diff --git a/src/mat2.rs b/src/mat2.rs
index 8d210ac..c91bc5f 100644
--- a/src/mat2.rs
+++ b/src/mat2.rs
@@ -70,7 +70,8 @@ pub impl<T:Copy Float> Mat2<T> {
      * ~~~
      */
     #[inline(always)]
-    static pure fn from_cols(c0: Vec2<T>, c1: Vec2<T>) -> Mat2<T> {
+    static pure fn from_cols(c0: Vec2<T>,
+                             c1: Vec2<T>) -> Mat2<T> {
         Mat2 { x: move c0,
                y: move c1 }
     }
diff --git a/src/mat3.rs b/src/mat3.rs
index 6501962..dd9ebd7 100644
--- a/src/mat3.rs
+++ b/src/mat3.rs
@@ -70,16 +70,18 @@ pub impl<T:Copy Float> Mat3<T> {
      * ~~~
      *        c0     c1     c2
      *     +------+------+------+
-     *  r0 | c0.x | c1.y | c2.z |
+     *  r0 | c0.x | c1.x | c2.x |
      *     +------+------+------+
-     *  r1 | c0.x | c1.y | c2.z |
+     *  r1 | c0.y | c1.y | c2.y |
      *     +------+------+------+
-     *  r2 | c0.x | c1.y | c2.z |
+     *  r2 | c0.z | c1.z | c2.z |
      *     +------+------+------+
      * ~~~
      */
     #[inline(always)]
-    static pure fn from_cols(c0: Vec3<T>, c1: Vec3<T>, c2: Vec3<T>) -> Mat3<T> {
+    static pure fn from_cols(c0: Vec3<T>,
+                             c1: Vec3<T>,
+                             c2: Vec3<T>) -> Mat3<T> {
         Mat3 { x: move c0,
                y: move c1,
                z: move c2 }
@@ -307,9 +309,17 @@ pub impl<T:Copy Float> Mat3<T>: Matrix<T, Vec3<T>> {
     
     #[inline(always)]
     pure fn mul_m(&self, other: &Mat3<T>) -> Mat3<T> {
-        Mat3::new(self.row(0).dot(&other.col(0)), self.row(1).dot(&other.col(0)), self.row(2).dot(&other.col(0)),
-                  self.row(0).dot(&other.col(1)), self.row(1).dot(&other.col(1)), self.row(2).dot(&other.col(1)),
-                  self.row(0).dot(&other.col(2)), self.row(1).dot(&other.col(2)), self.row(2).dot(&other.col(2)))
+        Mat3::new(self.row(0).dot(&other.col(0)),
+                  self.row(1).dot(&other.col(0)),
+                  self.row(2).dot(&other.col(0)),
+            
+                  self.row(0).dot(&other.col(1)),
+                  self.row(1).dot(&other.col(1)),
+                  self.row(2).dot(&other.col(1)),
+                  
+                  self.row(0).dot(&other.col(2)),
+                  self.row(1).dot(&other.col(2)),
+                  self.row(2).dot(&other.col(2)))
     }
     
     pure fn dot(&self, other: &Mat3<T>) -> T {
diff --git a/src/mat4.rs b/src/mat4.rs
index 59dc689..6833671 100644
--- a/src/mat4.rs
+++ b/src/mat4.rs
@@ -86,7 +86,10 @@ pub impl<T:Copy Float> Mat4<T> {
      * ~~~
      */
     #[inline(always)]
-    static pure fn from_cols(c0: Vec4<T>, c1: Vec4<T>, c2: Vec4<T>, c3: Vec4<T>) -> Mat4<T> {
+    static pure fn from_cols(c0: Vec4<T>,
+                             c1: Vec4<T>,
+                             c2: Vec4<T>,
+                             c3: Vec4<T>) -> Mat4<T> {
         Mat4 { x: move c0,
                y: move c1,
                z: move c2,
@@ -239,13 +242,26 @@ pub impl<T:Copy Float Sign> Mat4<T>: Matrix<T, Vec4<T>> {
     
     #[inline(always)]
     pure fn mul_m(&self, other: &Mat4<T>) -> Mat4<T> {
-        // Surprisingly when building with optimisation turned on this is actually
-        // faster than writing out the matrix multiplication in expanded form.
-        // If you don't believe me, see ./test/performance/matrix_mul.rs
-        Mat4::new(self.row(0).dot(&other.col(0)), self.row(1).dot(&other.col(0)), self.row(2).dot(&other.col(0)), self.row(3).dot(&other.col(0)),
-                  self.row(0).dot(&other.col(1)), self.row(1).dot(&other.col(1)), self.row(2).dot(&other.col(1)), self.row(3).dot(&other.col(1)),
-                  self.row(0).dot(&other.col(2)), self.row(1).dot(&other.col(2)), self.row(2).dot(&other.col(2)), self.row(3).dot(&other.col(2)),
-                  self.row(0).dot(&other.col(3)), self.row(1).dot(&other.col(3)), self.row(2).dot(&other.col(3)), self.row(3).dot(&other.col(3)))
+        Mat4::new(self.row(0).dot(&other.col(0)),
+                  self.row(1).dot(&other.col(0)),
+                  self.row(2).dot(&other.col(0)),
+                  self.row(3).dot(&other.col(0)),
+                  
+                  self.row(0).dot(&other.col(1)),
+                  self.row(1).dot(&other.col(1)),
+                  self.row(2).dot(&other.col(1)),
+                  self.row(3).dot(&other.col(1)),
+                  
+                  self.row(0).dot(&other.col(2)),
+                  self.row(1).dot(&other.col(2)),
+                  self.row(2).dot(&other.col(2)),
+                  self.row(3).dot(&other.col(2)),
+                  
+                  self.row(0).dot(&other.col(3)),
+                  self.row(1).dot(&other.col(3)),
+                  self.row(2).dot(&other.col(3)),
+                  self.row(3).dot(&other.col(3)))
+        
     }
     
     pure fn dot(&self, other: &Mat4<T>) -> T {
@@ -485,8 +501,7 @@ pub impl<T:Copy Float Sign> Mat4<T>: MutableMatrix<T, Vec4<T>> {
     }
 }
 
-pub impl<T> Mat4<T>: Matrix4<T, Vec4<T>> {
-}
+pub impl<T> Mat4<T>: Matrix4<T, Vec4<T>> {}
 
 pub impl<T:Copy Float> Mat4<T>: Neg<Mat4<T>> {
     #[inline(always)]