diff --git a/include/cglm-affine.h b/include/cglm-affine.h
index 5befb8a..0b7f7b9 100644
--- a/include/cglm-affine.h
+++ b/include/cglm-affine.h
@@ -74,25 +74,23 @@ glm_translate_z(mat4 m, float to) {
 CGLM_INLINE
 void
 glm_scale_to(mat4 m, vec3 v, mat4 dest) {
-  dest[0][0] = m[0][0] * v[0];
-  dest[1][1] = m[1][1] * v[1];
-  dest[2][2] = m[2][2] * v[2];
+  glm_vec_scale(m[0], v[0], dest[0]);
+  glm_vec_scale(m[1], v[1], dest[1]);
+  glm_vec_scale(m[2], v[2], dest[2]);
+
+  glm_vec_dup(m[3], dest[3]);
 }
 
 CGLM_INLINE
 void
 glm_scale(mat4 m, vec3 v) {
-  m[0][0] *= v[0];
-  m[1][1] *= v[1];
-  m[2][2] *= v[2];
+  glm_scale_to(m, v, m);
 }
 
 CGLM_INLINE
 void
 glm_scale1(mat4 m, float s) {
-  m[0][0] *= s;
-  m[1][1] *= s;
-  m[2][2] *= s;
+  glm_scale_to(m, (vec3){s, s, s}, m);
 }
 
 CGLM_INLINE
diff --git a/include/cglm-vec.h b/include/cglm-vec.h
index 95e86a1..1cc838e 100644
--- a/include/cglm-vec.h
+++ b/include/cglm-vec.h
@@ -10,6 +10,23 @@
 
 #include "cglm.h"
 
+CGLM_INLINE
+void
+glm_vec_dup(vec3 a, vec3 dest) {
+  dest[0] = a[0];
+  dest[1] = a[1];
+  dest[2] = a[2];
+}
+
+CGLM_INLINE
+void
+glm_vec_dup4(vec4 a, vec4 dest) {
+  dest[0] = a[0];
+  dest[1] = a[1];
+  dest[2] = a[2];
+  dest[3] = a[3];
+}
+
 CGLM_INLINE
 float
 glm_vec_dot(vec3 a, vec3 b) {