Refactor SkinnedMeshFeatureProcessor (DCO fixed) (#9171)

* Squash skinned mesh dispatch changes (#299) Squashing several changes from a staging branch into a single commit to get rid of some bad history. Each of these changes was individually reviewed: One skinning dispatch per mesh: Part 1 aws-lumberyard-dev#286 Removing support for morph targets that modify vertex colors aws-lumberyard-dev#288 One SkinnedMesh Dispatch Per Mesh Part 2: Morph targets aws-lumberyard-dev#294 Support for more or less than 4 influences per vertex aws-lumberyard-dev#296 Skip skinning cloth meshes aws-lumberyard-dev#297 Support varying influence counts per mesh aws-lumberyard-dev#298 * WIP use oned dispatch per mesh instead of per lod for skinning. Need to update Actor.cpp to account for the new padding in the bone Id's Signed-off-by: Tommy Walton <waltont@amazon.com> * Attempt to fix JointId buffer Signed-off-by: Tommy Walton <waltont@amazon.com> * WIP use oned dispatch per mesh instead of per lod for skinning. Need to update Actor.cpp to account for the new padding in the bone Id's Signed-off-by: Tommy Walton <waltont@amazon.com> * Attempt to fix JointId buffer Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix compile errors Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix Material->MaterialAsset name mixup Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix 3-component buffer views to work with metal Signed-off-by: Tommy Walton <waltont@amazon.com> * Fixing position history writing to the incorrect offset Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix JointId alignment issue Signed-off-by: Tommy Walton <waltont@amazon.com> * Moved jointId padding logic into a separate file that can be shared between EMFX and the ModelAssetBuilderComponent. Also simplified the loop for doing the padding in EMFX. Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix incorrect variable name after merge Signed-off-by: Tommy Walton <waltont@amazon.com> * Minor cleanup Signed-off-by: Tommy Walton <waltont@amazon.com> * More minor cleanup Signed-off-by: Tommy Walton <waltont@amazon.com> * Comment updates Signed-off-by: Tommy Walton <waltont@amazon.com> * Split giant CreateFromModel function into smaller functions that should be easier to follow. Signed-off-by: Tommy Walton <waltont@amazon.com> * More minor cleanup/re-naming Signed-off-by: Tommy Walton <waltont@amazon.com> * Switch from AZ::RoundUpToMultiple to AZ::SizeAlignUp since the alignment is a power of two Signed-off-by: Tommy Walton <waltont@amazon.com> * Minor PR feedback Signed-off-by: Tommy Walton <waltont@amazon.com> * Switch from size_t to uint32_t Signed-off-by: Tommy Walton <waltont@amazon.com> * Fixing a few integer conversion related failures on non-windows platforms Signed-off-by: Tommy Walton <waltont@amazon.com> * Removing support for morph targets that modify vertex colors Signed-off-by: Tommy Walton <waltont@amazon.com> * One SkinnedMesh Dispatch Per Mesh Part 2: Morph targets (#294) * Update morph targets to work with per-mesh dispatches Signed-off-by: Tommy Walton <waltont@amazon.com> * Moving some things around to make the updated render proxy logic easier to follow Signed-off-by: Tommy Walton <waltont@amazon.com> * Store morph targets in the order they were created instead of on individual meshes Signed-off-by: Tommy Walton <waltont@amazon.com> * Minor PR feedback Signed-off-by: Tommy Walton <waltont@amazon.com> * Support for more or less than 4 influences per vertex (#296) * Initial support for more than 4 influences per vertex Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix shader compile error Signed-off-by: Tommy Walton <waltont@amazon.com> * Add a settings registry setting to modify the default max influences per vertex Signed-off-by: Tommy Walton <waltont@amazon.com> * Add a comment explaining the jointId offset Signed-off-by: Tommy Walton <waltont@amazon.com> * PR Feedback Signed-off-by: Tommy Walton <waltont@amazon.com> * Support varying influence counts per mesh (#298) * Modify ModelAssetBuilder to allow for meshes to have different per-vertex influence counts Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix merge error. Signed-off-by: Tommy Walton <waltont@amazon.com> * Update EMFX::Mesh to work with sub-meshes that have varying influence counts Signed-off-by: Tommy Walton <waltont@amazon.com> * Use the max influence count across all meshes to