Index: source/maths/Matrix3D.h
===================================================================
--- source/maths/Matrix3D.h
+++ source/maths/Matrix3D.h
@@ -1,4 +1,4 @@
-/* Copyright (C) 2019 Wildfire Games.
+/* Copyright (C) 2021 Wildfire Games.
  * This file is part of 0 A.D.
  *
  * 0 A.D. is free software: you can redistribute it and/or modify
@@ -23,6 +23,7 @@
 #ifndef INCLUDED_MATRIX3D
 #define INCLUDED_MATRIX3D
 
+#include "lib/sysdep/arch/x86_x64/simd.h"
 #include "maths/Vector3D.h"
 #include "maths/Vector4D.h"
 
@@ -100,6 +101,50 @@
 	// matrix multiplication
 	CMatrix3D operator*(const CMatrix3D &matrix) const
 	{
+#if OS_WIN && COMPILER_HAS_SSE
+		CMatrix3D result;
+		__m128 col1 = _mm_loadu_ps(_data2d[0]);
+		__m128 col2 = _mm_loadu_ps(_data2d[1]);
+		__m128 col3 = _mm_loadu_ps(_data2d[2]);
+		__m128 col4 = _mm_loadu_ps(_data2d[3]);
+
+		__m128 mvec = _mm_set_ps1(matrix._11);
+		__m128 vec = _mm_mul_ps(mvec, col1);
+		mvec = _mm_set_ps1(matrix._21);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col2), vec);
+		mvec = _mm_set_ps1(matrix._31);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col3), vec);
+		mvec = _mm_set_ps1(matrix._41);
+		_mm_storeu_ps(result._data2d[0], _mm_add_ps(_mm_mul_ps(mvec, col4), vec));
+
+		mvec = _mm_set_ps1(matrix._12);
+		vec = _mm_mul_ps(mvec, col1);
+		mvec = _mm_set_ps1(matrix._22);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col2), vec);
+		mvec = _mm_set_ps1(matrix._32);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col3), vec);
+		mvec = _mm_set_ps1(matrix._42);
+		_mm_storeu_ps(result._data2d[1], _mm_add_ps(_mm_mul_ps(mvec, col4), vec));
+
+		mvec = _mm_set_ps1(matrix._13);
+		vec = _mm_mul_ps(mvec, col1);
+		mvec = _mm_set_ps1(matrix._23);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col2), vec);
+		mvec = _mm_set_ps1(matrix._33);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col3), vec);
+		mvec = _mm_set_ps1(matrix._43);
+		_mm_storeu_ps(result._data2d[2], _mm_add_ps(_mm_mul_ps(mvec, col4), vec));
+
+		mvec = _mm_set_ps1(matrix._14);
+		vec = _mm_mul_ps(mvec, col1);
+		mvec = _mm_set_ps1(matrix._24);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col2), vec);
+		mvec = _mm_set_ps1(matrix._34);
+		vec = _mm_add_ps(_mm_mul_ps(mvec, col3), vec);
+		mvec = _mm_set_ps1(matrix._44);
+		_mm_storeu_ps(result._data2d[3], _mm_add_ps(_mm_mul_ps(mvec, col4), vec));
+		return result;
+#else
 		return CMatrix3D(
 			_11*matrix._11 + _12*matrix._21 + _13*matrix._31 + _14*matrix._41,
 			_11*matrix._12 + _12*matrix._22 + _13*matrix._32 + _14*matrix._42,
@@ -121,6 +166,7 @@
 			_41*matrix._13 + _42*matrix._23 + _43*matrix._33 + _44*matrix._43,
 			_41*matrix._14 + _42*matrix._24 + _43*matrix._34 + _44*matrix._44
 		);
+#endif
 	}
 
 	// matrix multiplication/assignment
@@ -196,19 +242,54 @@
 	// blend matrix using only 4x3 subset
 	void Blend(const CMatrix3D& m, float f)
 	{
+#if OS_WIN && COMPILER_HAS_SSE
+		__m128 fvec = _mm_set_ps(0.f, f, f, f);
+		__m128 mask = _mm_set_ps(1.f, 0.f, 0.f, 0.f);
+
+		__m128 col = _mm_loadu_ps(_data2d[0]);
+		__m128 mcol = _mm_loadu_ps(m._data2d[0]);
+		_mm_storeu_ps(_data2d[0], _mm_add_ps(_mm_mul_ps(col, mask), _mm_mul_ps(mcol, fvec)));
+		col = _mm_loadu_ps(_data2d[1]);
+		mcol = _mm_loadu_ps(m._data2d[1]);
+		_mm_storeu_ps(_data2d[1], _mm_add_ps(_mm_mul_ps(col, mask), _mm_mul_ps(mcol, fvec)));
+		col = _mm_loadu_ps(_data2d[2]);
+		mcol = _mm_loadu_ps(m._data2d[2]);
+		_mm_storeu_ps(_data2d[2], _mm_add_ps(_mm_mul_ps(col, mask), _mm_mul_ps(mcol, fvec)));
+		col = _mm_loadu_ps(_data2d[3]);
+		mcol = _mm_loadu_ps(m._data2d[3]);
+		_mm_storeu_ps(_data2d[3], _mm_add_ps(_mm_mul_ps(col, mask), _mm_mul_ps(mcol, fvec)));
+#else
 		_11 = m._11*f; _21 = m._21*f; _31 = m._31*f;
 		_12 = m._12*f; _22 = m._22*f; _32 = m._32*f;
 		_13 = m._13*f; _23 = m._23*f; _33 = m._33*f;
 		_14 = m._14*f; _24 = m._24*f; _34 = m._34*f;
+#endif
 	}
 
 	// blend matrix using only 4x3 and add onto existing blend
 	void AddBlend(const CMatrix3D& m, float f)
 	{
+#if OS_WIN && COMPILER_HAS_SSE
+		__m128 fvec = _mm_set_ps(0.f, f, f, f);
+
+		__m128 col = _mm_loadu_ps(_data2d[0]);
+		__m128 mcol = _mm_loadu_ps(m._data2d[0]);
+		_mm_storeu_ps(_data2d[0], _mm_add_ps(_mm_mul_ps(mcol, fvec), col));
+		col = _mm_loadu_ps(_data2d[1]);
+		mcol = _mm_loadu_ps(m._data2d[1]);
+		_mm_storeu_ps(_data2d[1], _mm_add_ps(_mm_mul_ps(mcol, fvec), col));
+		col = _mm_loadu_ps(_data2d[2]);
+		mcol = _mm_loadu_ps(m._data2d[2]);
+		_mm_storeu_ps(_data2d[2], _mm_add_ps(_mm_mul_ps(mcol, fvec), col));
+		col = _mm_loadu_ps(_data2d[3]);
+		mcol = _mm_loadu_ps(m._data2d[3]);
+		_mm_storeu_ps(_data2d[3], _mm_add_ps(_mm_mul_ps(mcol, fvec), col));
+#else
 		_11 += m._11*f; _21 += m._21*f; _31 += m._31*f;
 		_12 += m._12*f; _22 += m._22*f; _32 += m._32*f;
 		_13 += m._13*f; _23 += m._23*f; _33 += m._33*f;
 		_14 += m._14*f; _24 += m._24*f; _34 += m._34*f;
+#endif
 	}
 
 	// set this matrix to a rotation matrix for a rotation about X axis of given angle