svn merge -r 59968:60342 svn+ssh://svn.lindenlab.com/svn/linden/branches/maintenance --> release

3 files changed, 55 insertions, 43 deletions

diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp
index 969c3663e6..8d5fc27776 100644
--- a/indra/llmath/m4math.cpp
+++ b/indra/llmath/m4math.cpp
@@ -654,31 +654,55 @@ LLVector4 operator*(const LLMatrix4 &a, const LLVector4 &b)
 }
 */
 
+// Operates "to the left" on row-vector a
+//
+// This used to be in the header file but was not actually inlined in practice.
+// When avatar vertex programs are off, this function is a hot spot in profiles
+// due to software skinning in LLViewerJointMesh::updateGeometry().  JC
+LLVector3 operator*(const LLVector3 &a, const LLMatrix4 &b)
+{
+	// This is better than making a temporary LLVector3.  This eliminates an
+	// unnecessary LLVector3() constructor and also helps the compiler to
+	// realize that the output floats do not alias the input floats, hence
+	// eliminating redundant loads of a.mV[0], etc.  JC
+	return LLVector3(a.mV[VX] * b.mMatrix[VX][VX] + 
+					 a.mV[VY] * b.mMatrix[VY][VX] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VX] +
+					 b.mMatrix[VW][VX],
+					 
+					 a.mV[VX] * b.mMatrix[VX][VY] + 
+					 a.mV[VY] * b.mMatrix[VY][VY] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VY] +
+					 b.mMatrix[VW][VY],
+					 
+					 a.mV[VX] * b.mMatrix[VX][VZ] + 
+					 a.mV[VY] * b.mMatrix[VY][VZ] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VZ] +
+					 b.mMatrix[VW][VZ]);
+}
 
 LLVector4 operator*(const LLVector4 &a, const LLMatrix4 &b)
 {
 	// Operate "to the left" on row-vector a
-	LLVector4	vec;
-	vec.mV[VX] = a.mV[VX] * b.mMatrix[VX][VX] + 
-				 a.mV[VY] * b.mMatrix[VY][VX] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VX] +
-				 a.mV[VW] * b.mMatrix[VW][VX];
-
-	vec.mV[VY] = a.mV[VX] * b.mMatrix[VX][VY] + 
-				 a.mV[VY] * b.mMatrix[VY][VY] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VY] +
-				 a.mV[VW] * b.mMatrix[VW][VY];
-
-	vec.mV[VZ] = a.mV[VX] * b.mMatrix[VX][VZ] + 
-				 a.mV[VY] * b.mMatrix[VY][VZ] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VZ] +
-				 a.mV[VW] * b.mMatrix[VW][VZ];
-
-	vec.mV[VW] = a.mV[VX] * b.mMatrix[VX][VW] + 
-				 a.mV[VY] * b.mMatrix[VY][VW] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VW] +
-				 a.mV[VW] * b.mMatrix[VW][VW];
-	return vec;
+	return LLVector4(a.mV[VX] * b.mMatrix[VX][VX] + 
+					 a.mV[VY] * b.mMatrix[VY][VX] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VX] +
+					 a.mV[VW] * b.mMatrix[VW][VX],
+
+					 a.mV[VX] * b.mMatrix[VX][VY] + 
+					 a.mV[VY] * b.mMatrix[VY][VY] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VY] +
+					 a.mV[VW] * b.mMatrix[VW][VY],
+
+					 a.mV[VX] * b.mMatrix[VX][VZ] + 
+					 a.mV[VY] * b.mMatrix[VY][VZ] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VZ] +
+					 a.mV[VW] * b.mMatrix[VW][VZ],
+
+					 a.mV[VX] * b.mMatrix[VX][VW] + 
+					 a.mV[VY] * b.mMatrix[VY][VW] + 
+					 a.mV[VZ] * b.mMatrix[VZ][VW] +
+					 a.mV[VW] * b.mMatrix[VW][VW]);
 }
 
 LLVector4 rotate_vector(const LLVector4 &a, const LLMatrix4 &b)
diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h
index 69463d7718..e95a40af16 100644
--- a/indra/llmath/m4math.h
+++ b/indra/llmath/m4math.h
@@ -247,27 +247,6 @@ inline const LLMatrix4&	LLMatrix4::identity()
 	return (*this);
 }
 
-inline LLVector3 operator*(const LLVector3 &a, const LLMatrix4 &b)
-{
-	// Converts a to LLVector4 and applies full transformation
-	// Operates "to the left" on row-vector a
-	LLVector3	vec;
-	vec.mV[VX] = a.mV[VX] * b.mMatrix[VX][VX] + 
-				 a.mV[VY] * b.mMatrix[VY][VX] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VX] +
-				 b.mMatrix[VW][VX];
-
-	vec.mV[VY] = a.mV[VX] * b.mMatrix[VX][VY] + 
-				 a.mV[VY] * b.mMatrix[VY][VY] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VY] +
-				 b.mMatrix[VW][VY];
-
-	vec.mV[VZ] = a.mV[VX] * b.mMatrix[VX][VZ] + 
-				 a.mV[VY] * b.mMatrix[VY][VZ] + 
-				 a.mV[VZ] * b.mMatrix[VZ][VZ] +
-				 b.mMatrix[VW][VZ];
-	return vec;
-}
 
 /*
 inline LLMatrix4 operator*(const LLMatrix4 &a, const LLMatrix4 &b)
diff --git a/indra/llmath/v4math.h b/indra/llmath/v4math.h
index e9b8d5b71f..ced7ef131a 100644
--- a/indra/llmath/v4math.h
+++ b/indra/llmath/v4math.h
@@ -26,7 +26,8 @@ class LLVector4
 	public:
 		F32 mV[LENGTHOFVECTOR4];
 		LLVector4();						// Initializes LLVector4 to (0, 0, 0, 1)
-		explicit LLVector4(const F32 *vec);			// Initializes LLVector4 to (vec[0]. vec[1], vec[2], 1)
+		explicit LLVector4(const F32 *vec);			// Initializes LLVector4 to (vec[0]. vec[1], vec[2], vec[3])
+		explicit LLVector4(const F64 *vec);			// Initialized LLVector4 to ((F32) vec[0], (F32) vec[1], (F32) vec[3], (F32) vec[4]);
 		explicit LLVector4(const LLVector3 &vec);			// Initializes LLVector4 to (vec, 1)
 		explicit LLVector4(const LLVector3 &vec, F32 w);	// Initializes LLVector4 to (vec, w)
 		LLVector4(F32 x, F32 y, F32 z);		// Initializes LLVector4 to (x. y, z, 1)
@@ -136,6 +137,14 @@ inline LLVector4::LLVector4(const F32 *vec)
 	mV[VW] = vec[VW];
 }
 
+inline LLVector4::LLVector4(const F64 *vec)
+{
+	mV[VX] = (F32) vec[VX];
+	mV[VY] = (F32) vec[VY];
+	mV[VZ] = (F32) vec[VZ];
+	mV[VW] = (F32) vec[VW];
+}
+
 inline LLVector4::LLVector4(const LLVector3 &vec)
 {
 	mV[VX] = vec.mV[VX];