diff --git a/assets/shaders/pbr/render.wgsl b/assets/shaders/pbr/render.wgsl
index 51f0df64..0abbc269 100644
--- a/assets/shaders/pbr/render.wgsl
+++ b/assets/shaders/pbr/render.wgsl
@@ -1,6 +1,6 @@
-#include "dither.wgsl"
-#include "tonemap.wgsl"
-#include "atmosphere.wgsl"
+#include "../dither.wgsl"
+#include "../tonemap.wgsl"
+#include "../atmosphere.wgsl"
 
 fn fresnelSchlick(cosTheta: f32, F0: vec3<f32>) -> vec3<f32> {
     return F0 + (1.0 - F0) * pow(clamp(1.0 - cosTheta, 0.0, 1.0), 5.0);
diff --git a/assets/shaders/terrain.frag.wgsl b/assets/shaders/terrain/terrain.frag.wgsl
similarity index 98%
rename from assets/shaders/terrain.frag.wgsl
rename to assets/shaders/terrain/terrain.frag.wgsl
index 21d2db3e..4836fbc0 100644
--- a/assets/shaders/terrain.frag.wgsl
+++ b/assets/shaders/terrain/terrain.frag.wgsl
@@ -1,6 +1,6 @@
-#include "render_params.wgsl"
-#include "shadow.wgsl"
-#include "pbr/render.wgsl"
+#include "../render_params.wgsl"
+#include "../shadow.wgsl"
+#include "../pbr/render.wgsl"
 
 struct FragmentOutput {
     @location(0) out_color: vec4<f32>,
diff --git a/assets/shaders/terrain.vert.wgsl b/assets/shaders/terrain/terrain.vert.wgsl
similarity index 75%
rename from assets/shaders/terrain.vert.wgsl
rename to assets/shaders/terrain/terrain.vert.wgsl
index 030e7c8d..fd8b1c69 100644
--- a/assets/shaders/terrain.vert.wgsl
+++ b/assets/shaders/terrain/terrain.vert.wgsl
@@ -1,4 +1,5 @@
-#include "render_params.wgsl"
+#include "../render_params.wgsl"
+#include "unpack.wgsl"
 
 struct Uniforms {
     u_view_proj: mat4x4<f32>,
@@ -36,20 +37,6 @@ normal: vec3(self.cell_size * scale as f32, 0.0, hx - height)
                             .normalize(),
 */
 
-const MAX_HEIGHT: f32 = 1024.0;
-const MAX_DIFF: f32 = 32.0;
-
-fn unpack_height(h: u32) -> f32 {
-    return ((f32(h) - 32768.0) / 32767.0 ) * MAX_HEIGHT;
-}
-
-fn unpack_diffs(v: u32) -> vec2<f32> {
-    let x = v & 0xFFu;
-    let y = (v & 0xFF00u) >> 8u;
-    return vec2<f32>((f32(x) - 128.0) / 127.0 * MAX_DIFF,
-                     (f32(y) - 128.0) / 127.0 * MAX_DIFF);
-}
-
 @vertex
 fn vert(@builtin(vertex_index) vid: u32,
         @location(0) in_off: vec2<f32>,
@@ -75,19 +62,15 @@ fn vert(@builtin(vertex_index) vid: u32,
 
     let tpos: vec2<i32> = in_position * i32(cdata.lod_pow2) + vec2<i32>(in_off * cdata.inv_cell_size);
 
-    let texLoad: vec2<u32> = textureLoad(t_terraindata, tpos, 0).rg;
-
-    let height: f32 = unpack_height(texLoad.r);
-    let diffs: vec2<f32> = unpack_diffs(texLoad.g);
+    let h_dx_dy: vec3<f32> = unpack(textureLoad(t_terraindata, tpos, 0).r, 1.0);
 
-    let world_pos: vec3<f32> = vec3(vec2<f32>(in_position * i32(cdata.lod_pow2)) * cdata.cell_size + in_off, height);
+    let world_pos: vec3<f32> = vec3(vec2<f32>(in_position * i32(cdata.lod_pow2)) * cdata.cell_size + in_off, h_dx_dy.x);
+    let clip_pos: vec4<f32> = global.u_view_proj * vec4(world_pos, 1.0);
 
     //let dist_to_cam: f32 = length(params.cam_pos.xyz - vec3(pos.xy, 0.0));
     //let transition_alpha: f32 = smoothstep(cdata.distance_lod_cutoff * 0.8, cdata.distance_lod_cutoff, dist_to_cam);
 
-    var out_normal: vec3<f32> = normalize(vec3(diffs.x, diffs.y, cdata.cell_size * 2.0)); // https://stackoverflow.com/questions/49640250/calculate-normals-from-heightmap
-
-    let position: vec4<f32> = global.u_view_proj * vec4(world_pos, 1.0);
+    var out_normal: vec3<f32> = normalize(vec3(h_dx_dy.yz, cdata.cell_size * 2.0)); // https://stackoverflow.com/questions/49640250/calculate-normals-from-heightmap
 