pre-allocate memory for the EMFX skin layer Signed-off-by: Tommy Walton <waltont@amazon.com> * Get rid of local vector in ProcessSkinInfluences to improve performance during level load Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix an incorrect comment Signed-off-by: Tommy Walton <waltont@amazon.com> * PR feedback Signed-off-by: Tommy Walton <waltont@amazon.com> * Skip skinning cloth meshes (#297) * Skip skinning dispatch entirely for certain meshes(like those with cloth) instead of modifying the blendweights to all be 0 Signed-off-by: Tommy Walton <waltont@amazon.com> * Working but somewhat hacky attempt to push the skip skinning logic to the component Signed-off-by: Tommy Walton <waltont@amazon.com> * Disable via ebus instead Signed-off-by: amzn-tommy <waltont@amazon.com> * Rename shouldSkipSkinning to isSkinningEnabled which is easier to follow with set/get methods Signed-off-by: amzn-tommy <waltont@amazon.com> * Add SkinnedMeshOverrideBus, which was missing from an earlier commit Signed-off-by: amzn-tommy <waltont@amazon.com> * Update FillSkinnedMeshInstanceBuffers to only fill buffers for meshes that won't be skinned. Signed-off-by: amzn-tommy <waltont@amazon.com> * Allow cloth to tell skinning to be enabled/disabled at any time instead of requesting it from cloth at a particular point of initialization Signed-off-by: Tommy Walton <waltont@amazon.com> * Remove function for filling buffers, since now they are either filled by simulation or filled by skinning Signed-off-by: Tommy Walton <waltont@amazon.com> * Minor PR feedback Signed-off-by: Tommy Walton <waltont@amazon.com> * Support varying influence counts per mesh (#298) * Modify ModelAssetBuilder to allow for meshes to have different per-vertex influence counts Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix merge error. Signed-off-by: Tommy Walton <waltont@amazon.com> * Update EMFX::Mesh to work with sub-meshes that have varying influence counts Signed-off-by: Tommy Walton <waltont@amazon.com> * Use the max influence count across all meshes to pre-allocate memory for the EMFX skin layer Signed-off-by: Tommy Walton <waltont@amazon.com> * Get rid of local vector in ProcessSkinInfluences to improve performance during level load Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix an incorrect comment Signed-off-by: Tommy Walton <waltont@amazon.com> * PR feedback Signed-off-by: Tommy Walton <waltont@amazon.com> Co-authored-by: Jon B <jonbold@amazon.com> Signed-off-by: Tommy Walton <waltont@amazon.com> * Redcode old SkinnedMesh interfaces (#301) * Redcode Signed-off-by: Tommy Walton <waltont@amazon.com> * More redcode Signed-off-by: Tommy Walton <waltont@amazon.com> * Even more redcode Signed-off-by: Tommy Walton <waltont@amazon.com> * Still more redcode Signed-off-by: Tommy Walton <waltont@amazon.com> * Modified the handle to be opaque to the underlying implemention, and have operations on the handle done through the feature processor interface instead of directly via the handle (#302) Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix cloth skinning for assets that have different influence counts for each sub-mesh (#313) * Fix cloth skinning for assets that have different influence counts for each sub-mesh Signed-off-by: Tommy Walton <waltont@amazon.com> * Removed check for invalid jointId, which no longer occurs and should be handled by the higher level actor initialization. Also renamed a few variables based on PR feedback. Signed-off-by: Tommy Walton <waltont@amazon.com> * Fixed a miss-spelling of the word skinning Signed-off-by: Tommy Walton <waltont@amazon.com> * Fixing a couple print statements using %zu for a variable that was changed to uint32_t Signed-off-by: Tommy Walton <waltont@amazon.com> * Fix missing commits for unity build and added spaces around PRIu32 Signed-off-by: Tommy Walton <waltont@amazon.com> * Made other uses of PRIu32 have spaces as well for consistency Signed-off-by: Tommy Walton <waltont@amazon.com> * Minor build fixes: A type mis-match between AZ::u64 and uint64_t, an unnecessary call to reserve on a fixed_vector, a missing namespace in a unit test, and an unused variable. Signed-off-by: Tommy Walton <waltont@amazon.com> * Remove unused variable Signed-off-by: Tommy Walton <waltont@amazon.com> Co-authored-by: Jon B <jonbold@amazon.com>
o3de · Apr 27, 2022 · 3ff2f29 · 3ff2f29
1 parent e125689
commit 3ff2f29
Show file tree

