// j <- random integer 0 <= j <= i
      size_t j = random::New64() % (i + 1);
      // swap values at i, j
      T tmp = tensor_flat(i);
      tensor_flat(i) = tensor_flat(j);
      tensor_flat(j) = tmp;
    }
  }

  Scope scope_;
};

TEST_F(NNGradTest, SoftmaxGrad) {
  TensorShape shape({32, 10});
  auto x = Placeholder(scope_, DT_FLOAT, Placeholder::Shape(shape));

      ASSERT_EQ(errors::OK, s.code()) << s.message();
      ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr ctx_;

 public:

      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr immediate_execution_ctx_;
  GradientRegistry registry_;

      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr immediate_execution_ctx_;
  GradientRegistry registry_;

      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr immediate_execution_ctx_;
  GradientRegistry registry_;

        [TensorFloat-32](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/).
        Specifically, inputs to such ops are rounded from 23 bits of precision
        to 10 bits of precision. This is unlikely to cause issues in practice
        for deep learning models. In some cases, TensorFloat-32 is also used for
        complex64 ops. TensorFloat-32 can be disabled by running