applied a fix for https://github.com/NVlabs/cub/issues/162 to cub_sem…

…iring
seunghwak · May 10, 2019 · abb3dc8 · abb3dc8
1 parent e3399a2
commit abb3dc8
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 6 deletions.
diff --git a/cpp/nvgraph/external/cub_semiring/device/dispatch/dispatch_spmv_orig.cuh b/cpp/nvgraph/external/cub_semiring/device/dispatch/dispatch_spmv_orig.cuh
@@ -762,7 +762,7 @@ struct DispatchSpmv
 
 #if (CUB_PTX_ARCH == 0)
             // 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
 
             if (search_grid_size < sm_count)

diff --git a/cpp/nvgraph/external/cub_semiring/iterator/tex_obj_input_iterator.cuh b/cpp/nvgraph/external/cub_semiring/iterator/tex_obj_input_iterator.cuh
@@ -161,7 +161,7 @@ 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);

diff --git a/cpp/nvgraph/external/cub_semiring/iterator/tex_ref_input_iterator.cuh b/cpp/nvgraph/external/cub_semiring/iterator/tex_ref_input_iterator.cuh
@@ -91,13 +91,13 @@ struct 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;
@@ -245,12 +245,12 @@ 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;
     }