diff --git a/cuBQL/builder/cuda.h b/cuBQL/builder/cuda.h
index 76a7f27..f15dad8 100644
--- a/cuBQL/builder/cuda.h
+++ b/cuBQL/builder/cuda.h
@@ -12,7 +12,6 @@
 # include <mutex>
 
 namespace cuBQL {
-
   // ------------------------------------------------------------------
   /*! defines a 'memory resource' that can be used for allocating gpu
     memory; this allows the user to switch between usign
diff --git a/cuBQL/builder/cuda/builder_common.h b/cuBQL/builder/cuda/builder_common.h
index 27d50df..e71b901 100644
--- a/cuBQL/builder/cuda/builder_common.h
+++ b/cuBQL/builder/cuda/builder_common.h
@@ -191,11 +191,6 @@ namespace cuBQL {
       
       typename int_type_of<scalar_t>::type lower[box_t::numDims];
       typename int_type_of<scalar_t>::type upper[box_t::numDims];
-      // int32_t lower[box_t::numDims];
-      // int32_t upper[box_t::numDims];
-      
-      // inline static __device__ int32_t encode(float f);
-      // inline static __device__ float   decode(int32_t bits);
     };
     
     template<typename box_t>
@@ -203,14 +198,7 @@ namespace cuBQL {
     AtomicBox<box_t>::get_center(int dim) const
     {
       return (get_lower(dim)+get_upper(dim))/(AtomicBox<box_t>::scalar_t)2;
-      // return 0.5f*(decode(lower[dim])+decode(upper[dim]));
     }
-    // template<typename box_t>
-    // inline __device__ float AtomicBox<box_t>::get_center(int dim) const
-    // {
-    //   return 0.5f*(get_lower(dim)+get_upper(dim));
-    //   // return 0.5f*(decode(lower[dim])+decode(upper[dim]));
-    // }
 
     template<typename box_t>
     inline __device__ box_t AtomicBox<box_t>::make_box() const
@@ -225,23 +213,6 @@ namespace cuBQL {
       return box;
     }
     
-    // template<typename box_t>
-    // inline __device__ int32_t AtomicBox<box_t>::encode(float f)
-    // {
-    //   const int32_t sign = 0x80000000;
-    //   int32_t bits = __float_as_int(f);
-    //   if (bits & sign) bits ^= 0x7fffffff;
-    //   return bits;
-    // }
-      
-    // template<typename box_t>
-    // inline __device__ float AtomicBox<box_t>::decode(int32_t bits)
-    // {
-    //   const int32_t sign = 0x80000000;
-    //   if (bits & sign) bits ^= 0x7fffffff;
-    //   return __int_as_float(bits);
-    // }
-    
     template<typename box_t>
     inline __device__ void AtomicBox<box_t>::set_empty()
     {
diff --git a/cuBQL/builder/cuda/gpu_builder.h b/cuBQL/builder/cuda/gpu_builder.h
index 37aee9f..18365f7 100644
--- a/cuBQL/builder/cuda/gpu_builder.h
+++ b/cuBQL/builder/cuda/gpu_builder.h
@@ -9,7 +9,6 @@
 #include "cuBQL/builder/cuda/elh_builder.h"
 
 namespace cuBQL {
-
   template<typename T, int D>
   struct is3f { enum { value = false }; };
   template<>
@@ -69,7 +68,6 @@ namespace cuBQL {
       bvh.primIDs = 0;
     }
   }
-}
-
-
+  
+} // ::cuBQL
 
diff --git a/cuBQL/builder/cuda/instantiate_builders.cu b/cuBQL/builder/cuda/instantiate_builders.cu
index b54d5ef..7d9c667 100644
--- a/cuBQL/builder/cuda/instantiate_builders.cu
+++ b/cuBQL/builder/cuda/instantiate_builders.cu
@@ -88,4 +88,3 @@ CUBQL_INSTANTIATE_WIDE_BVH(float,3,8)
 CUBQL_INSTANTIATE_WIDE_BVH(float,3,16)
 #endif
   
- 
diff --git a/cuBQL/bvh.h b/cuBQL/bvh.h
index 73fe0cd..1ab1941 100644
--- a/cuBQL/bvh.h
+++ b/cuBQL/bvh.h
@@ -6,9 +6,9 @@
 #include "cuBQL/math/common.h"
 #include "cuBQL/math/vec.h"
 #include "cuBQL/math/box.h"
-#ifdef __CUDACC__
-# include <cuda.h>
-#endif
+// #ifdef __CUDACC__
+// # include <cuda.h>
+// #endif
 
 namespace cuBQL {
 
@@ -156,8 +156,10 @@ namespace cuBQL {
 #ifdef __CUDACC__
 # include "cuBQL/builder/cuda.h"
 #endif
-# include "cuBQL/builder/cpu.h"
 
+#if XXX
+# include "cuBQL/builder/cpu.h"
+#endif
 
 
   
diff --git a/cuBQL/math/common.h b/cuBQL/math/common.h
index 710786e..a3063dd 100644
--- a/cuBQL/math/common.h
+++ b/cuBQL/math/common.h
@@ -30,6 +30,8 @@
 #  include <hip/hip_runtime.h>
 #elif defined(__CUDACC__)
 #  include <cuda_runtime.h>
+// on cuda 13.2/gcc 15.2/ubuntu 25.10 we need to include that before cuda.h gets included
+#  include <cub/cub.cuh>
 #endif
 
 #ifdef _WIN32