Added support for hardware queue class on passes (#7339)

* Added support for hardware queue class on passes. Compute pass data and copy pass data can now specify to run on their appropriate queue.

Signed-off-by: antonmic <56370189+antonmic@users.noreply.github.com>

* addressed PR feedaback

Signed-off-by: antonmic <56370189+antonmic@users.noreply.github.com>

* fixing python test failures on AR

Signed-off-by: antonmic <56370189+antonmic@users.noreply.github.com>

Gems/Atom/Feature/Common/Assets/Passes/SsaoCompute.pass

@@ -50,7 +50,8 @@
                     "FilePath": "Shaders/PostProcessing/SsaoCompute.shader"
                 },
                 "Make Fullscreen Pass": true,
-                "PipelineViewTag": "MainCamera"
+                "PipelineViewTag": "MainCamera",
+                "Use Async Compute":  true
             },
             "FallbackConnections": [
                 {

Gems/Atom/Feature/Common/Code/Source/DiffuseGlobalIllumination/DiffuseProbeGridVisualizationAccelerationStructurePass.cpp

@@ -72,7 +72,7 @@ namespace AZ
 
         void DiffuseProbeGridVisualizationAccelerationStructurePass::BuildInternal()
         {
-            SetScopeId(RHI::ScopeId(GetPathName()));
+            InitScope(RHI::ScopeId(GetPathName()));
         }
 
         void DiffuseProbeGridVisualizationAccelerationStructurePass::FrameBeginInternal(FramePrepareParams params)

Gems/Atom/Feature/Common/Code/Source/PostProcessing/DepthOfFieldCopyFocusDepthToCpuPass.cpp

@@ -50,7 +50,7 @@ namespace AZ
 
         void DepthOfFieldCopyFocusDepthToCpuPass::BuildInternal()
         {
-            SetScopeId(RHI::ScopeId(GetPathName()));
+            InitScope(RHI::ScopeId(GetPathName()));
         }
 
         void DepthOfFieldCopyFocusDepthToCpuPass::FrameBeginInternal(FramePrepareParams params)

Gems/Atom/Feature/Common/Code/Source/RayTracing/RayTracingAccelerationStructurePass.cpp

@@ -40,7 +40,7 @@ namespace AZ
 
         void RayTracingAccelerationStructurePass::BuildInternal()
         {
-            SetScopeId(RHI::ScopeId(GetPathName()));
+            InitScope(RHI::ScopeId(GetPathName()));
         }
 
         void RayTracingAccelerationStructurePass::FrameBeginInternal(FramePrepareParams params)

Gems/Atom/RHI/Code/Include/Atom/RHI/Scope.h

@@ -76,6 +76,9 @@ namespace AZ
             /// Returns the hardware queue class for this scope.
             HardwareQueueClass GetHardwareQueueClass() const;
 
+            /// Sets the hardware queue class for this scope.
+            void SetHardwareQueueClass(HardwareQueueClass hardwareQueueClass);
+
             /**
              * Returns the estimated number of draw / dispatch / copy items that the user will submit
              * while in this scope. This is an estimation intended to be used by the platform-specific
@@ -120,7 +123,7 @@ namespace AZ
             const AZStd::vector<Ptr<Fence>>& GetFencesToSignal() const;
 
             /// Initializes the scope.
-            void Init(const ScopeId& scopeId);
+            void Init(const ScopeId& scopeId, HardwareQueueClass hardwareQueueClass = HardwareQueueClass::Graphics);
 
             /// Activates the scope for the current frame.
             void Activate(const FrameGraph* frameGraph, uint32_t index, const GraphGroupId& groupId);

Gems/Atom/RHI/Code/Include/Atom/RHI/ScopeProducer.h

@@ -86,11 +86,22 @@ namespace AZ
             ScopeProducer();
 
             /**
-             *  Sets ID of the scope producer. Used by class that inherit from
+             *  Sets the HardwareQueueClass on the scope
-             *  ScopeProducer but that can't supply a ScopeId at construction.
+             */
+            void SetHardwareQueueClass(HardwareQueueClass hardwareQueueClass);
+
+            /**
+             *  DEPRECATED.
+             *  @deprecated Use InitScope instead
              */
             void SetScopeId(const ScopeId& scopeId);
 
+            /**
+             *  Initializes the scope with a ScopeId and HardwareQueueClass. 
+             *  Used by classes that inherit from ScopeProducer but can't supply a ScopeId at construction.
+             */
+            void InitScope(const ScopeId& scopeId, HardwareQueueClass hardwareQueueClass = HardwareQueueClass::Graphics);
+
         private:
             //////////////////////////////////////////////////////////////////////////
             // User Overrides - Derived classes should override from these methods.

