Add support for rendering into 3d textures on Vulkan

#136

Source/Engine/Graphics/Materials/MaterialShader.cpp

@@ -145,6 +145,11 @@ MaterialShader* MaterialShader::CreateDummy(MemoryReadStream& shaderCacheStream,
     return material;
 }
 
+GPUShader* MaterialShader::GetShader() const
+{
+    return _shader;
+}
+
 const MaterialInfo& MaterialShader::GetInfo() const
 {
     return _info;

Source/Engine/Graphics/Materials/MaterialShader.h

@@ -100,6 +100,8 @@ public:
     /// <returns>The created and loaded material or null if failed.</returns>
     static MaterialShader* CreateDummy(MemoryReadStream& shaderCacheStream, const MaterialInfo& info);
 
+    GPUShader* GetShader() const;
+
     /// <summary>
     /// Clears the loaded data.
     /// </summary>

Source/Engine/Graphics/Shaders/GPUShader.cpp

@@ -82,6 +82,9 @@ bool GPUShader::Create(MemoryReadStream& stream)
     int32 shadersCount;
     stream.ReadInt32(&shadersCount);
     GPUShaderProgramInitializer initializer;
+#if !BUILD_RELEASE
+    initializer.Owner = this;
+#endif
     for (int32 i = 0; i < shadersCount; i++)
     {
         const ShaderStage type = static_cast<ShaderStage>(stream.ReadByte());

Source/Engine/Graphics/Shaders/GPUShaderProgram.h

@@ -6,6 +6,8 @@
 #include "Engine/Core/Types/String.h"
 #include "Config.h"
 
+class GPUShader;
+
 /// <summary>
 /// The shader program metadata container. Contains description about resources used by the shader.
 /// </summary>
@@ -37,6 +39,9 @@ struct GPUShaderProgramInitializer
     StringAnsi Name;
     ShaderBindings Bindings;
     ShaderFlags Flags;
+#if !BUILD_RELEASE
+    GPUShader* Owner;
+#endif
 };
 
 /// <summary>
@@ -49,12 +54,18 @@ protected:
     StringAnsi _name;
     ShaderBindings _bindings;
     ShaderFlags _flags;
+#if !BUILD_RELEASE
+    GPUShader* _owner;
+#endif
 
     void Init(const GPUShaderProgramInitializer& initializer)
     {
         _name = initializer.Name;
         _bindings = initializer.Bindings;
         _flags = initializer.Flags;
+#if !BUILD_RELEASE
+        _owner = initializer.Owner;
+#endif
     }
 
 public:
@@ -69,9 +80,8 @@ public:
 public:
 
     /// <summary>
-    /// Gets name of the shader program
+    /// Gets name of the shader program.
     /// </summary>
-    /// <returns>Name</returns>
     FORCE_INLINE const StringAnsi& GetName() const
     {
         return _name;
@@ -80,7 +90,6 @@ public:
     /// <summary>
     /// Gets the shader resource bindings.
     /// </summary>
-    /// <returns>The bindings.</returns>
     FORCE_INLINE const ShaderBindings& GetBindings() const
     {
         return _bindings;
@@ -89,7 +98,6 @@ public:
     /// <summary>
     /// Gets the shader flags.
     /// </summary>
-    /// <returns>The flags.</returns>
     FORCE_INLINE ShaderFlags GetFlags() const
     {
         return _flags;
@@ -98,21 +106,18 @@ public:
 public:
 
     /// <summary>
-    /// Gets shader program stage type
+    /// Gets shader program stage type.
     /// </summary>
-    /// <returns>Shader Stage type</returns>
     virtual ShaderStage GetStage() const = 0;
 
     /// <summary>
-    /// Gets buffer handle (platform dependent)
+    /// Gets buffer handle (platform dependent).
     /// </summary>
-    /// <returns>Handle</returns>
     virtual void* GetBufferHandle() const = 0;
 