 #ifdef DEBUG
     var debug = 0.0;
@@ -104,5 +87,5 @@ fn vert(@builtin(vertex_index) vid: u32,
                         #ifdef DEBUG
                         debug,
                         #endif
-                        position);
+                        clip_pos);
 }
diff --git a/assets/shaders/terrain/unpack.wgsl b/assets/shaders/terrain/unpack.wgsl
new file mode 100644
index 00000000..3100acad
--- /dev/null
+++ b/assets/shaders/terrain/unpack.wgsl
@@ -0,0 +1,18 @@
+const MAX_HEIGHT: f32 = 1024.0;
+const MAX_DIFF: f32 = 32.0;
+
+fn unpack_height(h: u32) -> f32 {
+    return ((f32(h) - 32768.0) / 32767.0 ) * MAX_HEIGHT;
+}
+
+fn unpack_diffs(v: u32, lod_pow2: f32) -> vec2<f32> {
+    let x = (f32(v >> 8u) - 128.0) / 127.0 * (MAX_DIFF * lod_pow2);
+    let y = (f32(v & 0xFFu) - 128.0) / 127.0 * (MAX_DIFF * lod_pow2);
+    return vec2<f32>(x, y);
+}
+
+fn unpack(v: u32, lod_pow2: f32) -> vec3<f32> {
+    let h = unpack_height(v & 0xFFFFu);
+    let d = unpack_diffs(v >> 16u, lod_pow2);
+    return vec3(h, d);
+}
\ No newline at end of file
diff --git a/engine/src/drawables/terrain.rs b/engine/src/drawables/terrain.rs
index 5af803d2..c327ec1b 100644
--- a/engine/src/drawables/terrain.rs
+++ b/engine/src/drawables/terrain.rs
@@ -50,7 +50,7 @@ impl<const CSIZE: usize, const CRESOLUTION: usize> TerrainRender<CSIZE, CRESOLUT
             w * CRESOLUTION as u32,
             h * CRESOLUTION as u32,
             1,
-            TextureFormat::Rg16Uint,
+            TextureFormat::R32Uint,
         )
         .with_usage(TextureUsages::COPY_DST | TextureUsages::TEXTURE_BINDING)
         .with_fixed_mipmaps(LOD as u32)
@@ -123,17 +123,24 @@ impl<const CSIZE: usize, const CRESOLUTION: usize> TerrainRender<CSIZE, CRESOLUT
         get_left: impl Fn(usize) -> Option<f32>,
     ) -> bool {
         fn pack(height: f32, diffx: f32, diffy: f32) -> [u8; 4] {
-            let a = ((height.clamp(-MAX_HEIGHT, MAX_HEIGHT) / MAX_HEIGHT * i16::MAX as f32
-                + 32768.0) as u16)
-                .to_le_bytes();
+            let h_encoded = ((height.clamp(-MAX_HEIGHT, MAX_HEIGHT) / MAX_HEIGHT * i16::MAX as f32
+                + 32768.0) as u16);
+
+            let dx_encoded: u8;
+            let dy_encoded: u8;
 
             if height >= MAX_HEIGHT || height <= -MAX_HEIGHT {
-                return [a[0], a[1], 128, 128]; // normal is zero if we hit max height
+                dx_encoded = 128;
+                dy_encoded = 128; // normal is zero if we hit max height
+            } else {
+                dx_encoded =
+                    (diffx.clamp(-MAX_DIFF, MAX_DIFF) / MAX_DIFF * i8::MAX as f32 + 128.0) as u8;
+                dy_encoded =
+                    (diffy.clamp(-MAX_DIFF, MAX_DIFF) / MAX_DIFF * i8::MAX as f32 + 128.0) as u8;
             }
 
-            let b = (diffx.clamp(-MAX_DIFF, MAX_DIFF) / MAX_DIFF * i8::MAX as f32 + 128.0) as u8;
-            let c = (diffy.clamp(-MAX_DIFF, MAX_DIFF) / MAX_DIFF * i8::MAX as f32 + 128.0) as u8;
-            [a[0], a[1], b, c]
+            let packed = (dx_encoded as u32) << 24 | (dy_encoded as u32) << 16 | h_encoded as u32;
+            packed.to_le_bytes()
         }
 
         let mut contents = Vec::with_capacity(CRESOLUTION * CRESOLUTION);
@@ -386,10 +393,10 @@ impl PipelineBuilder for TerrainPipeline {
                     .collect::<Vec<_>>(),
                 label: Some("terrain bindgroup layout"),
             });
-        let vert = &mk_module("terrain.vert");
+        let vert = &mk_module("terrain/terrain.vert");
 
         if !self.depth {
-            let frag = &mk_module("terrain.frag");
+            let frag = &mk_module("terrain/terrain.frag");
 
             return gfx.color_pipeline(
                 "terrain",
diff --git a/engine/src/shader.rs b/engine/src/shader.rs
index 2d000862..212795aa 100644
--- a/engine/src/shader.rs
+++ b/engine/src/shader.rs
@@ -134,9 +134,9 @@ fn replace_imports(base: &Path, src: &str, deps: &mut Vec<String>) -> String {
                 let mut p = base.to_path_buf();
                 p.pop();
                 p.push(loc);
-                let mut s = std::fs::read_to_string(p)
-                    .unwrap_or_else(|_| panic!("could not find included file {loc}"));
-                s = replace_imports(base, &s, deps);
+                let mut s = std::fs::read_to_string(&p)
+                    .unwrap_or_else(|_| panic!("could not find included file {loc} for {base:?}"));
+                s = replace_imports(&p, &s, deps);
                 return Cow::Owned(s);
             }
             Cow::Borrowed(x)