Hide file tree

Showing 58 changed files with 1,659 additions and 1,576 deletions.
diff --git a/Code/Tools/SceneAPI/SceneCore/DataTypes/Rules/ISkinRule.h b/Code/Tools/SceneAPI/SceneCore/DataTypes/Rules/ISkinRule.h
@@ -9,6 +9,7 @@
  */
 
 #include <AzCore/RTTI/RTTI.h>
+#include <AzCore/Settings/SettingsRegistry.h>
 #include <SceneAPI/SceneCore/DataTypes/Rules/IRule.h>
 
 namespace AZ
@@ -17,6 +18,36 @@ namespace AZ
     {
         namespace DataTypes
         {
+            inline static constexpr AZStd::string_view DefaultMaxSkinInfluencesPerVertexKey{ "/O3DE/SceneAPI/SkinRule/DefaultMaxSkinInfluencesPerVertex" };
+            inline static constexpr AZStd::string_view DefaultWeightThresholdKey{ "/O3DE/SceneAPI/SkinRule/DefaultWeightThreshold" };
+
+            struct SkinRuleSettings
+            {
+                AZ::u32 m_maxInfluencesPerVertex = 8;
+                float m_weightThreshold = 0.001;
+            };
+
+            inline SkinRuleSettings GetDefaultSkinRuleSettings()
+            {
+                SkinRuleSettings defaultSettings;
+                // Get the project wide default settings for the skin rule
+                if (auto settingsRegistry = AZ::SettingsRegistry::Get(); settingsRegistry != nullptr)
+                {
+                    AZ::u64 defaultMaxSkinInfluences = 0;
+                    if (settingsRegistry->Get(defaultMaxSkinInfluences, SceneAPI::DataTypes::DefaultMaxSkinInfluencesPerVertexKey))
+                    {
+                        defaultSettings.m_maxInfluencesPerVertex = aznumeric_caster(defaultMaxSkinInfluences);
+                    }
+
+                    double defaultWeightThreshold = 0.0f;
+                    if (settingsRegistry->Get(defaultWeightThreshold, SceneAPI::DataTypes::DefaultWeightThresholdKey))
+                    {
+                        defaultSettings.m_weightThreshold = aznumeric_caster(defaultWeightThreshold);
+                    }
+                }
+                return defaultSettings;
+            }
+
             class ISkinRule
                 : public IRule
             {

diff --git a/Code/Tools/SceneAPI/SceneData/Rules/SkinRule.cpp b/Code/Tools/SceneAPI/SceneData/Rules/SkinRule.cpp
@@ -21,10 +21,12 @@ namespace AZ
             AZ_CLASS_ALLOCATOR_IMPL(SkinRule, AZ::SystemAllocator, 0)
 
             SkinRule::SkinRule()
-                : m_maxWeightsPerVertex(4)
-                , m_weightThreshold(0.001f)
             {
+                DataTypes::SkinRuleSettings defaultSettings = DataTypes::GetDefaultSkinRuleSettings();
+                m_maxWeightsPerVertex = defaultSettings.m_maxInfluencesPerVertex;
+                m_weightThreshold = defaultSettings.m_weightThreshold;
             }
+
             AZ::u32 SkinRule::GetMaxWeightsPerVertex() const
             {
                 return m_maxWeightsPerVertex;

diff --git a/Gems/Atom/Feature/Common/Assets/Shaders/MorphTargets/MorphTargetCS.azsl b/Gems/Atom/Feature/Common/Assets/Shaders/MorphTargets/MorphTargetCS.azsl
@@ -17,9 +17,6 @@ rootconstant uint s_targetPositionOffset;
 rootconstant uint s_targetNormalOffset;
 rootconstant uint s_targetTangentOffset;
 rootconstant uint s_targetBitangentOffset;
-rootconstant uint s_targetColorOffset;
-
-option bool o_hasColorDeltas = false;
 
