gistfile1.txt

diff --git a/torch/csrc/jit/autodiff.cpp b/torch/csrc/jit/autodiff.cpp
index 59eb7ca11..75abe0097 100644
--- a/torch/csrc/jit/autodiff.cpp
+++ b/torch/csrc/jit/autodiff.cpp
@@ -77,7 +77,8 @@ bool isDifferentiable(Node * n) {
     "aten::trunc(Tensor self) -> Tensor",
     "aten::log_softmax(Tensor self, int dim) -> Tensor",
     "aten::avg_pool2d(Tensor self, int[] kernel_size, int[] stride, int[] padding, bool ceil_mode, bool count_include_pad) -> Tensor",
-    "aten::max_pool2d_with_indices(Tensor self, int[] kernel_size, int[] stride, int[] padding, int[] dilation, bool ceil_mode) -> (Tensor, Tensor)"
+    "aten::max_pool2d_with_indices(Tensor self, int[] kernel_size, int[] stride, int[] padding, int[] dilation, bool ceil_mode) -> (Tensor, Tensor)",
+    "aten::thnn_conv2d_forward(Tensor self, Tensor weight, int[] kernel_size, Tensor bias, int[] stride, int[] padding) -> (Tensor, Tensor, Tensor)",
   };
 
   // TODO: add support for the following fusible operators.
@@ -88,7 +89,8 @@ bool isDifferentiable(Node * n) {
 
   if (n->kind() == prim::Constant ||
       n->kind() == prim::AutogradAdd ||
-      n->kind() == prim::ConstantChunk)
+      n->kind() == prim::ConstantChunk ||
+      n->kind() == prim::TupleConstruct)
     return true;
   if (differentiable_ops.find(n))
     return true;
@@ -144,6 +146,17 @@ static std::vector<Value*> gradientForNode(Node* node, ArrayRef<Value*> grad_val
     } else if (node->kind() == prim::AutogradAdd) {
       return {grads.at(0), grads.at(0)};
 
+    } else if (node->kind() == prim::TupleConstruct) {
+      auto graph = node->owningGraph();
+      auto node = graph->createTupleUnpack(grads.at(0).value());
+      graph->insertNode(node);
+      auto outputs = node->outputs();
+      std::vector<SymbolicVariable> sym_outputs;
+      for (auto output : outputs) {
+	sym_outputs.emplace_back(output);
+      }
+      return sym_outputs;
+
     } else if (node->matches("aten::sub(Tensor self, Tensor other, *, Scalar alpha) -> Tensor")) {
       return {grads.at(0), -grads.at(0) * node->namedInput(attr::alpha), nullptr};
 
@@ -395,6 +408,25 @@ static std::vector<Value*> gradientForNode(Node* node, ArrayRef<Value*> grad_val
       });
       return {backward_value->node()->output(0), nullptr, nullptr, nullptr, nullptr, nullptr};
 
+    } else if (node->matches("aten::thnn_conv2d_forward(Tensor self, Tensor weight, int[] kernel_size, Tensor bias, int[] stride, int[] padding) -> (Tensor, Tensor, Tensor)")) {
+      auto graph = node->owningGraph();
+
+      auto convNode = graph->create(aten::thnn_conv2d_backward, 3);
+      convNode->addInput(grads.at(0).value());
+      convNode->addInput(inputs.at(0));
+      convNode->addInput(inputs.at(1));
+      convNode->addInput(node->namedInput(attr::kernel_size));
+      convNode->addInput(node->namedInput(attr::stride));
+      convNode->addInput(node->namedInput(attr::padding));
+      convNode->addInput(outputs.at(1));
+      convNode->addInput(outputs.at(2));
+      convNode->addInput(graph->insertConstant(std::vector<bool>{true, true, true}));
+      graph->insertNode(convNode);
+      auto outputs = convNode->outputs();
+      JIT_ASSERT(outputs.size() == size_t(3));
+
+      return {outputs[0], outputs[1], nullptr, outputs[2], nullptr, nullptr};
+
     } else if (node->matches("aten::log_softmax(Tensor self, int dim) -> Tensor")) {
       JIT_ASSERT(grads.size() == 1);
       auto graph = node->owningGraph();