fix(vk): create the bindless image heap at device setup (#1088)

The heap was created lazily by pipeline creation and by the shader prewarm,
but a draw binds its bindless slot table before it creates its pipeline. A
title whose first draw samples images, with no pending prewarm work, hit a
null heap in BindFlattenedResourceTable.

The heap depends only on a device capability, so create it with the stores
and drop the lazy calls.

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.Device.Setup.cs

@@ -1474,6 +1474,7 @@ internal static unsafe partial class VulkanVideoPresenter
             CreateScheduler();
             CreateBufferCache();
             CreateImageCache();
+            CreateBindlessImageHeap();
             _descriptorHeap = new DescriptorHeap(_deviceInfo, _scheduler);
             CreateCommandStream();
             LoadDebugUtilsCommands();

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.Pipelines.cs

@@ -622,17 +622,11 @@ internal static unsafe partial class VulkanVideoPresenter
             }
         }
 
-        private void EnsureBindlessImageHeap(params ShaderProgramInfo?[] programs)
-            => EnsureBindlessImageHeapForMode(programs.Any(program => program?.Bindings?.UsesBindlessImages == true));
-
-        private void EnsureBindlessImageHeapForMode(bool usesBindlessImages)
+        // Every translated program samples through this one set when the device supports it,
+        // and a draw binds its slots before it creates its pipeline. Create it with the stores.
+        private void CreateBindlessImageHeap()
         {
-            if (_bindlessImageHeap is not null)
-            {
-                return;
-            }
-
-            if (!usesBindlessImages)
+            if (!BindlessImageHeapEnabled)
             {
                 return;
             }
@@ -730,7 +724,6 @@ internal static unsafe partial class VulkanVideoPresenter
                 throw SubmissionScheduler.Fatal($"The draw binds more color attachments than the device supports: count={description.Rendering.ColorCount} max={_maxColorAttachments}.");
             }
 
-            EnsureBindlessImageHeap(description.VertexStage, description.PixelStage);
             var usesBindlessImages = description.VertexStage.Bindings!.UsesBindlessImages ||
                 description.PixelStage?.Bindings?.UsesBindlessImages == true;
 
@@ -1197,7 +1190,6 @@ internal static unsafe partial class VulkanVideoPresenter
         public PipelineHandle CreateComputePipeline(ComputePipelineDescription description)
         {
             using var profileScope = RenderPhaseProfile.MeasureDetail(RenderPhaseProfile.Phase.PipelineSetup);
-            EnsureBindlessImageHeap(description.Stage);
             var bindings = new List<DescriptorSetLayoutBinding>();
             CollectLayoutBindings(bindings, description.Stage, ShaderStage.Compute);
             var setLayout = CreateDescriptorSetLayout(bindings, out var usesPushDescriptors, out var demand);
@@ -1336,7 +1328,6 @@ internal static unsafe partial class VulkanVideoPresenter
             var bindings = new List<DescriptorSetLayoutBinding>();
             CollectLayoutBindings(bindings, description.Stage, ShaderStage.Compute);
             var setLayout = CreateDescriptorSetLayout(bindings, out var usesPushDescriptors, out var demand);
-            EnsureBindlessImageHeap(description.Stage);
             var layout = CreatePipelineLayout(setLayout, ShaderStageFlags.ComputeBit, description.Stage.Bindings!.UsesBindlessImages);
             var computeModule = new ShaderModule(description.Program.Module);
             if (computeModule.Handle == 0)

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.ShaderPrewarm.cs

@@ -93,9 +93,6 @@ internal static unsafe partial class VulkanVideoPresenter
             var workerCount = Math.Clamp(Environment.ProcessorCount / 4, 1, 4);
             var remaining = workerCount;
             var host = (IShaderPipelineHost)this;
-            // The workers create layouts that include the global set. Initialize
-            // it once on the device setup thread, before any worker can use it.
-            EnsureBindlessImageHeapForMode(host.UsesBindlessImages);
             Volatile.Write(ref _shaderPrewarmTotal, pending.Count);
             Volatile.Write(ref _shaderPrewarmProgress, 0);
             Console.Error.WriteLine(