 void WriteDeltaToAccumulationBuffer(float3 delta, uint offset, uint morphedVertexIndex)
 {
@@ -87,18 +84,5 @@ void MainCS(uint3 thread_id: SV_DispatchThreadID)
         // Now that we have the compressed bitangents, unpack them and write them to the accumulation buffer      
         float3 bitangentDelta = DecodeTBNDelta(compressedBitangentDelta) * s_weight;
         WriteDeltaToAccumulationBuffer(bitangentDelta, s_targetBitangentOffset, morphedVertexIndex);
-
-        if (o_hasColorDeltas)
-        {
-            uint4 compressedColorDelta;
-            // Colors are in the least significant 24 bits (8 bits per channel)
-            compressedColorDelta.r = (delta.m_compressedColorDeltaRGBA >> 24) & 0x000000FF;
-            compressedColorDelta.g = (delta.m_compressedColorDeltaRGBA >> 16) & 0x000000FF;
-            compressedColorDelta.b = (delta.m_compressedColorDeltaRGBA >> 8)  & 0x000000FF;
-            compressedColorDelta.a =  delta.m_compressedColorDeltaRGBA        & 0x000000FF;
-
-            float4 colorDelta = DecodeColorDelta(compressedColorDelta) * s_weight;
-            WriteDeltaToAccumulationBuffer(colorDelta, s_targetColorOffset, morphedVertexIndex);
-        }
     }
 }
diff --git a/Gems/Atom/Feature/Common/Assets/Shaders/MorphTargets/MorphTargetSRG.azsli b/Gems/Atom/Feature/Common/Assets/Shaders/MorphTargets/MorphTargetSRG.azsli
@@ -32,10 +32,8 @@ struct MorphTargetDelta
     uint m_compressedNormalDeltaZTangentDelta;
     // 8 bit padding plus 8 bits per component for bitangent deltas
     uint m_compressedPadBitangentDeltaXYZ;
-    // 8 bits per component for color delta
-    uint m_compressedColorDeltaRGBA;
     // Extra padding so the struct is 16 byte aligned for structured buffers
-    uint2 m_pad;
+    uint3 m_pad;
 };
 
 // Input to the morph target compute shader

diff --git a/Gems/Atom/Feature/Common/Assets/Shaders/SkinnedMesh/LinearSkinningCS.azsl b/Gems/Atom/Feature/Common/Assets/Shaders/SkinnedMesh/LinearSkinningCS.azsl
@@ -13,7 +13,6 @@
 
 option enum class SkinningMethod { LinearSkinning, DualQuaternion } o_skinningMethod = SkinningMethod::LinearSkinning;
 option bool o_applyMorphTargets = false;
-option bool o_applyColorMorphTargets = false;
 
 float3 ReadFloat3FromFloatBuffer(Buffer<float> buffer, uint index)
 {
@@ -92,14 +91,37 @@ void SkinTBN(float3x3 skinToWorldMatrix, inout float3 normal, inout float4 tange
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Skinning support
-void SkinVertexLinear(int4 indices, float4 weights, inout float3 position, inout float3 normal, inout float4 tangent, inout float3 bitangent)
+
+// Get two weights and two joint ids
+void GetInfluences(in uint weightStartOffsetForVertex, in uint jointIdStartOffsetForVertex, in uint influenceIndex, out float2 weights, out uint2 jointIds)
+{
+    weights.x = InstanceSrg::m_sourceBlendWeights[weightStartOffsetForVertex + influenceIndex];
+    weights.y = InstanceSrg::m_sourceBlendWeights[weightStartOffsetForVertex + influenceIndex + 1];
+
+    // Two indices, 16-bits each, stored in a 32-bit uint
+    uint rawIndex = InstanceSrg::m_sourceBlendIndices.Load(jointIdStartOffsetForVertex + influenceIndex);
+
+    // The first index in each 32-bit pair is in the most significant bits in the buffer
+    jointIds.x = rawIndex >> 16 & 0x0000FFFF;
+    jointIds.y = rawIndex & 0x0000FFFF;
+}
+
+void SkinVertexLinear(uint vertexIndex, inout float3 position, inout float3 normal, inout float4 tangent, inout float3 bitangent)
 {
     float3x4 skinToWorldMatrix = (float3x4)0;
 
