Conv node bug, cached state was incoherent (microsoft#10041)

* Moved the init earlier to keep the cache coherent
* Move setting of w_desc later, and zero shape check later to catch all cacheable changes.
* Add comment

onnxruntime/core/providers/cuda/nn/conv.cc

@@ -176,9 +176,6 @@ Status Conv<T>::UpdateState(OpKernelContext* context, bool bias_expected) const
     s_.slice_axes = slice_axes;
 
     s_.Y = context->Output(0, TensorShape(s_.y_dims));
-    if (s_.Y->Shape().Size() == 0) {
-      return Status::OK();
-    }
     if (post_slicing_required) {
       // Post slicing needed. Create and fill in the Conv results in an intermediate buffer.
       s_.memory_for_cudnn_conv_results = GetScratchBuffer<void>(TensorShape(y_dims_with_adjusted_pads).Size() * s_.element_size);
@@ -206,9 +203,14 @@ Status Conv<T>::UpdateState(OpKernelContext* context, bool bias_expected) const
       dilations.push_back(1);
     }
 
-    if (w_dims_changed) {
+    if (w_dims_changed)
       ORT_RETURN_IF_ERROR(s_.w_desc.Set(w_dims, CudnnTensor::GetDataType<CudaT>()));
+
+    // We must delay returning early until here so that the weight dims have been cached properly
+    if (s_.Y->Shape().Size() == 0) {
+      return Status::OK();
     }
+
     ORT_RETURN_IF_ERROR(s_.x_tensor.Set(x_dims_cudnn, CudnnTensor::GetDataType<CudaT>()));
     ORT_RETURN_IF_ERROR(s_.y_tensor.Set(y_dims_cudnn, CudnnTensor::GetDataType<CudaT>()));
     ORT_RETURN_IF_ERROR(s_.conv_desc.Set(kernel_shape.size(), pads, strides, dilations,