /// </remarks>
     public BenchmarkRenderResult Render(ReadOnlySpan<VisualLine> lines, int width, int height, bool capturePreview)
     {
         this.ThrowIfUnavailable();

src/ImageSharp.Drawing.WebGPU/Shaders/WgslSource/Shared/pathtag.wgsl

@@ -21,7 +21,7 @@ const PATH_TAG_STYLE = 0x40u;
 const PATH_TAG_SUBPATH_END = 4u;
 
 // Size of the `Style` data structure in words
-const STYLE_SIZE_IN_WORDS: u32 = 2u;
+const STYLE_SIZE_IN_WORDS: u32 = 3u;
 
 const STYLE_FLAGS_STYLE: u32 = 0x80000000u;
 const STYLE_FLAGS_FILL: u32 = 0x40000000u;
@@ -38,6 +38,8 @@ const STYLE_FLAGS_JOIN_MASK: u32 = 0x30000000u;
 const STYLE_FLAGS_JOIN_BEVEL: u32 = 0u;
 const STYLE_FLAGS_JOIN_MITER: u32 = 0x10000000u;
 const STYLE_FLAGS_JOIN_ROUND: u32 = 0x20000000u;
+const STYLE_FLAGS_JOIN_MITER_REVERT: u32 = 0x00400000u;
+const STYLE_FLAGS_JOIN_MITER_ROUND: u32 = 0x00800000u;
 
 // TODO: Declare the remaining STYLE flags here.
 

src/ImageSharp.Drawing.WebGPU/Shaders/WgslSource/flatten.wgsl

@@ -567,6 +567,11 @@ fn draw_join(
         case STYLE_FLAGS_JOIN_MITER: {
             let hypot = length(vec2f(cr, d));
             let miter_limit = unpack2x16float(style_flags & STYLE_MITER_LIMIT_MASK)[0];
+            let is_backside = cr > 0.;
+            let outer_prev = select(front0, back1, is_backside);
+            let outer_next = select(front1, back0, is_backside);
+            let inner_prev = select(back0, front0, is_backside);
+            let inner_next = select(back1, front1, is_backside);
 
             var line_ix: u32;
             // Given the two tangents `tan_prev` and `tan_next` arranged tail-to-tail, the
@@ -583,29 +588,43 @@ fn draw_join(
             if 2. * hypot < (hypot + d) * miter_limit * miter_limit
                 && abs(cr) > TANGENT_THRESH * TANGENT_THRESH
             {
-                let is_backside = cr > 0.;
-                let fp_last = select(front0, back1, is_backside);
-                let fp_this = select(front1, back0, is_backside);
-                let p = select(front0, back0, is_backside);
-
-                let v = fp_this - fp_last;
+                let v = outer_next - outer_prev;
                 let h = (tan_prev.x * v.y - tan_prev.y * v.x) / cr;
-                let miter_pt = fp_this - tan_next * h;
+                let miter_pt = outer_next - tan_next * h;
 