-    [unroll]
-    for(uint i = 0; i < 4; ++i)
+    uint weightStartOffsetForVertex = vertexIndex * InstanceSrg::m_numInfluencesPerVertex;
+    // Multiply by two here since the jointId offset is in bytes, and there are two bytes per 16-bit jointId
+    uint jointIdStartOffsetForVertex = vertexIndex * InstanceSrg::m_numInfluencesPerVertex * 2;
+    [loop]
+    for(uint i = 0; i < InstanceSrg::m_numInfluencesPerVertex; i+=2)
     {
-        skinToWorldMatrix += InstanceSrg::m_boneTransformsLinear[ indices[i] ] * weights[i];
+        float2 weights;
+        float2 jointIds;
+        GetInfluences(weightStartOffsetForVertex, jointIdStartOffsetForVertex, i, weights, jointIds);
+
+        skinToWorldMatrix += InstanceSrg::m_boneTransformsLinear[ jointIds.x ] * weights.x;
+        skinToWorldMatrix += InstanceSrg::m_boneTransformsLinear[ jointIds.y ] * weights.y;
     }
 
     position = mul(skinToWorldMatrix, float4(position, 1.0));
@@ -120,14 +142,22 @@ void NormalizeDualQuaternion(inout float2x4 dualQuaternion)
     dualQuaternion *= invLength;
 }
 
-void SkinVertexDualQuaternion(int4 indices, float4 weights, inout float3 position, inout float3 normal, inout float4 tangent, inout float3 bitangent)
+void SkinVertexDualQuaternion(uint vertexIndex, inout float3 position, inout float3 normal, inout float4 tangent, inout float3 bitangent)
 {
     float2x4 skinToWorldDualQuaternion = (float2x4)0;
 
-    [unroll]
-    for(uint i = 0; i < 4; ++i)
-    {
-        AddWeightedDualQuaternion(skinToWorldDualQuaternion, InstanceSrg::m_boneTransformsDualQuaternion[ indices[i] ], weights[i]);
+    uint weightStartOffsetForVertex = vertexIndex * InstanceSrg::m_numInfluencesPerVertex;
+    // Multiply by two here since the jointId offset is in bytes, and there are two bytes per 16-bit jointId
+    uint jointIdStartOffsetForVertex = vertexIndex * InstanceSrg::m_numInfluencesPerVertex * 2;
+    [loop]
+    for(uint i = 0; i < InstanceSrg::m_numInfluencesPerVertex; i+=2)
+    {        
+        float2 weights;
+        float2 jointIds;
+        GetInfluences(weightStartOffsetForVertex, jointIdStartOffsetForVertex, i, weights, jointIds);
+
+        AddWeightedDualQuaternion(skinToWorldDualQuaternion, InstanceSrg::m_boneTransformsDualQuaternion[ jointIds.x ], weights.x);
+        AddWeightedDualQuaternion(skinToWorldDualQuaternion, InstanceSrg::m_boneTransformsDualQuaternion[ jointIds.y ], weights.y);
     }
 
     NormalizeDualQuaternion(skinToWorldDualQuaternion);
@@ -154,46 +184,15 @@ void MainCS(uint3 thread_id: SV_DispatchThreadID)
     if(i < InstanceSrg::m_numVertices)
     {
         // Moving current vertex position updated last frame to a predefined location to maintain a vertex history between two frames
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3];
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3 + 1] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 1];
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3 + 2] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 2];
-
-        float4 blendWeights;
-        blendWeights.x = InstanceSrg::m_sourceBlendWeights[i * 4];
-        blendWeights.y = InstanceSrg::m_sourceBlendWeights[i * 4 + 1];
-        blendWeights.z = InstanceSrg::m_sourceBlendWeights[i * 4 + 2];
-        blendWeights.w = InstanceSrg::m_sourceBlendWeights[i * 4 + 3];
-
-        // [TODO ATOM-15288]
-        // Temporary workaround. When all the blend weights of a vertex are zero it means its data is set by the CPU directly
-        // and skinning must be skipped to not overwrite it (e.g. cloth simulation).
-        if(!any(blendWeights))
-        {
-            return;
-        }
+        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositionHistory + i * 3] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3];
+        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositionHistory + i * 3 + 1] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 1];
+        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositionHistory + i * 3 + 2] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 2];
 
         float3 position = ReadFloat3FromFloatBuffer(InstanceSrg::m_sourcePositions, i);
         float3 normal = ReadFloat3FromFloatBuffer(InstanceSrg::m_sourceNormals, i);    
         float4 tangent = InstanceSrg::m_sourceTangents[i];      
         float3 bitangent = ReadFloat3FromFloatBuffer(InstanceSrg::m_sourceBiTangents, i);
 