     /// <summary>
-    /// Gets buffer size (in bytes)
+    /// Gets buffer size (in bytes).
     /// </summary>
-    /// <returns>Size of the buffer in bytes</returns>
     virtual uint32 GetBufferSize() const = 0;
 };
 
@@ -124,15 +129,13 @@ class GPUShaderProgramVS : public GPUShaderProgram
 public:
 
     /// <summary>
-    /// Gets input layout description handle (platform dependent)
+    /// Gets input layout description handle (platform dependent).
     /// </summary>
-    /// <returns>Input layout</returns>
     virtual void* GetInputLayout() const = 0;
 
     /// <summary>
-    /// Gets input layout description size (in bytes)
+    /// Gets input layout description size (in bytes).
     /// </summary>
-    /// <returns>Input layout description size in bytes</returns>
     virtual byte GetInputLayoutSize() const = 0;
 
 public:

Source/Engine/GraphicsDevice/Vulkan/GPUContextVulkan.cpp

@@ -346,7 +346,6 @@ void GPUContextVulkan::BeginRenderPass()
         {
             layout.RTVsFormats[i] = handle->GetFormat();
             framebufferKey.Attachments[i] = handle->GetFramebufferView();
-
             AddImageBarrier(handle, handle->LayoutRTV);
         }
         else
@@ -355,36 +354,31 @@ void GPUContextVulkan::BeginRenderPass()
             framebufferKey.Attachments[i] = VK_NULL_HANDLE;
         }
     }
+    GPUTextureViewVulkan* handle;
     if (_rtDepth)
     {
-        auto handle = _rtDepth;
+        handle = _rtDepth;
-        layout.MSAA = handle->GetMSAA();
-        layout.Extent = handle->Extent;
         layout.ReadDepth = true; // TODO: use proper depthStencilAccess flags
         layout.WriteDepth = handle->LayoutRTV == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; // TODO: do it in a proper way
         framebufferKey.AttachmentCount++;
         framebufferKey.Attachments[_rtCount] = handle->GetFramebufferView();
-
         AddImageBarrier(handle, handle->LayoutRTV);
     }
-    else if (_rtHandles[0])
-    {
-        layout.MSAA = _rtHandles[0]->GetMSAA();
-        layout.Extent = _rtHandles[0]->Extent;
-        layout.ReadDepth = false;
-        layout.WriteDepth = false;
-    }
     else
     {
-        // No depth or render target binded?
+        handle = _rtHandles[0];
-        CRASH;
+        layout.ReadDepth = false;
+        layout.WriteDepth = false;
     }
+    layout.MSAA = handle->GetMSAA();
+    layout.Extent.width = handle->Extent.width;
+    layout.Extent.height = handle->Extent.height;
+    layout.Layers = handle->Layers;
 
     // Get or create objects
     auto renderPass = _device->GetOrCreateRenderPass(layout);
     framebufferKey.RenderPass = renderPass;
-    uint32 layers = 1; // TODO: support rendering to many layers (eg. texture array)
+    auto framebuffer = _device->GetOrCreateFramebuffer(framebufferKey, layout.Extent, layout.Layers);
-    auto framebuffer = _device->GetOrCreateFramebuffer(framebufferKey, layout.Extent, layers);
     _renderPass = renderPass;
 
     FlushBarriers();
@@ -519,7 +513,7 @@ void GPUContextVulkan::UpdateDescriptorSets(const SpirvShaderDescriptorInfo& des
         }
         default:
             // Unknown or invalid descriptor type
-        CRASH;
+            CRASH;
             break;
         }
     }

