diff --git a/src/linmath.h b/src/linmath.h
index b1a2a8e..4ce880f 100644
--- a/src/linmath.h
+++ b/src/linmath.h
@@ -17,6 +17,7 @@
 #define REPEAT_2(FN) REPEAT_1(FN) FN(1)
 #define REPEAT_3(FN) REPEAT_2(FN) FN(2)
 #define REPEAT_4(FN) REPEAT_3(FN) FN(3)
+#define REPEAT_4N(FN) REPEAT_3(FN) FN(3)
 
 #define add(i) r[i] = a[i] + b[i];
 #define set(i) r[i] = v;
@@ -130,11 +131,7 @@ LINMATH_H_FUNC void vec4_reflect(vec4 r, vec4 const v, vec4 const n) {
 
 typedef vec4 mat4x4[4];
 
-LINMATH_H_FUNC void mat4x4_getPos(vec3 r, mat4x4 const M) {
-  r[0] = M[3][0];
-  r[1] = M[3][1];
-  r[2] = M[3][2];
-}
+LINMATH_H_FUNC float *mat4x4_getPos(mat4x4 M) { return M[3]; }
 
 LINMATH_H_FUNC void mat4x4_identity(mat4x4 M) {
   int i, j;
@@ -198,14 +195,36 @@ LINMATH_H_FUNC void mat4x4_mul(mat4x4 M, mat4x4 const a, mat4x4 const b) {
     }
   mat4x4_dup(M, temp);
 }
+
 LINMATH_H_FUNC void mat4x4_mul_vec4(vec4 r, mat4x4 const M, vec4 const v) {
-  int i, j;
-  for (j = 0; j < 4; ++j) {
-    r[j] = 0.f;
-    for (i = 0; i < 4; ++i)
-      r[j] += M[i][j] * v[i];
-  }
+  r[0] = M[0][0] * v[0];
+  r[0] += M[1][0] * v[1];
+  r[0] += M[2][0] * v[2];
+  r[0] += M[3][0] * v[3];
+
+  r[1] = M[0][1] * v[0];
+  r[1] += M[1][1] * v[1];
+  r[1] += M[2][1] * v[2];
+  r[1] += M[3][1] * v[3];
+
+  r[2] = M[0][2] * v[0];
+  r[2] += M[1][2] * v[1];
+  r[2] += M[2][2] * v[2];
+  r[2] += M[3][2] * v[3];
+
+  r[3] = M[0][3] * v[0];
+  r[3] += M[1][3] * v[1];
+  r[3] += M[2][3] * v[2];
+  r[3] += M[3][3] * v[3];
 }
+// LINMATH_H_FUNC void mat4x4_mul_vec4(vec4 r, mat4x4 const M, vec4 const v) {
+//   int i, j;
+//   for (j = 0; j < 4; ++j) {
+//     r[j] = 0.f;
+//     for (i = 0; i < 4; ++i)
+//       r[j] += M[i][j] * v[i];
+//   }
+// }
 LINMATH_H_FUNC void mat4x4_translate(mat4x4 T, float x, float y, float z) {
   mat4x4_identity(T);
   T[3][0] = x;