Begin work on bindless texturing

DX12: add ability to bind the entire CBV_SRV_UAV heap at once
RavEngine · Jul 4, 2024 · a83433f · a83433f
1 parent 2df1b38
commit a83433f
Show file tree

Hide file tree

Showing 15 changed files with 64 additions and 6 deletions.
diff --git a/include/RGL/CommandBuffer.hpp b/include/RGL/CommandBuffer.hpp
@@ -129,6 +129,12 @@ struct TextureView;
         virtual void SetVertexTexture(const TextureView& texture, uint32_t index) = 0;
         virtual void SetFragmentTexture(const TextureView& texture, uint32_t index) = 0;
         virtual void SetComputeTexture(const TextureView& texture, uint32_t index) = 0;
+
+		struct MakeResidentConfig {
+			bool written = false;
+		};
+		virtual void MakeResident(const TextureView& texture, const MakeResidentConfig& config) = 0;
+
 
         virtual void Draw(uint32_t nVertices, const DrawInstancedConfig& = {}) = 0;
         virtual void DrawIndexed(uint32_t nIndices, const DrawIndexedInstancedConfig& = {}) = 0;

diff --git a/include/RGL/Device.hpp b/include/RGL/Device.hpp
@@ -29,6 +29,8 @@ namespace RGL {
         } mtlData;
 	};
 
+	struct TextureView;
+
 	struct IDevice {
 		virtual ~IDevice() {}
 		static RGLDevicePtr CreateSystemDefaultDevice();
@@ -53,6 +55,8 @@ namespace RGL {
 
 		virtual RGLCommandQueuePtr CreateCommandQueue(QueueType type) = 0;
 
+		virtual TextureView GetGlobalBindlessTextureHeap() const = 0;
+
 		virtual RGLComputePipelinePtr CreateComputePipeline(const struct ComputePipelineDescriptor&) = 0;
 
 		virtual size_t GetTotalVRAM() const = 0;

diff --git a/include/RGL/Pipeline.hpp b/include/RGL/Pipeline.hpp
@@ -59,6 +59,7 @@ namespace RGL {
 	struct PipelineLayoutDescriptor {
 		struct LayoutBindingDesc {
 			uint32_t binding = 0;
+			uint32_t count = 1;
 			BindingType type;
 			BindingVisibility stageFlags;
             bool writable = false;

diff --git a/include/RGL/Texture.hpp b/include/RGL/Texture.hpp
@@ -3,6 +3,7 @@
 #include <cstdint>
 #include <limits>
 #include <string_view>
+#include <array>
 #include "TextureFormat.hpp"
 #include "SubresourceRange.hpp"
 
@@ -125,6 +126,7 @@ namespace RGL {
 		TextureUsage usage;
 		TextureAspect aspect;
 		uint32_t width = 0, height = 0, depth = 1, mipLevels = 1, arrayLayers = 1;
+		std::array<float, 4> optimizedClearValue{ 0,0,0,0 };
 		TextureType imageType = decltype(imageType)::T2D;
 		TextureFormat format;
 		TilingMode mode = decltype(mode)::Optimal;

diff --git a/src/D3D12CommandBuffer.cpp b/src/D3D12CommandBuffer.cpp
@@ -284,7 +284,7 @@ namespace RGL {
 		else {
 			assert(thisTexture.srvAllocated(), "Cannot bind this texture because it is not in a SRV heap!");
 		}
-		auto& heap = thisTexture.parentResource->owningDevice->CBV_SRV_UAVHeap;
+		auto& heap = owningQueue->owningDevice->CBV_SRV_UAVHeap;
 