Source/Engine/GraphicsDevice/Vulkan/GPUDeviceVulkan.cpp

@@ -61,7 +61,6 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugReportFunction(VkDebugReportFlagsEXT m
     if (msgFlags & VK_DEBUG_REPORT_ERROR_BIT_EXT)
     {
         msgPrefix = TEXT("ERROR");
-
         if (!StringUtils::Compare(layerPrefix, "SC"))
         {
             if (msgCode == 3)
@@ -79,7 +78,6 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugReportFunction(VkDebugReportFlagsEXT m
     else if (msgFlags & VK_DEBUG_REPORT_WARNING_BIT_EXT)
     {
         msgPrefix = TEXT("WARN");
-
         if (!StringUtils::Compare(layerPrefix, "SC"))
         {
             if (msgCode == 2)
@@ -92,7 +90,6 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugReportFunction(VkDebugReportFlagsEXT m
     else if (msgFlags & VK_DEBUG_REPORT_PERFORMANCE_WARNING_BIT_EXT)
     {
         msgPrefix = TEXT("PERF");
-
         if (!StringUtils::Compare(layerPrefix, "SC"))
         {
             if (msgCode == 2)
@@ -118,14 +115,8 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugReportFunction(VkDebugReportFlagsEXT m
     {
         msgPrefix = TEXT("DEBUG");
     }
-    else
-    {
-        CRASH;
-    }
 
-    // Send info
     LOG(Info, "[Vulkan] {0}:{1}:{2} {3}", msgPrefix, String(layerPrefix), msgCode, String(msg));
-
     return VK_FALSE;
 }
 
@@ -147,6 +138,9 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugUtilsCallback(VkDebugUtilsMessageSever
         case 5: // SPIR-V module not valid: MemoryBarrier: Vulkan specification requires Memory Semantics to have one of the following bits set: Acquire, Release, AcquireRelease or SequentiallyConsistent
         case -1666394502: // After query pool creation, each query must be reset before it is used. Queries must also be reset between uses.
         case 602160055: // Attachment 4 not written by fragment shader; undefined values will be written to attachment. TODO: investigate it for PS_GBuffer shader from Deferred material with USE_LIGHTMAP=1
+        case 7060244: //  Image Operand Offset can only be used with OpImage*Gather operations
+        case -1539028524: // SortedIndices is null so Vulkan backend sets it to default R32_SFLOAT format which is not good for UINT format of the buffer
+        case -1810835948: // SortedIndices is null so Vulkan backend sets it to default R32_SFLOAT format which is not good for UINT format of the buffer
             return VK_FALSE;
         }
         break;
@@ -214,15 +208,10 @@ static VKAPI_ATTR VkBool32 VKAPI_PTR DebugUtilsCallback(VkDebugUtilsMessageSever
         type = TEXT("Perf");
     }
 
-    // Send info
     if (callbackData->pMessageIdName)
-    {
         LOG(Info, "[Vulkan] {0} {1}:{2}({3}) {4}", type, severity, callbackData->messageIdNumber, String(callbackData->pMessageIdName), String(callbackData->pMessage));
-    }
     else
-    {
         LOG(Info, "[Vulkan] {0} {1}:{2} {3}", type, severity, callbackData->messageIdNumber, String(callbackData->pMessage));
-    }
 
     return VK_FALSE;
 }
@@ -440,10 +429,11 @@ uint32 GetHash(const RenderTargetLayoutVulkan& key)
     CombineHash(hash, (uint32)key.WriteDepth);
     CombineHash(hash, (uint32)key.DepthFormat * 93473262);
     CombineHash(hash, key.RTsCount * 136);
+    CombineHash(hash, key.Extent.width);
+    CombineHash(hash, key.Extent.height);
+    CombineHash(hash, key.Layers);
     for (int32 i = 0; i < ARRAY_COUNT(key.RTVsFormats); i++)
-    {
         CombineHash(hash, (uint32)key.RTVsFormats[i]);
-    }
     return hash;
 }
 
@@ -452,13 +442,11 @@ uint32 GetHash(const FramebufferVulkan::Key& key)
     uint32 hash = (int32)(intptr)key.RenderPass;
     CombineHash(hash, (uint32)key.AttachmentCount * 136);
     for (int32 i = 0; i < ARRAY_COUNT(key.Attachments); i++)
-    {
         CombineHash(hash, (uint32)(intptr)key.Attachments[i]);
-    }
     return hash;
 }
 