-        // Four indices, 16-bits each, stored in 2 32-bit uints
-        uint2 rawIndices = InstanceSrg::m_sourceBlendIndices.Load2(i * 8);
-
-        uint4 blendIndices;
-        // Although the first index in each 32-bit pair is in the most significant bits in the cpu buffer,
-        // the indices get swapped within the 32-bit uint when they are loaded so we treat them
-        // as if the first index is in the least significant bits
-        blendIndices.x = rawIndices.x >> 16;
-        blendIndices.y = rawIndices.x & 0x0000FFFF;
-        blendIndices.z = rawIndices.y >> 16;
-        blendIndices.w = rawIndices.y & 0x0000FFFF;
-
-
-        // Moving current vertex position updated last frame to a predefined location to maintain a vertex history between two frames
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3];
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3 + 1] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 1];
-        PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + (InstanceSrg::m_numVertices + i) * 3 + 2] = PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetPositions + i * 3 + 2];
 
         if(o_applyMorphTargets)
         {
@@ -202,24 +201,14 @@ void MainCS(uint3 thread_id: SV_DispatchThreadID)
             ApplyMorphTargetDelta(InstanceSrg::m_morphTargetTangentDeltaOffset, i, tangent.xyz);
             ApplyMorphTargetDelta(InstanceSrg::m_morphTargetBitangentDeltaOffset, i, bitangent);
         }
-
-        if (o_applyColorMorphTargets)
-        {
-            float4 color = InstanceSrg::m_sourceColors[i];
-            ApplyMorphTargetDelta(InstanceSrg::m_morphTargetColorDeltaOffset, i, color);
-            PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetColors + i * 4] = color.r;
-            PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetColors + i * 4 + 1] = color.g;
-            PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetColors + i * 4 + 2] = color.b;
-            PassSrg::m_skinnedMeshOutputStream[InstanceSrg::m_targetColors + i * 4 + 3] = color.a;
-        }
 
         switch(o_skinningMethod)
         {
         case SkinningMethod::LinearSkinning:
-            SkinVertexLinear(blendIndices, blendWeights, position, normal, tangent, bitangent);
+            SkinVertexLinear(i, position, normal, tangent, bitangent);
             break;
         case SkinningMethod::DualQuaternion:
-            SkinVertexDualQuaternion(blendIndices, blendWeights, position, normal, tangent, bitangent);
+            SkinVertexDualQuaternion(i, position, normal, tangent, bitangent);
             break;
         }
 

diff --git a/Gems/Atom/Feature/Common/Assets/Shaders/SkinnedMesh/LinearSkinningPassSRG.azsli b/Gems/Atom/Feature/Common/Assets/Shaders/SkinnedMesh/LinearSkinningPassSRG.azsli
@@ -18,6 +18,7 @@ ShaderResourceGroup PassSrg : SRG_PerPass
 ShaderResourceGroup InstanceSrg : SRG_PerDraw
 {    
     uint m_numVertices;
+    uint m_numInfluencesPerVertex;
     uint m_totalNumberOfThreadsX;
 
     // Per-model input
@@ -62,6 +63,10 @@ ShaderResourceGroup InstanceSrg : SRG_PerDraw
     uint m_targetTangents;
     uint m_targetBiTangents;
 
+    // Offset to the location in the output stream with the start of the position history
+    // for the particular mesh the dispatch is operating on
+    uint m_targetPositionHistory;
+
     // Optional color output, if colors are being morphed by morph targets
     uint m_targetColors;
 }
diff --git a/Gems/Atom/Feature/Common/Code/Include/Atom/Feature/MorphTargets/MorphTargetInputBuffers.h b/Gems/Atom/Feature/Common/Code/Include/Atom/Feature/MorphTargets/MorphTargetInputBuffers.h
@@ -52,15 +52,17 @@ namespace AZ
             Data::Instance<RPI::Buffer> m_vertexDeltaBuffer;
         };
 
