SL-16006 and SL-16009 Rigged mesh rendering optimization pass

3 files changed, 20 insertions, 13 deletions

diff --git a/indra/llprimitive/lldaeloader.cpp b/indra/llprimitive/lldaeloader.cpp
index dfa29fb539..8343de0cbc 100644
--- a/indra/llprimitive/lldaeloader.cpp
+++ b/indra/llprimitive/lldaeloader.cpp
@@ -1173,17 +1173,19 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 
 			LLMeshSkinInfo& skin_info = model->mSkinInfo;
 
+            LLMatrix4 mat;
 			for (int i = 0; i < 4; i++)
 			{
 				for(int j = 0; j < 4; j++)
 				{
-					skin_info.mBindShapeMatrix.mMatrix[i][j] = dom_value[i + j*4];
+                    mat.mMatrix[i][j] = dom_value[i + j*4];
 				}
 			}
 
-			LLMatrix4 trans = normalized_transformation;
-			trans *= skin_info.mBindShapeMatrix;
-			skin_info.mBindShapeMatrix = trans;							
+            skin_info.mBindShapeMatrix.loadu(mat);
+
+			LLMatrix4a trans(normalized_transformation);
+            matMul(trans, skin_info.mBindShapeMatrix, skin_info.mBindShapeMatrix);
 		}
 
 
@@ -1401,7 +1403,7 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 									mat.mMatrix[i][j] = transform[k*16 + i + j*4];
 								}
 							}
-							model->mSkinInfo.mInvBindMatrix.push_back(mat);
+							model->mSkinInfo.mInvBindMatrix.push_back(LLMatrix4a(mat));
 						}
 					}
 				}
@@ -1475,9 +1477,9 @@ void LLDAELoader::processDomModel(LLModel* model, DAE* dae, daeElement* root, do
 			if (mJointMap.find(lookingForJoint) != mJointMap.end()
 				&& model->mSkinInfo.mInvBindMatrix.size() > i)
 			{
-				LLMatrix4 newInverse = model->mSkinInfo.mInvBindMatrix[i];
+				LLMatrix4 newInverse = LLMatrix4(model->mSkinInfo.mInvBindMatrix[i].getF32ptr());
 				newInverse.setTranslation( mJointList[lookingForJoint].getTranslation() );
-				model->mSkinInfo.mAlternateBindMatrix.push_back( newInverse );
+				model->mSkinInfo.mAlternateBindMatrix.push_back( LLMatrix4a(newInverse) );
             }
 			else
 			{
diff --git a/indra/llprimitive/llmodel.cpp b/indra/llprimitive/llmodel.cpp
index 702a1b5238..a23b991f1d 100644
--- a/indra/llprimitive/llmodel.cpp
+++ b/indra/llprimitive/llmodel.cpp
@@ -1396,7 +1396,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 				}
 			}
 
-			mInvBindMatrix.push_back(mat);
+			mInvBindMatrix.push_back(LLMatrix4a(mat));
 		}
 
         if (mJointNames.size() != mInvBindMatrix.size())
@@ -1410,13 +1410,15 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 
 	if (skin.has("bind_shape_matrix"))
 	{
+        LLMatrix4 mat;
 		for (U32 j = 0; j < 4; j++)
 		{
 			for (U32 k = 0; k < 4; k++)
 			{
-				mBindShapeMatrix.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal();
+				mat.mMatrix[j][k] = skin["bind_shape_matrix"][j*4+k].asReal();
 			}
 		}
+        mBindShapeMatrix.loadu(mat);
 	}
 
 	if (skin.has("alt_inverse_bind_matrix"))
@@ -1432,7 +1434,7 @@ void LLMeshSkinInfo::fromLLSD(LLSD& skin)
 				}
 			}
 			
-			mAlternateBindMatrix.push_back(mat);
+			mAlternateBindMatrix.push_back(LLMatrix4a(mat));
 		}
 	}
 
diff --git a/indra/llprimitive/llmodel.h b/indra/llprimitive/llmodel.h
index 51fa2f8079..96d4582b4f 100644
--- a/indra/llprimitive/llmodel.h
+++ b/indra/llprimitive/llmodel.h
@@ -33,6 +33,8 @@
 #include "m4math.h"
 #include <queue>
 
+#include <boost/align/aligned_allocator.hpp>
+
 class daeElement;
 class domMesh;
 
@@ -49,10 +51,11 @@ public:
 	LLUUID mMeshID;
 	std::vector<std::string> mJointNames;
     mutable std::vector<S32> mJointNums;
-	std::vector<LLMatrix4> mInvBindMatrix;
-	std::vector<LLMatrix4> mAlternateBindMatrix;
+    typedef std::vector<LLMatrix4a, boost::alignment::aligned_allocator<LLMatrix4a, 16>> matrix_list_t;
+	matrix_list_t mInvBindMatrix;
+	matrix_list_t mAlternateBindMatrix;
 
-	LLMatrix4 mBindShapeMatrix;
+	LLMatrix4a mBindShapeMatrix;
 	float mPelvisOffset;
     bool mLockScaleIfJointPosition;
     bool mInvalidJointsScrubbed;