-FramebufferVulkan::FramebufferVulkan(GPUDeviceVulkan* device, Key& key, VkExtent3D& extent, uint32 layers)
+FramebufferVulkan::FramebufferVulkan(GPUDeviceVulkan* device, Key& key, VkExtent2D& extent, uint32 layers)
     : _device(device)
     , _handle(VK_NULL_HANDLE)
     , Extent(extent)
@@ -1271,33 +1259,23 @@ GPUDeviceVulkan::~GPUDeviceVulkan()
 
 RenderPassVulkan* GPUDeviceVulkan::GetOrCreateRenderPass(RenderTargetLayoutVulkan& layout)
 {
-    // Try reuse cached version
     RenderPassVulkan* renderPass;
     if (_renderPasses.TryGet(layout, renderPass))
-    {
         return renderPass;
-    }
 
     PROFILE_CPU_NAMED("Create Render Pass");
-
-    // Create object and cache it
     renderPass = New<RenderPassVulkan>(this, layout);
     _renderPasses.Add(layout, renderPass);
     return renderPass;
 }
 
-FramebufferVulkan* GPUDeviceVulkan::GetOrCreateFramebuffer(FramebufferVulkan::Key& key, VkExtent3D& extent, uint32 layers)
+FramebufferVulkan* GPUDeviceVulkan::GetOrCreateFramebuffer(FramebufferVulkan::Key& key, VkExtent2D& extent, uint32 layers)
 {
-    // Try reuse cached version
     FramebufferVulkan* framebuffer;
     if (_framebuffers.TryGet(key, framebuffer))
-    {
         return framebuffer;
-    }
 
     PROFILE_CPU_NAMED("Create Framebuffer");
-
-    // Create object and cache it
     framebuffer = New<FramebufferVulkan>(this, key, extent, layers);
     _framebuffers.Add(key, framebuffer);
     return framebuffer;
@@ -1305,16 +1283,11 @@ FramebufferVulkan* GPUDeviceVulkan::GetOrCreateFramebuffer(FramebufferVulkan::Ke
 
 PipelineLayoutVulkan* GPUDeviceVulkan::GetOrCreateLayout(DescriptorSetLayoutInfoVulkan& key)
 {
-    // Try reuse cached version
     PipelineLayoutVulkan* layout;
     if (_layouts.TryGet(key, layout))
-    {
         return layout;
-    }
 
     PROFILE_CPU_NAMED("Create Pipeline Layout");
-
-    // Create object and cache it
     layout = New<PipelineLayoutVulkan>(this, key);
     _layouts.Add(key, layout);
     return layout;
@@ -1698,7 +1671,11 @@ bool GPUDeviceVulkan::Init()
         auto& limits = Limits;
         limits.HasCompute = GetShaderProfile() == ShaderProfile::Vulkan_SM5 && PhysicalDeviceLimits.maxComputeWorkGroupCount[0] >= GPU_MAX_CS_DISPATCH_THREAD_GROUPS && PhysicalDeviceLimits.maxComputeWorkGroupCount[1] >= GPU_MAX_CS_DISPATCH_THREAD_GROUPS;
         limits.HasTessellation = !!PhysicalDeviceFeatures.tessellationShader && PhysicalDeviceLimits.maxBoundDescriptorSets > (uint32_t)DescriptorSet::Domain;
-        limits.HasGeometryShaders = false; // TODO: add geometry shaders support for Vulkan
+#if PLATFORM_ANDROID
+        limits.HasGeometryShaders = false; // Don't even try GS on mobile
+#else
+        limits.HasGeometryShaders = !!PhysicalDeviceFeatures.geometryShader;
+#endif
         limits.HasInstancing = true;
         limits.HasVolumeTextureRendering = true;
         limits.HasDrawIndirect = PhysicalDeviceLimits.maxDrawIndirectCount >= 1;