                 line_ix = atomicAdd(&bump.lines, 3u);
-                write_line_with_transform(line_ix, path_ix, p, miter_pt, transform);
-                line_ix += 1u;
-
-                if is_backside {
-                    back0 = miter_pt;
+                write_line_with_transform(line_ix, path_ix, outer_prev, miter_pt, transform);
+                write_line_with_transform(line_ix + 1u, path_ix, miter_pt, outer_next, transform);
+                write_line_with_transform(line_ix + 2u, path_ix, inner_prev, inner_next, transform);
+            } else {
+                let use_round_overflow = (style_flags & STYLE_FLAGS_JOIN_MITER_ROUND) != 0u;
+                let use_revert_overflow = (style_flags & STYLE_FLAGS_JOIN_MITER_REVERT) != 0u;
+                if use_round_overflow {
+                    flatten_arc(path_ix, outer_prev, outer_next, p0, abs(atan2(cr, d)), transform);
+                    output_line_with_transform(path_ix, inner_prev, inner_next, transform);
+                } else if use_revert_overflow || abs(cr) <= TANGENT_THRESH * TANGENT_THRESH {
+                    output_two_lines_with_transform(path_ix, front0, front1, back0, back1, transform);
                 } else {
-                    front0 = miter_pt;
+                    let v = outer_next - outer_prev;
+                    let h = (tan_prev.x * v.y - tan_prev.y * v.x) / cr;
+                    let miter_pt = outer_next - tan_next * h;
+                    let limit = length(n_prev) * miter_limit;
+                    let bevel_distance = length(((outer_prev - p0) + (outer_next - p0)) * 0.5);
+                    let intersection_distance = distance(p0, miter_pt);
+                    if intersection_distance <= bevel_distance + TANGENT_THRESH {
+                        output_two_lines_with_transform(path_ix, front0, front1, back0, back1, transform);
+                    } else {
+                        let ratio = (limit - bevel_distance) / (intersection_distance - bevel_distance);
+                        let clipped_prev = outer_prev + ((miter_pt - outer_prev) * ratio);
+                        let clipped_next = outer_next + ((miter_pt - outer_next) * ratio);
+                        line_ix = atomicAdd(&bump.lines, 4u);
+                        write_line_with_transform(line_ix, path_ix, outer_prev, clipped_prev, transform);
+                        write_line_with_transform(line_ix + 1u, path_ix, clipped_prev, clipped_next, transform);
+                        write_line_with_transform(line_ix + 2u, path_ix, clipped_next, outer_next, transform);
+                        write_line_with_transform(line_ix + 3u, path_ix, inner_prev, inner_next, transform);
+                    }
                 }
-            } else {
-                line_ix = atomicAdd(&bump.lines, 2u);
             }
-            write_line_with_transform(line_ix, path_ix, front0, front1, transform);
-            write_line_with_transform(line_ix + 1u, path_ix, back0, back1, transform);
         }
         case STYLE_FLAGS_JOIN_ROUND: {
             var arc0: vec2f;
@@ -691,10 +710,8 @@ fn compute_tag_monoid(ix: u32) -> PathTagData {
     // TODO: this can be a read buf overflow. Conditionalize by tag byte?
     tm = combine_tag_monoid(tag_monoids[ix >> 2u], tm);
     var tag_byte = (tag_word >> shift) & 0xffu;
-    // We no longer encode an initial transform and style so these
-    // are off by one.
-    // Note: an alternative would be to adjust config.transform_base and
-    // config.style_base.
+    // The encoded streams begin after the implicit identity transform and current style,
+    // so these indices are rebased to the actual payload starts.
     tm.trans_ix -= 1u;
     tm.style_ix -= STYLE_SIZE_IN_WORDS;
     return PathTagData(tag_byte, tm);
@@ -847,10 +864,9 @@ fn main(
 
     let out = &path_bboxes[path_ix];
     let style_flags = scene[config.style_base + style_ix];
+    let style_draw_flags = scene[config.style_base + style_ix + 2u];
     let fill_rule = select(DRAW_INFO_FLAGS_FILL_RULE_BIT, 0u, (style_flags & STYLE_FLAGS_FILL) == 0u);
-    let blend_mode = ((style_flags & 0x00003ffeu) >> 1u) << DRAW_FLAGS_BLEND_MODE_SHIFT;
-    let blend_alpha = ((style_flags & 0x3fffc000u) >> 14u) << DRAW_FLAGS_BLEND_ALPHA_SHIFT;
-    let draw_flags = fill_rule | blend_mode | blend_alpha;
+    let draw_flags = style_draw_flags | fill_rule;
     if (tag.tag_byte & PATH_TAG_PATH) != 0u {
         (*out).draw_flags = draw_flags;
         (*out).trans_ix = trans_ix;