Gems/Atom/RHI/Code/Source/RHI/Scope.cpp

@@ -24,12 +24,13 @@ namespace AZ
             return m_isActive;
         }
 
-        void Scope::Init(const ScopeId& scopeId)
+        void Scope::Init(const ScopeId& scopeId, HardwareQueueClass hardwareQueueClass)
         {
             AZ_Assert(!scopeId.IsEmpty(), "Scope id is not valid.");
             AZ_Assert(IsInitialized() == false, "Scope was previously initialized.");
             SetName(scopeId);
             m_id = scopeId;
+            m_hardwareQueueClass = hardwareQueueClass;
             InitInternal();
             m_isInitialized = true;
         }
@@ -134,6 +135,11 @@ namespace AZ
             return m_hardwareQueueClass;
         }
 
+        void Scope::SetHardwareQueueClass(HardwareQueueClass hardwareQueueClass)
+        {
+            m_hardwareQueueClass = hardwareQueueClass;
+        }
+
         uint32_t Scope::GetEstimatedItemCount() const
         {
             return m_estimatedItemCount;

Gems/Atom/RHI/Code/Source/RHI/ScopeProducer.cpp

@@ -40,7 +40,18 @@ namespace AZ
             return m_scope.get();
         }
 
+        void ScopeProducer::SetHardwareQueueClass(HardwareQueueClass hardwareQueueClass)
+        {
+            m_scope->SetHardwareQueueClass(hardwareQueueClass);
+        }
+
+        // DEPRECATED: use InitScope instead
         void ScopeProducer::SetScopeId(const ScopeId& scopeId)
+        {
+            InitScope(scopeId);
+        }
+
+        void ScopeProducer::InitScope(const ScopeId& scopeId, HardwareQueueClass hardwareQueueClass)
         {
             m_scopeId = scopeId;
             
@@ -49,7 +60,7 @@ namespace AZ
                 m_scope->Shutdown();
             }
             
-            m_scope->Init(scopeId);
+            m_scope->Init(scopeId, hardwareQueueClass);
         }
     }
 }

Gems/Atom/RPI/Code/Include/Atom/RPI.Public/Pass/RenderPass.h

@@ -113,6 +113,9 @@ namespace AZ
             // The shader resource group for this pass
             Data::Instance<ShaderResourceGroup> m_shaderResourceGroup = nullptr;
 
+            // Determines which hardware queue the pass will run on
+            RHI::HardwareQueueClass m_hardwareQueueClass = RHI::HardwareQueueClass::Graphics;
+
         private:
             // Helper function that binds a single attachment to the pass shader resource group
             void BindAttachment(const RHI::FrameGraphCompileContext& context, PassAttachmentBinding& binding, int16_t& imageIndex, int16_t& bufferIndex);

Gems/Atom/RPI/Code/Include/Atom/RPI.Reflect/Pass/ComputePassData.h

@@ -29,12 +29,13 @@ namespace AZ
                 if (auto* serializeContext = azrtti_cast<SerializeContext*>(context))
                 {
                     serializeContext->Class<ComputePassData, RenderPassData>()
-                        ->Version(1)
+                        ->Version(2)
                         ->Field("ShaderAsset", &ComputePassData::m_shaderReference)
                         ->Field("Target Thread Count X", &ComputePassData::m_totalNumberOfThreadsX)
                         ->Field("Target Thread Count Y", &ComputePassData::m_totalNumberOfThreadsY)
                         ->Field("Target Thread Count Z", &ComputePassData::m_totalNumberOfThreadsZ)
                         ->Field("Make Fullscreen Pass", &ComputePassData::m_makeFullscreenPass)
+                        ->Field("Use Async Compute", &ComputePassData::m_useAsyncCompute)
                         ;
                 }
             }
@@ -46,6 +47,9 @@ namespace AZ
             uint32_t m_totalNumberOfThreadsZ = 0;
 
             bool m_makeFullscreenPass = false;
+
+            // Whether the pass should use async compute and run on the compute hardware queue.
+            bool m_useAsyncCompute = false;
         };
     } // namespace RPI
 } // namespace AZ