Source/Engine/GraphicsDevice/Vulkan/GPUDeviceVulkan.h

@@ -237,13 +237,14 @@ public:
     bool WriteDepth;
     PixelFormat DepthFormat;
     PixelFormat RTVsFormats[GPU_MAX_RT_BINDED];
-    VkExtent3D Extent;
+    VkExtent2D Extent;
+    uint32 Layers;
 
 public:
 
     bool operator==(const RenderTargetLayoutVulkan& other) const
     {
-        return Platform::MemoryCompare((void*)this, &other, sizeof(RenderTargetLayoutVulkan)) == 0;
+        return Platform::MemoryCompare(this, &other, sizeof(RenderTargetLayoutVulkan)) == 0;
     }
 };
 
@@ -263,7 +264,7 @@ public:
 
         bool operator==(const Key& other) const
         {
-            return Platform::MemoryCompare((void*)this, &other, sizeof(Key)) == 0;
+            return Platform::MemoryCompare(this, &other, sizeof(Key)) == 0;
         }
     };
 
@@ -274,13 +275,13 @@ private:
 
 public:
 
-    FramebufferVulkan(GPUDeviceVulkan* device, Key& key, VkExtent3D& extent, uint32 layers);
+    FramebufferVulkan(GPUDeviceVulkan* device, Key& key, VkExtent2D& extent, uint32 layers);
     ~FramebufferVulkan();
 
 public:
 
     VkImageView Attachments[GPU_MAX_RT_BINDED + 1];
-    VkExtent3D Extent;
+    VkExtent2D Extent;
     uint32 Layers;
 
 public:
@@ -498,8 +499,6 @@ private:
 
 public:
 
-    // Create new graphics device (returns Vulkan if failed)
-    // @returns Created device or Vulkan
     static GPUDevice* Create();
 
     /// <summary>
@@ -685,7 +684,7 @@ public:
     }
 
     RenderPassVulkan* GetOrCreateRenderPass(RenderTargetLayoutVulkan& layout);
-    FramebufferVulkan* GetOrCreateFramebuffer(FramebufferVulkan::Key& key, VkExtent3D& extent, uint32 layers);
+    FramebufferVulkan* GetOrCreateFramebuffer(FramebufferVulkan::Key& key, VkExtent2D& extent, uint32 layers);
     PipelineLayoutVulkan* GetOrCreateLayout(DescriptorSetLayoutInfoVulkan& key);
     void OnImageViewDestroy(VkImageView imageView);
 

Source/Engine/GraphicsDevice/Vulkan/GPUTextureVulkan.cpp

@@ -21,6 +21,7 @@ void GPUTextureViewVulkan::Init(GPUDeviceVulkan* device, ResourceOwnerVulkan* ow
     Extent.width = Math::Max<uint32_t>(1, extent.width >> firstMipIndex);
     Extent.height = Math::Max<uint32_t>(1, extent.height >> firstMipIndex);
     Extent.depth = Math::Max<uint32_t>(1, extent.depth >> firstMipIndex);
+    Layers = arraySize;
 
     RenderToolsVulkan::ZeroStruct(Info, VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO);
     Info.image = image;
@@ -74,16 +75,30 @@ VkImageView GPUTextureViewVulkan::GetFramebufferView()
     if (ViewFramebuffer)
         return ViewFramebuffer;
 
-    if (Info.subresourceRange.levelCount == 1)
+    if (Info.viewType == VK_IMAGE_VIEW_TYPE_3D)
-        return View;
+    {
-
+        // Special case:
-    // Special case:
+        // Render Target Handle to a 3D Volume texture.
-    // Render Target Handle can be created for full texture including its mip maps but framebuffer image view can use only a single surface
+        // Use it as Texture2D Array with layers.
-    // Use an additional view for that case with modified level count to 1.
+        VkImageViewCreateInfo createInfo = Info;
-
+        createInfo.viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY;
-    VkImageViewCreateInfo createInfo = Info;
+        createInfo.subresourceRange.layerCount = Extent.depth;
-    createInfo.subresourceRange.levelCount = 1;
+        Layers = Extent.depth;
-    VALIDATE_VULKAN_RESULT(vkCreateImageView(Device->Device, &createInfo, nullptr, &ViewFramebuffer));
+        VALIDATE_VULKAN_RESULT(vkCreateImageView(Device->Device, &createInfo, nullptr, &ViewFramebuffer));
+    }
+    else if (Info.subresourceRange.levelCount != 1)
+    {
+        // Special case:
+        // Render Target Handle can be created for full texture including its mip maps but framebuffer image view can use only a single surface
+        // Use an additional view for that case with modified level count to 1.
+        VkImageViewCreateInfo createInfo = Info;
+        createInfo.subresourceRange.levelCount = 1;
+        VALIDATE_VULKAN_RESULT(vkCreateImageView(Device->Device, &createInfo, nullptr, &ViewFramebuffer));
+    }
+    else
+    {
+        ViewFramebuffer = View;
+    }
 
