[FEA] Add method to copy device_buffer back to host memory

include/rmm/device_buffer.hpp

@@ -383,24 +383,6 @@ class device_buffer {
    *-------------------------------------------------------------------------**/
   mr::device_memory_resource* memory_resource() const noexcept { return _mr; }
 
-  /**--------------------------------------------------------------------------*
-   * @brief Copies rmm::device_buffer to a preallocated host buffer.
-   *-------------------------------------------------------------------------**/
-  void copy_to_host(void* host_buffer, cudaStream_t stream = 0) const {
-    if (host_buffer == nullptr) {
-      throw std::runtime_error{"Cannot copy to `nullptr`."};
-    }
-    set_stream(stream);
-    cudaError_t err = cudaMemcpyAsync(host_buffer,
-                                      _data,
-                                      _size,
-                                      cudaMemcpyDeviceToHost,
-                                      this->stream());
-    if (err != cudaSuccess) {
-      throw std::runtime_error{"Failed to copy to host."};
-    }
-  }
-
  private:
   void* _data{nullptr};     ///< Pointer to device memory allocation
   std::size_t _size{};      ///< Requested size of the device memory allocation
@@ -410,4 +392,22 @@ class device_buffer {
       mr::get_default_resource()};  ///< The memory resource used to
                                     ///< allocate/deallocate device memory
 };
+
+/**--------------------------------------------------------------------------*
+ * @brief Copies rmm::device_buffer to a preallocated host buffer.
+ *-------------------------------------------------------------------------**/
+void copy_to_host(device_buffer& db, void* hb, cudaStream_t stream = 0) {
+  if (hb == nullptr) {
+    throw std::runtime_error{"Cannot copy to `nullptr`."};
+  }
+  db.set_stream(stream);
+  cudaError_t err = cudaMemcpyAsync(hb,
+                                    db.data(),
+                                    db.size(),
+                                    cudaMemcpyDeviceToHost,
+                                    db.stream());
+  if (err != cudaSuccess) {
+    throw std::runtime_error{"Failed to copy to host."};
+  }
+}
 }  // namespace rmm

python/rmm/_lib/device_buffer.pxd

@@ -13,7 +13,8 @@ cdef extern from "rmm/device_buffer.hpp" namespace "rmm" nogil:
         void* data()
         size_t size()
         size_t capacity()
-        void copy_to_host(void* host_buffer) except *
+
+    void copy_to_host(device_buffer& db, void* hb) except *
 
 cdef class DeviceBuffer:
     cdef unique_ptr[device_buffer] c_obj

python/rmm/_lib/device_buffer.pyx

@@ -56,7 +56,7 @@ cdef class DeviceBuffer:
         cdef bytes b = PyBytes_FromStringAndSize(NULL, self.c_size())
         cdef void* p = <void*>PyBytes_AS_STRING(b)
 
-        self.c_obj.get()[0].copy_to_host(p)
+        copy_to_host(self.c_obj.get()[0], p)
 
         return b