Gems/Atom/RPI/Code/Include/Atom/RPI.Reflect/Pass/CopyPassData.h

@@ -32,7 +32,7 @@ namespace AZ
                 if (auto* serializeContext = azrtti_cast<SerializeContext*>(context))
                 {
                     serializeContext->Class<CopyPassData, PassData>()
-                        ->Version(0)
+                        ->Version(1)
                         ->Field("BufferSize", &CopyPassData::m_bufferSize)
                         ->Field("BufferSourceOffset", &CopyPassData::m_bufferSourceOffset)
                         ->Field("BufferSourceBytesPerRow", &CopyPassData::m_bufferSourceBytesPerRow)
@@ -46,6 +46,7 @@ namespace AZ
                         ->Field("ImageDestinationSubresource", &CopyPassData::m_imageDestinationSubresource)
                         ->Field("ImageDestinationOrigin", &CopyPassData::m_imageDestinationOrigin)
                         ->Field("CloneInput", &CopyPassData::m_cloneInput)
+                        ->Field("UseCopyQueue", &CopyPassData::m_useCopyQueue)
                         ;
                 }
             }
@@ -75,6 +76,9 @@ namespace AZ
             // If set to true, pass will automatically create a transient output attachment based on input
             // If false, the output target of the copy will need to be specified
             bool m_cloneInput = true;
+
+            // Whether the pass should use the copy queue.
+            bool m_useCopyQueue = false;
         };
     } // namespace RPI
 } // namespace AZ

Gems/Atom/RPI/Code/Source/RPI.Public/Pass/ComputePass.cpp

@@ -58,6 +58,12 @@ namespace AZ
                 return;
             }
 
+            // Hardware Queue Class
+            if (passData->m_useAsyncCompute)
+            {
+                m_hardwareQueueClass = RHI::HardwareQueueClass::Compute;
+            }
+
             // Load Shader
             Data::Asset<ShaderAsset> shaderAsset;
             if (passData->m_shaderReference.m_assetId.IsValid())

Gems/Atom/RPI/Code/Source/RPI.Public/Pass/CopyPass.cpp

@@ -38,6 +38,11 @@ namespace AZ
             if (copyData)
             {
                 m_data = *copyData;
+
+                if (copyData->m_useCopyQueue)
+                {
+                    m_hardwareQueueClass = RHI::HardwareQueueClass::Copy;
+                }
             }
         }
 

Gems/Atom/RPI/Code/Source/RPI.Public/Pass/RenderPass.cpp

@@ -183,7 +183,7 @@ namespace AZ
         {
             if (GetScopeId().IsEmpty())
             {
-                SetScopeId(RHI::ScopeId(GetPathName()));
+                InitScope(RHI::ScopeId(GetPathName()), m_hardwareQueueClass);
             }
 
             params.m_frameGraphBuilder->ImportScopeProducer(*this);

Gems/Atom/RPI/Code/Source/RPI.Public/Pass/Specific/ImageAttachmentPreviewPass.cpp

@@ -32,7 +32,7 @@ namespace AZ
             m_destAttachmentId = destAttachmentId;
 
             // Use the unique destination attachment id as scope id
-            SetScopeId(m_destAttachmentId);
+            InitScope(m_destAttachmentId);
 
             // Clear the previous attachment and copy item
             m_copyItem = {};
@@ -124,7 +124,7 @@ namespace AZ
         ImageAttachmentPreviewPass::ImageAttachmentPreviewPass(const PassDescriptor& descriptor)
             : Pass(descriptor)
         {
-            SetScopeId(RHI::ScopeId(GetPathName()));
+            InitScope(RHI::ScopeId(GetPathName()));
         }
 
         ImageAttachmentPreviewPass::~ImageAttachmentPreviewPass()