src/ImageSharp.Drawing.WebGPU/WebGPUDrawingBackend.cs

@@ -1,7 +1,6 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
-using System.Collections.Generic;
 using System.Runtime.CompilerServices;
 using Silk.NET.WebGPU;
 using SixLabors.ImageSharp.Memory;
@@ -217,7 +216,7 @@ public void FlushCompositions<TPixel>(
         where TPixel : unmanaged, IPixel<TPixel>
     {
         this.ThrowIfDisposed();
-        if (compositionScene.Commands.Count == 0)
+        if (compositionScene.CommandCount == 0)
         {
             return;
         }
@@ -227,7 +226,7 @@ public void FlushCompositions<TPixel>(
         WebGPUSceneBumpSizes currentBumpSizes = this.bumpSizes;
         for (int attempt = 0; attempt < MaxDynamicGrowthAttempts; attempt++)
         {
-            if (!WebGPUSceneDispatch.TryCreateStagedScene(configuration, target, compositionScene.Commands, currentBumpSizes, out bool exceedsBindingLimit, out WebGPUSceneDispatch.BindingLimitFailure bindingLimitFailure, out WebGPUStagedScene stagedScene, out string? error))
+            if (!WebGPUSceneDispatch.TryCreateStagedScene(configuration, target, compositionScene, currentBumpSizes, out bool exceedsBindingLimit, out WebGPUSceneDispatch.BindingLimitFailure bindingLimitFailure, out WebGPUStagedScene stagedScene, out string? error))
             {
                 this.TestingLastGPUInitializationFailure = exceedsBindingLimit
                     ? error ?? "The staged WebGPU scene exceeded the current binding limits."

src/ImageSharp.Drawing.WebGPU/WebGPUSceneDispatch.cs

@@ -1,7 +1,7 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
-#pragma warning disable SA1201 // Phase-1 staged scene types are grouped by pipeline role.
+#pragma warning disable SA1201 // Staged scene types are grouped by pipeline role.
 
 using System.Diagnostics;
 using System.IO;
@@ -100,14 +100,14 @@ public static bool TryCreateStagedScene<TPixel>(
         out WebGPUStagedScene stagedScene,
         out string? error)
         where TPixel : unmanaged, IPixel<TPixel>
-        => TryCreateStagedScene(configuration, target, scene.Commands, bumpSizes, out _, out _, out stagedScene, out error);
+        => TryCreateStagedScene(configuration, target, scene, bumpSizes, out _, out _, out stagedScene, out error);
 
     /// <summary>
     /// Builds the flush-scoped encoded scene and uploads its GPU resources.
     /// </summary>
     /// <param name="configuration">The drawing configuration that owns allocators and backend settings.</param>
     /// <param name="target">The flush target that exposes the native WebGPU surface.</param>
-    /// <param name="commands">The prepared composition commands for this flush.</param>
+    /// <param name="scene">The prepared composition scene for this flush.</param>
     /// <param name="bumpSizes">The current dynamic scratch capacities to use for this attempt.</param>
     /// <param name="exceedsBindingLimit">Receives whether creation failed because a single WebGPU binding would be too large.</param>
     /// <param name="bindingLimitFailure">Receives the exact binding-limit failure when <paramref name="exceedsBindingLimit"/> is <see langword="true"/>.</param>
@@ -118,7 +118,7 @@ public static bool TryCreateStagedScene<TPixel>(
     public static bool TryCreateStagedScene<TPixel>(
         Configuration configuration,
         ICanvasFrame<TPixel> target,
-        IReadOnlyList<CompositionCommand> commands,
+        CompositionScene scene,
         WebGPUSceneBumpSizes bumpSizes,
         out bool exceedsBindingLimit,
         out BindingLimitFailure bindingLimitFailure,
@@ -137,13 +137,6 @@ public static bool TryCreateStagedScene<TPixel>(
             return false;
         }
 
-        WebGPUSceneSupportResult support = WebGPUSceneEncoder.ValidateSceneSupport(commands);
-        if (!support.IsSupported)
-        {
-            error = support.Error;
-            return false;
-        }
-
         TextureFormat textureFormat = WebGPUTextureFormatMapper.ToSilk(formatId);
 
         WebGPUFlushContext? flushContext = WebGPUFlushContext.Create(
@@ -160,7 +153,14 @@ public static bool TryCreateStagedScene<TPixel>(
 
         try
         {
-            encodedScene = WebGPUSceneEncoder.Encode(commands, support, flushContext.TargetBounds, flushContext.MemoryAllocator);
+            if (!WebGPUSceneEncoder.TryEncode(scene, flushContext.TargetBounds, flushContext.MemoryAllocator, out WebGPUEncodedScene createdScene, out error))
+            {
+                flushContext.Dispose();
+                stagedScene = default;
+                return false;
+            }
+
+            encodedScene = createdScene;
             WebGPUSceneBumpSizes sceneBumpSizes = bumpSizes.WithSceneLowerBounds(encodedScene);
             WebGPUSceneConfig config = WebGPUSceneConfig.Create(encodedScene, sceneBumpSizes);
             uint baseColor = 0U;