 		if (isGraphics) {
 			commandList->SetGraphicsRootDescriptorTable(textureSlot.slot, heap->GetGpuHandle(textureSlot.isUAV ? thisTexture.uavIDX : thisTexture.srvIDX));
@@ -297,6 +297,12 @@ namespace RGL {
 	{
 		SetFragmentTexture(texture, index);
 	}
+
+	void CommandBufferD3D12::MakeResident(const TextureView& texture, const MakeResidentConfig& config)
+	{
+
+	}
+
 	void CommandBufferD3D12::Draw(uint32_t nVertices, const DrawInstancedConfig& config)
 	{
 		commandList->DrawInstanced(nVertices, config.nInstances, config.startVertex, config.firstInstance);
@@ -625,6 +631,10 @@ namespace RGL {
 	}
 	void CommandBufferD3D12::SyncIfNeeded(TextureView texture, D3D12_RESOURCE_STATES needed, bool written)
 	{
+		if (texture.texture.dx.parentResource == nullptr) {
+			return;
+		}
+
 		D3D12TextureLastUseKey key{texture.texture.dx.parentResource, texture.texture.dx.coveredMips, texture.texture.dx.coveredLayers};
 		auto parent = texture.texture.dx.parentResource;
 

diff --git a/src/D3D12CommandBuffer.hpp b/src/D3D12CommandBuffer.hpp
@@ -91,6 +91,8 @@ namespace RGL {
 		void SetFragmentTexture(const TextureView& texture, uint32_t index) final;
 		void SetComputeTexture(const TextureView& texture, uint32_t index) final;
 
+		void MakeResident(const TextureView& texture, const MakeResidentConfig& config) final;
+
 		void Draw(uint32_t nVertices, const DrawInstancedConfig & = {}) final;
 		void DrawIndexed(uint32_t nIndices, const DrawIndexedInstancedConfig & = {}) final;
 

diff --git a/src/D3D12Device.cpp b/src/D3D12Device.cpp
@@ -293,6 +293,20 @@ namespace RGL {
         };
     }
 
+    TextureView DeviceD3D12::GetGlobalBindlessTextureHeap() const
+    {
+
+        return TextureView{ {
+            .dsvIDX = 0,
+            .rtvIDX = 0,
+            .srvIDX = 0,
+            .uavIDX = 0,
+            .parentResource = nullptr,      // bindless must set barriers elsewhere
+            .coveredMips = ALL_MIPS,
+            .coveredLayers = ALL_LAYERS
+        } };
+    }
+
     RGLCommandQueuePtr RGL::DeviceD3D12::CreateCommandQueue(QueueType type)
     {
         return std::make_shared<CommandQueueD3D12>(this, type);

diff --git a/src/D3D12Device.hpp b/src/D3D12Device.hpp
@@ -63,6 +63,8 @@ namespace RGL {
 
 		DeviceData GetDeviceData() final;
 
+		TextureView GetGlobalBindlessTextureHeap() const final;
+
 		RGLCommandQueuePtr CreateCommandQueue(QueueType type) final;
 		RGLFencePtr CreateFence(bool preSignaled) final;
 		void BlockUntilIdle() final;

diff --git a/src/D3D12RenderPipeline.cpp b/src/D3D12RenderPipeline.cpp
@@ -187,7 +187,7 @@ namespace RGL {
                     {
                         auto& range = ranges.emplace_back(D3D12_DESCRIPTOR_RANGE1{
                             .RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SAMPLER,
-                            .NumDescriptors = 1,
+                            .NumDescriptors = item.count,
                             .BaseShaderRegister = item.binding,
                             .RegisterSpace = 0,
                             .OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND,
@@ -202,9 +202,10 @@ namespace RGL {
                     {
                         auto& range = ranges.emplace_back(D3D12_DESCRIPTOR_RANGE1{
                             .RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV,
-                            .NumDescriptors = 1,
+                            .NumDescriptors = item.count,
                             .BaseShaderRegister = item.binding,
                             .RegisterSpace = 0,
+                            .Flags = D3D12_DESCRIPTOR_RANGE_FLAG_DESCRIPTORS_VOLATILE,
                             .OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND,
                             });
                         textureBindingToRootSlot[item.binding] = { uint32_t(rootParameters.size()), false };
@@ -216,9 +217,10 @@ namespace RGL {
                     // UAV
                     auto& range = ranges.emplace_back(D3D12_DESCRIPTOR_RANGE1{
                         .RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV,
-                        .NumDescriptors = 1,
+                        .NumDescriptors = item.count,
                         .BaseShaderRegister = item.binding,
                         .RegisterSpace = 0,
+                        .Flags = D3D12_DESCRIPTOR_RANGE_FLAG_DESCRIPTORS_VOLATILE,
                         .OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND,
                     });
                     textureBindingToRootSlot[item.binding] = { uint32_t(rootParameters.size()), true };

