:github_url: https://github.com/pytorch/pytorch


.. _program_listing_file_aten_src_ATen_cudnn_Utils.h:

Program Listing for File Utils.h
================================

|exhale_lsh| :ref:`Return to documentation for file <file_aten_src_ATen_cudnn_Utils.h>` (``aten/src/ATen/cudnn/Utils.h``)

.. |exhale_lsh| unicode:: U+021B0 .. UPWARDS ARROW WITH TIP LEFTWARDS

.. code-block:: cpp

   #pragma once
   
   #include <ATen/core/Tensor.h>
   #include <ATen/cuda/Exceptions.h>
   #include <ATen/cudnn/Handle.h>
   #include <ATen/cudnn/cudnn-wrapper.h>
   
   namespace at::native {
   
   // cuDNN has a buggy check for tensor being contiguous (that is, it does
   // not ignore stride for dimension that is equal to 0).  This function
   // makes tensors which have zero stride contiguous, by setting the
   // strides to 1 as cuDNN likes.
   inline Tensor contiguousIfZeroInStrides(const Tensor& t) {
     for (auto s : t.strides()) {
       if (s == 0)
         return t.contiguous();
     }
     return t;
   }
   
   } // namespace at::native