fix bloom chain mip offset

src/gpu/gpu_shared.cgh

@@ -95,7 +95,7 @@ Struct(G_IndexBufferDesc)
 //~ Resource dereference
 
 #if IsGpu
-  // TODO: Add explicit uniform dereference functions, since on AMD hardware
+  // TODO: Add explicit uniform-dereference variations, since on AMD hardware
   // non-uniform is slower and there are some shader-compilation issues in older
   // driver versions
 

src/pp/pp_vis/pp_vis_gpu.g

@@ -1111,7 +1111,7 @@ ComputeShader(V_BloomDownCS)
   Texture2D<Vec4> bloom_up = (
     is_first_pass ?
     G_Deref(frame.screen, Texture2D<Vec4>) :
-    G_Deref(frame.bloom_chain, Texture2D<Vec4>, mip_idx - 1)
+    G_Deref(frame.bloom_chain, Texture2D<Vec4>, mip_idx - 2)
   );
 
   Vec2 down_dims = G_Count2D(bloom_down);