diff --git a/GL/draw.c b/GL/draw.c
index b6e5086..a8dade5 100644
--- a/GL/draw.c
+++ b/GL/draw.c
@@ -615,9 +615,13 @@ static void transform(SubmissionTarget* target) {
     TRACE();
 
     /* Perform modelview transform, storing W */
-    Vertex* vertex = _glSubmissionTargetStart(target);
+    Vertex* it = _glSubmissionTargetStart(target);
+    int count  = target->count;
 
-    TransformVertices(vertex, target->count);
+    for(int i = 0; i < count; ++i, ++it) {
+        TransformVertex(it->xyz[0], it->xyz[1], it->xyz[2], it->w, 
+                        it->xyz, &it->w);
+    }
 }
 
 static void mat_transform_normal3(const float* xyz, const float* xyzOut, const uint32_t count, const uint32_t inStride, const uint32_t outStride) {
diff --git a/GL/platforms/sh4.h b/GL/platforms/sh4.h
index 6b9bad0..1139283 100644
--- a/GL/platforms/sh4.h
+++ b/GL/platforms/sh4.h
@@ -124,28 +124,6 @@ GL_FORCE_INLINE void TransformVertex(float x, float y, float z, float w, float*
     *ow = __w;
 }
 
-static inline void TransformVertices(Vertex* vertices, const int count) {
-    Vertex* it = vertices;
-    for(int i = 0; i < count; ++i, ++it) {
-        register float __x __asm__("fr12") = (it->xyz[0]);
-        register float __y __asm__("fr13") = (it->xyz[1]);
-        register float __z __asm__("fr14") = (it->xyz[2]);
-        register float __w __asm__("fr15") = (it->w);
-
-        __asm__ __volatile__(
-            "fldi1 fr15\n"
-            "ftrv   xmtrx,fv12\n"
-            : "=f" (__x), "=f" (__y), "=f" (__z), "=f" (__w)
-            : "0" (__x), "1" (__y), "2" (__z), "3" (__w)
-        );
-
-        it->xyz[0] = __x;
-        it->xyz[1] = __y;
-        it->xyz[2] = __z;
-        it->w = __w;
-    }
-}
-
 void InitGPU(_Bool autosort, _Bool fsaa);
 
 static inline size_t GPUMemoryAvailable() {
diff --git a/GL/platforms/software.c b/GL/platforms/software.c
index 9a2b4e5..57940b6 100644
--- a/GL/platforms/software.c
+++ b/GL/platforms/software.c
@@ -619,34 +619,17 @@ void TransformVec4(float* v) {
     FASTCPY(v, ret, sizeof(float) * 4);
 }
 
-void TransformVertices(Vertex* vertices, const int count) {
-    float ret[4];
-    for(int i = 0; i < count; ++i, ++vertices) {
-        ret[0] = vertices->xyz[0];
-        ret[1] = vertices->xyz[1];
-        ret[2] = vertices->xyz[2];
-        ret[3] = 1.0f;
-
-        TransformVec4(ret);
-
-        vertices->xyz[0] = ret[0];
-        vertices->xyz[1] = ret[1];
-        vertices->xyz[2] = ret[2];
-        vertices->w = ret[3];
-    }
-}
-
 void TransformVertex(float x, float y, float z, float w, float* oxyz, float* ow) {
-    float ret[4];
-    ret[0] = x;
-    ret[1] = y;
-    ret[2] = z;
-    ret[3] = w;
+    float vec[4], ret[4];
+    vec[0] = x;
+    vec[1] = y;
+    vec[2] = z;
+    vec[3] = w;
 
-    TransformVec4(ret);
+    TransformVec4NoMod(vec, ret);
 
     oxyz[0] = ret[0];
     oxyz[1] = ret[1];
     oxyz[2] = ret[2];
-    *ow = ret[3];
+    *ow     = ret[3];
 }
diff --git a/GL/platforms/software.h b/GL/platforms/software.h
index cc89f97..744a9bc 100644
--- a/GL/platforms/software.h
+++ b/GL/platforms/software.h
@@ -52,7 +52,6 @@ static inline void TransformNormalNoMod(const float* xIn, float* xOut) {
     (void) xOut;
 }
 
-void TransformVertices(Vertex* vertices, const int count);
 void TransformVertex(float x, float y, float z, float w, float* oxyz, float* ow);
 
 void InitGPU(_Bool autosort, _Bool fsaa);