-        struct MorphTargetMetaData
+        struct MorphTargetComputeMetaData
         {
             float m_minWeight;
             float m_maxWeight;
             float m_minDelta;
             float m_maxDelta;
             uint32_t m_vertexCount;
-            uint32_t m_positionOffset;
-            bool m_hasColorDeltas;
+            // Each morph target dispatch is associated with a single mesh. We need to keep track of which mesh
+            // so that we can calculate the maximum range a given mesh might be morphed if all of the morph targets
+            // associated with it were active at once.
+            uint32_t m_meshIndex;
         };
 
         namespace MorphTargetConstants
@@ -80,7 +82,6 @@ namespace AZ
             uint32_t m_accumulatedNormalDeltaOffsetInBytes;
             uint32_t m_accumulatedTangentDeltaOffsetInBytes;
             uint32_t m_accumulatedBitangentDeltaOffsetInBytes;
-            uint32_t m_accumulatedColorDeltaOffsetInBytes;
         };
 
     }// Render

diff --git a/...ature/Common/Code/Include/Atom/Feature/SkinnedMesh/SkinnedMeshFeatureProcessorInterface.h b/...ature/Common/Code/Include/Atom/Feature/SkinnedMesh/SkinnedMeshFeatureProcessorInterface.h
@@ -7,7 +7,6 @@
  */
 #pragma once
 
-#include <Atom/Feature/SkinnedMesh/SkinnedMeshRenderProxyInterface.h>
 #include <Atom/Feature/SkinnedMesh/SkinnedMeshInstance.h>
 #include <Atom/Feature/SkinnedMesh/SkinnedMeshShaderOptions.h>
 #include <Atom/Feature/Mesh/MeshFeatureProcessorInterface.h>
@@ -19,6 +18,7 @@ namespace AZ
     namespace Render
     {
         class SkinnedMeshInputBuffers;
+        class SkinnedMeshRenderProxy;
 
         //! SkinnedMeshFeatureProcessorInterface provides an interface to acquire and release a SkinnedMeshRenderProxy from the underlying SkinnedMeshFeatureProcessor
         class SkinnedMeshFeatureProcessorInterface
@@ -27,16 +27,30 @@ namespace AZ
         public:
             AZ_RTTI(AZ::Render::SkinnedMeshFeatureProcessorInterface, "{6BE6D9D7-FFD7-4C35-9A84-4EFDE730F06B}");
 
-            struct SkinnedMeshRenderProxyDesc
+            using SkinnedMeshHandle = StableDynamicArrayHandle<SkinnedMeshRenderProxy>;
+
+            struct SkinnedMeshHandleDescriptor
             {
                 Data::Instance<SkinnedMeshInputBuffers> m_inputBuffers;
                 AZStd::intrusive_ptr<SkinnedMeshInstance> m_instance;
                 AZStd::shared_ptr<MeshFeatureProcessorInterface::MeshHandle> m_meshHandle;
                 Data::Instance<RPI::Buffer> m_boneTransforms;
                 SkinnedMeshShaderOptions m_shaderOptions;
             };
-            virtual SkinnedMeshRenderProxyInterfaceHandle AcquireRenderProxyInterface(const SkinnedMeshRenderProxyDesc& desc) = 0;
-            virtual bool ReleaseRenderProxyInterface(SkinnedMeshRenderProxyInterfaceHandle& handle) = 0;
+
+            //! Given a descriptor of the input and output for skinning, acquire a handle to the instance that will be skinned
+            virtual SkinnedMeshHandle AcquireSkinnedMesh(const SkinnedMeshHandleDescriptor& desc) = 0;
+            //! Releases the skinned mesh handle
+            virtual bool ReleaseSkinnedMesh(SkinnedMeshHandle& handle) = 0;
+            //! Updates the data for the skinning transforms of a given skinned mesh handle
+            virtual void SetSkinningMatrices(const SkinnedMeshHandle& handle, const AZStd::vector<float>& data) = 0;
+            //! Updates the morph target weights for all meshes of a given lod of a skinned mesh handle
+            //! The weights should be in the order that the morph targets were initially added to the SkinnedMeshInputBuffers for the handle
+            virtual void SetMorphTargetWeights(const SkinnedMeshHandle& handle, uint32_t lodIndex, const AZStd::vector<float>& weights) = 0;
+            //! Enable skinning for a given mesh and lod of a skinned mesh handle
+            virtual void EnableSkinning(const SkinnedMeshHandle& handle, uint32_t lodIndex, uint32_t meshIndex) = 0;
+            //! Disable skinning for a given mesh and lod of a skinned mesh handle
+            virtual void DisableSkinning(const SkinnedMeshHandle& handle, uint32_t lodIndex, uint32_t meshIndex) = 0;
         };
     } // namespace Render
 } // namespace AZ