Changeset 8868 for OpenSceneGraph/trunk/include/osg/Matrixf

Timestamp:

09/17/08 18:14:28 (5 years ago)

Author:

robert

Message:

From Mathias Froehlich, "This is a generic optimization that does not depend on any cpu or instruction
set.

The optimization is based on the observation that matrix matrix multiplication
with a dense matrix 4x4 is 4^{3 Operations whereas multiplication with a
transform, or scale matrix is only 4}2 operations. Which is a gain of a
*FACTOR*4* for these special cases.
The change implements these special cases, provides a unit test for these
implementation and converts uses of the expensiver dense matrix matrix
routine with the specialized versions.

Depending on the transform nodes in the scenegraph this change gives a
noticable improovement.
For example the osgforest code using the MatrixTransform? is about 20% slower
than the same codepath using the PositionAttitudeTransform? instead of the
MatrixTransform? with this patch applied.

If I remember right, the sse type optimizations did *not* provide a factor 4
improovement. Also these changes are totally independent of any cpu or
instruction set architecture. So I would prefer to have this current kind of
change instead of some hand coded and cpu dependent assembly stuff. If we
need that hand tuned stuff, these can go on top of this changes which must
provide than hand optimized additional variants for the specialized versions
to give a even better result in the end.

An other change included here is a change to rotation matrix from quaterion
code. There is a sqrt call which couold be optimized away. Since we divide in
effect by sqrt(length)*sqrt(length) which is just length ...
"

Files:

• OpenSceneGraph/trunk/include/osg/Matrixf (modified) (2 diffs)

OpenSceneGraph/trunk/include/osg/Matrixf

@@ -350 +350 @@
         void postMult( const Matrixf& );
 
+        /** Optimized version of preMult(translate(v)); */
+        inline void preMultTranslate( const Vec3d& v );
+        inline void preMultTranslate( const Vec3f& v );
+        /** Optimized version of postMult(translate(v)); */
+        inline void postMultTranslate( const Vec3d& v );
+        inline void postMultTranslate( const Vec3f& v );
+
+        /** Optimized version of preMult(scale(v)); */
+        inline void preMultScale( const Vec3d& v );
+        inline void preMultScale( const Vec3f& v );
+        /** Optimized version of postMult(scale(v)); */
+        inline void postMultScale( const Vec3d& v );
+        inline void postMultScale( const Vec3f& v );
+
+        /** Optimized version of preMult(rotate(q)); */
+        inline void preMultRotate( const Quat& q );
+        /** Optimized version of postMult(rotate(q)); */
+        inline void postMultRotate( const Quat& q );
+
         inline void operator *= ( const Matrixf& other ) 
         {    if( this == &other ) {
@@ -642 +661 @@
 }
 
+inline void Matrixf::preMultTranslate( const Vec3d& v )
+{
+    for (unsigned i = 0; i < 3; ++i)
+    {
+        double tmp = v[i];
+        if (tmp == 0)
+            continue;
+        _mat[3][0] += tmp*_mat[i][0];
+        _mat[3][1] += tmp*_mat[i][1];
+        _mat[3][2] += tmp*_mat[i][2];
+        _mat[3][3] += tmp*_mat[i][3];
+    }
+}
+
+inline void Matrixf::preMultTranslate( const Vec3f& v )
+{
+    for (unsigned i = 0; i < 3; ++i)
+    {
+        float tmp = v[i];
+        if (tmp == 0)
+            continue;
+        _mat[3][0] += tmp*_mat[i][0];
+        _mat[3][1] += tmp*_mat[i][1];
+        _mat[3][2] += tmp*_mat[i][2];
+        _mat[3][3] += tmp*_mat[i][3];
+    }
+}
+
+inline void Matrixf::postMultTranslate( const Vec3d& v )
+{
+    for (unsigned i = 0; i < 3; ++i)
+    {
+        double tmp = v[i];
+        if (tmp == 0)
+            continue;
+        _mat[0][i] += tmp*_mat[0][3];
+        _mat[1][i] += tmp*_mat[1][3];
+        _mat[2][i] += tmp*_mat[2][3];
+        _mat[3][i] += tmp*_mat[3][3];
+    }
+}
+
+inline void Matrixf::postMultTranslate( const Vec3f& v )
+{
+    for (unsigned i = 0; i < 3; ++i)
+    {
+        float tmp = v[i];
+        if (tmp == 0)
+            continue;
+        _mat[0][i] += tmp*_mat[0][3];
+        _mat[1][i] += tmp*_mat[1][3];
+        _mat[2][i] += tmp*_mat[2][3];
+        _mat[3][i] += tmp*_mat[3][3];
+    }
+}
+
+inline void Matrixf::preMultScale( const Vec3d& v )
+{
+    _mat[0][0] *= v[0]; _mat[0][1] *= v[0]; _mat[0][2] *= v[0]; _mat[0][3] *= v[0];
+    _mat[1][0] *= v[1]; _mat[1][1] *= v[1]; _mat[1][2] *= v[1]; _mat[1][3] *= v[1];
+    _mat[2][0] *= v[2]; _mat[2][1] *= v[2]; _mat[2][2] *= v[2]; _mat[2][3] *= v[2];
+}
+
+inline void Matrixf::preMultScale( const Vec3f& v )
+{
+    _mat[0][0] *= v[0]; _mat[0][1] *= v[0]; _mat[0][2] *= v[0]; _mat[0][3] *= v[0];
+    _mat[1][0] *= v[1]; _mat[1][1] *= v[1]; _mat[1][2] *= v[1]; _mat[1][3] *= v[1];
+    _mat[2][0] *= v[2]; _mat[2][1] *= v[2]; _mat[2][2] *= v[2]; _mat[2][3] *= v[2];
+}
+
+inline void Matrixf::postMultScale( const Vec3d& v )
+{
+    _mat[0][0] *= v[0]; _mat[1][0] *= v[0]; _mat[2][0] *= v[0]; _mat[3][0] *= v[0];
+    _mat[0][1] *= v[1]; _mat[1][1] *= v[1]; _mat[2][1] *= v[1]; _mat[3][1] *= v[1];
+    _mat[0][2] *= v[2]; _mat[1][2] *= v[2]; _mat[2][2] *= v[2]; _mat[3][2] *= v[2];
+}
+
+inline void Matrixf::postMultScale( const Vec3f& v )
+{
+    _mat[0][0] *= v[0]; _mat[1][0] *= v[0]; _mat[2][0] *= v[0]; _mat[3][0] *= v[0];
+    _mat[0][1] *= v[1]; _mat[1][1] *= v[1]; _mat[2][1] *= v[1]; _mat[3][1] *= v[1];
+    _mat[0][2] *= v[2]; _mat[1][2] *= v[2]; _mat[2][2] *= v[2]; _mat[3][2] *= v[2];
+}
+
+
+inline void Matrixf::preMultRotate( const Quat& q )
+{
+    if (q.zeroRotation())
+        return;
+    Matrixf r;
+    r.setRotate(q);
+    preMult(r);
+}
+
+inline void Matrixf::postMultRotate( const Quat& q )
+{
+    if (q.zeroRotation())
+        return;
+    Matrixf r;
+    r.setRotate(q);
+    postMult(r);
+}
 
 inline Vec3f operator* (const Vec3f& v, const Matrixf& m )