diff --git a/src/D3D12Texture.cpp b/src/D3D12Texture.cpp
@@ -115,7 +115,7 @@ namespace RGL {
 		nativeState = D3D12_RESOURCE_STATE_COMMON;
 		if (isDS) {
 			resourceDesc.Flags |= D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL;
-			optimizedClearValue.DepthStencil = { 0,0 };
+			optimizedClearValue.DepthStencil = { config.optimizedClearValue[0], 0};
 			if (!config.usage.Sampled) {
 				nativeState |= D3D12_RESOURCE_STATE_DEPTH_WRITE;
 			}

diff --git a/src/RGLD3D12Common.cpp b/src/RGLD3D12Common.cpp
@@ -11,7 +11,7 @@
 #include <atlbase.h>
 #include <dxcapi.h>
 
-#define DX12_USE_AGILITY 0
+#define DX12_USE_AGILITY 1
 
 // Exports for the Agility SDK. For Windows 10 users, Go here: https://www.nuget.org/packages/Microsoft.Direct3D.D3D12/1.614.0 then unzip it, and place 
 // D3D12Core.dll and d3d12SDKLayers.dll in a folder named D3D12 next to the executable.

diff --git a/src/VkCommandBuffer.cpp b/src/VkCommandBuffer.cpp
@@ -234,6 +234,12 @@ namespace RGL {
 	{
 		SetFragmentTexture(texture, index);
 	}
+
+	void CommandBufferVk::MakeResident(const TextureView& texture, const MakeResidentConfig& config)
+	{
+
+	}
+
 	void CommandBufferVk::Draw(uint32_t nVertices, const DrawInstancedConfig& config)
 	{
 		EncodeCommand(CmdDraw{ nVertices, config });

diff --git a/src/VkCommandBuffer.hpp b/src/VkCommandBuffer.hpp
@@ -251,6 +251,8 @@ namespace RGL {
 		void SetFragmentTexture(const TextureView& texture, uint32_t index) final;
 		void SetComputeTexture(const TextureView& texture, uint32_t index) final;
 
+		void MakeResident(const TextureView& texture, const MakeResidentConfig& config) final;
+
 		void Draw(uint32_t nVertices, const DrawInstancedConfig & = {}) final;
 		void DrawIndexed(uint32_t nIndices, const DrawIndexedInstancedConfig & = {}) final;
 

diff --git a/src/VkDevice.cpp b/src/VkDevice.cpp
@@ -388,6 +388,11 @@ namespace RGL {
         };
     }
 
+    TextureView DeviceVk::GetGlobalBindlessTextureHeap() const
+    {
+        return TextureView();
+    }
+
     RGLCommandQueuePtr DeviceVk::CreateCommandQueue(QueueType type)
     {
         return std::make_shared<CommandQueueVk>(shared_from_this());    // vulkan does not use the queue type

diff --git a/src/VkDevice.hpp b/src/VkDevice.hpp
@@ -57,6 +57,8 @@ namespace RGL {
 
 		DeviceData GetDeviceData() final;
 
+		TextureView GetGlobalBindlessTextureHeap() const final;
+
 		RGLCommandQueuePtr CreateCommandQueue(QueueType type) final;
 		RGLFencePtr CreateFence(bool preSignaled) final;
 		void BlockUntilIdle() final;