Small tweaks:

- Math functions return results in 32-bit int, cast back to short is done only where necessary
- Changed GTE screen distance to obtain an exact 90 degree horizontal FOV in 4:3 mode

display.c

@@ -40,7 +40,8 @@ MATRIX light_dirs = {
 MATRIX vp_matrix;
 u_long *curOT;
 
-// Scale X coordinates to correct the aspect ratio for the chosen resolution
+// We're using a virtual screen resolution of 320x240, which is then scaled according to display resolution, 
+// video mode and widescreen options to maintain a consistent aspect ratio.
 VECTOR aspect_scale = { SCREENWIDTH * ONE / 320, ONE, ONE };
 
 u_short polyCount;
@@ -53,7 +54,10 @@ void display_init()
 
 	// Initialize GTE
 	InitGeom();
-	gte_SetGeomScreen(180);	// Screen depth for FOV control. Determines the distance of the camera to the near plane.
+	
+	// Screen depth for FOV control. Determines the distance of the camera to the near plane.
+	// Setting this to half the virtual screen width gives us an exact 90 degree horizontal FOV when playing in 4:3 mode.
+	gte_SetGeomScreen(160);
 
 	// Start the display in progressive mode
 	int screenHeight;

qmath.h

@@ -3,18 +3,18 @@
 
 MATRIX *RotMatrixQ(SVECTOR *r, MATRIX *m);
 
-INLINE short m_dot12s(const SVECTOR a, const SVECTOR b)
+INLINE int m_dot12s(const SVECTOR a, const SVECTOR b)
 {
-    return (short)(((int)a.vx * b.vx) >> 12) + (((int)a.vy * b.vy) >> 12) + (((int)a.vz * b.vz) >> 12);
+    return (((int)a.vx * b.vx) >> 12) + (((int)a.vy * b.vy) >> 12) + (((int)a.vz * b.vz) >> 12);
 }
 
 // TODO: worth a benchmark: is it faster to copy these vectors and use them from the stack, or to do six pointer dereferences?
-INLINE short m_pointPlaneDist2(const VECTOR point2, const SVECTOR normal12, int dist2)
+INLINE int m_pointPlaneDist2(const VECTOR point2, const SVECTOR normal12, int dist2)
 {
     int x = ((int)point2.vx * normal12.vx) >> 12;
     int y = ((int)point2.vy * normal12.vy) >> 12;
     int z = ((int)point2.vz * normal12.vz) >> 12;
-    return (short)(x + y + z - dist2);
+    return (x + y + z - dist2);
 }
 
 #endif // __QMATH_H__

world.c

@@ -274,7 +274,7 @@ static INLINE short world_pointPlaneDist(const VECTOR *point, const ps1bsp_plane
     if (plane->type < 3)
         return (short)(((int*)point)[plane->type] - plane->dist);
     
-    return m_pointPlaneDist2(*point, plane->normal, plane->dist);
+    return (short)m_pointPlaneDist2(*point, plane->normal, plane->dist);
 }
 
 static INLINE short world_planeDot(const SVECTOR *dir, const ps1bsp_plane_t *plane)
@@ -283,7 +283,7 @@ static INLINE short world_planeDot(const SVECTOR *dir, const ps1bsp_plane_t *pla
     if (plane->type < 3)
         return ((short*)dir)[plane->type];
 
-    return m_dot12s(*dir, plane->normal);
+    return (short)m_dot12s(*dir, plane->normal);
 }
 
 static void world_drawface(const world_t *world, const ps1bsp_face_t *face)