If there are too many wrinkle masks, discard the lowest weight ones (#988)

Since the shader supports a fixed maximum number of active wrinkle masks, if the number of active wrinkle masks is larger than that, it will crash where wrinkleMaskObjectSrg->SetImageArray is called with an array_view which is larger than the target image array.

We need a sensible method of truncating the number of masks. One such method is to discard the lowest weight masks.

Gems/AtomLyIntegration/EMotionFXAtom/Code/Source/AtomActorInstance.cpp

@@ -526,6 +526,20 @@ namespace AZ
             }
         }
 
+        template<class X>
+        void swizzle_unique(AZStd::vector<X>& values, const AZStd::vector<size_t>& indices)
+        {
+            AZStd::vector<X> out;
+            out.reserve(indices.size());
+
+            for (size_t i : indices)
+            {
+                out.push_back(AZStd::move(values[i]));
+            }
+
+            values = AZStd::move(out);
+        }
+
         void AtomActorInstance::OnUpdateSkinningMatrices()
         {
             if (m_skinnedMeshRenderProxy.IsValid())
@@ -584,6 +598,30 @@ namespace AZ
                                 }
                             }
                         }
+
+                        AZ_Assert(m_wrinkleMasks.size() == m_wrinkleMaskWeights.size(), "Must have equal # of masks and weights");
+
+                        // If there's too many masks, truncate
+                        if (m_wrinkleMasks.size() > s_maxActiveWrinkleMasks)
+                        {
+                            // Build a remapping of indices (because we want to sort two vectors)
+                            AZStd::vector<size_t> remapped;
+                            remapped.resize_no_construct(m_wrinkleMasks.size());
+                            std::iota(remapped.begin(), remapped.end(), 0);
+
+                            // Sort index remapping by weight (highest first)
+                            std::sort(remapped.begin(), remapped.end(), [&](size_t ia, size_t ib) {
+                                return m_wrinkleMaskWeights[ia] > m_wrinkleMaskWeights[ib];
+                            });
+
+                            // Truncate indices list
+                            remapped.resize(s_maxActiveWrinkleMasks);
+
+                            // Remap wrinkle masks list and weights list
+                            swizzle_unique(m_wrinkleMasks, remapped);
+                            swizzle_unique(m_wrinkleMaskWeights, remapped);
+                        }
+
                         m_skinnedMeshRenderProxy->SetMorphTargetWeights(lodIndex, m_morphTargetWeights);
 
                         // Until EMotionFX and Atom lods are synchronized [ATOM-13564] we don't know which EMotionFX lod to pull the weights from