Fixes Issues #161 and #162

cub/device/dispatch/dispatch_spmv_orig.cuh

@@ -539,7 +539,7 @@ struct DispatchSpmv
 
             // Get max x-dimension of grid
             int max_dim_x;
-            if (CubDebug(error = cudaDeviceGetAttribute(&max_dim_x, cudaDevAttrMaxGridDimX, device_ordinal))) break;;
+            if (CubDebug(error = cudaDeviceGetAttribute(&max_dim_x, cudaDevAttrMaxGridDimX, device_ordinal))) break;
 
             // Total number of spmv work items
             int num_merge_items = spmv_params.num_rows + spmv_params.num_nonzeros;
@@ -607,7 +607,7 @@ struct DispatchSpmv
                 if (CUB_IS_HOST_CODE)
                 {
                     // Init textures
-                    if (CubDebug(error = spmv_params.t_vector_x.BindTexture(spmv_params.d_vector_x))) break;
+                    if (CubDebug(error = spmv_params.t_vector_x.BindTexture(spmv_params.d_vector_x, spmv_params.num_cols * sizeof(ValueT)))) break;
                 }
             #endif
 

cub/iterator/tex_obj_input_iterator.cuh

@@ -157,7 +157,7 @@ public:
     template <typename QualifiedT>
     cudaError_t BindTexture(
         QualifiedT      *ptr,               ///< Native pointer to wrap that is aligned to cudaDeviceProp::textureAlignment
-        size_t          bytes = size_t(-1),         ///< Number of bytes in the range
+        size_t          bytes,              ///< Number of bytes in the range
         size_t          tex_offset = 0)     ///< OffsetT (in items) from \p ptr denoting the position of the iterator
     {
         this->ptr = const_cast<typename RemoveQualifiers<QualifiedT>::Type *>(ptr);

cub/iterator/tex_ref_input_iterator.cuh

@@ -99,13 +99,13 @@ struct CUB_DEPRECATED IteratorTexRef
         static TexRef ref;
 
         /// Bind texture
-        static cudaError_t BindTexture(void *d_in, size_t &offset)
+        static cudaError_t BindTexture(void *d_in, size_t &bytes, size_t &offset)
         {
             if (d_in)
             {
                 cudaChannelFormatDesc tex_desc = cudaCreateChannelDesc<TextureWord>();
                 ref.channelDesc = tex_desc;
-                return (CubDebug(cudaBindTexture(&offset, ref, d_in)));
+                return (CubDebug(cudaBindTexture(&offset, ref, d_in, bytes)));
             }
 
             return cudaSuccess;
@@ -275,12 +275,12 @@ public:
     template <typename QualifiedT>
     cudaError_t BindTexture(
         QualifiedT      *ptr,                   ///< Native pointer to wrap that is aligned to cudaDeviceProp::textureAlignment
-        size_t          /*bytes*/ = size_t(-1), ///< Number of bytes in the range
+        size_t          bytes,                  ///< Number of bytes in the range
         size_t          tex_offset = 0)         ///< OffsetT (in items) from \p ptr denoting the position of the iterator
     {
         this->ptr = const_cast<typename RemoveQualifiers<QualifiedT>::Type *>(ptr);
         size_t offset;
-        cudaError_t retval = TexId::BindTexture(this->ptr + tex_offset, offset);
+        cudaError_t retval = TexId::BindTexture(this->ptr + tex_offset, bytes, offset);
         this->tex_offset = (difference_type) (offset / sizeof(QualifiedT));
         return retval;
     }

experimental/spmv_compare.cu

@@ -193,7 +193,7 @@ float TestGpuCsrIoProxy(
 
     typedef TexRefInputIterator<ValueT, 1234, int> TexItr;
     TexItr x_itr;
-    CubDebugExit(x_itr.BindTexture(params.d_vector_x));
+    CubDebugExit(x_itr.BindTexture(params.d_vector_x, params.num_cols * sizeof(ValueT)));
 
     // Get device ordinal
     int device_ordinal;