     return ViewFramebuffer;
 }
@@ -92,15 +107,15 @@ void GPUTextureViewVulkan::Release()
 {
     if (View != VK_NULL_HANDLE)
     {
-        Device->OnImageViewDestroy(View);
+        if (ViewFramebuffer != View && ViewFramebuffer != VK_NULL_HANDLE)
-        Device->DeferredDeletionQueue.EnqueueResource(DeferredDeletionQueueVulkan::Type::ImageView, View);
-
-        if (ViewFramebuffer != VK_NULL_HANDLE)
         {
             Device->OnImageViewDestroy(ViewFramebuffer);
             Device->DeferredDeletionQueue.EnqueueResource(DeferredDeletionQueueVulkan::Type::ImageView, ViewFramebuffer);
         }
 
+        Device->OnImageViewDestroy(View);
+        Device->DeferredDeletionQueue.EnqueueResource(DeferredDeletionQueueVulkan::Type::ImageView, View);
+
         View = VK_NULL_HANDLE;
         ViewFramebuffer = VK_NULL_HANDLE;
 
@@ -236,32 +251,20 @@ bool GPUTextureVulkan::OnInit()
     imageInfo.extent.depth = Depth();
     imageInfo.flags = IsCubeMap() ? VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT : 0;
     if (IsSRGB())
-    {
         imageInfo.flags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
-    }
 #if VK_KHR_maintenance1
     if (_device->OptionalDeviceExtensions.HasKHRMaintenance1 && imageInfo.imageType == VK_IMAGE_TYPE_3D)
-    {
         imageInfo.flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT_KHR;
-    }
 #endif
     imageInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
     if (useSRV)
-    {
         imageInfo.usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
-    }
     if (useDSV)
-    {
         imageInfo.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
-    }
     if (useRTV)
-    {
         imageInfo.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
-    }
     if (useUAV)
-    {
         imageInfo.usage |= VK_IMAGE_USAGE_STORAGE_BIT;
-    }
     imageInfo.tiling = optimalTiling ? VK_IMAGE_TILING_OPTIMAL : VK_IMAGE_TILING_LINEAR;
     imageInfo.samples = (VkSampleCountFlagBits)MultiSampleLevel();
     // TODO: set initialLayout to VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL for IsRegularTexture() ???

Source/Engine/GraphicsDevice/Vulkan/GPUTextureVulkan.h

@@ -51,6 +51,7 @@ public:
     VkImageView View = VK_NULL_HANDLE;
     VkImageView ViewFramebuffer = VK_NULL_HANDLE;
     VkExtent3D Extent;
+    uint32 Layers;
     VkImageViewCreateInfo Info;
     int32 SubresourceIndex;
     VkImageLayout LayoutRTV;

Source/Engine/Renderer/RenderList.cpp

@@ -695,7 +695,7 @@ DRAW:
             if (drawCall.InstanceCount == 0)
             {
                 // No support for batching indirect draw calls
-                ASSERT(batch.BatchSize == 1);
+                ASSERT_LOW_LAYER(batch.BatchSize == 1);
 
                 context->BindVB(ToSpan(vb, vbCount), vbOffsets);
                 context->DrawIndexedInstancedIndirect(drawCall.Draw.IndirectArgsBuffer, drawCall.Draw.IndirectArgsOffset);