moved inherit_stack_trace inside output_merge decorator

theano/gpuarray/dnn.py

@@ -3424,7 +3424,6 @@ def local_dnn_convi_alpha_merge(node, *inputs):
 @register_opt('cudnn')
 @output_merge(GpuDnnConv, alpha_in=4, beta_in=5, out_in=2)
 def local_dnn_conv_output_merge(node, *inputs):
-    with inherit_stack_trace(node.outputs):
     inputs = inputs[0:2] + (gpu_contiguous(inputs[2]),) + inputs[3:]
     return [GpuDnnConv(algo=node.op.algo)(*inputs)]
 
@@ -3432,7 +3431,6 @@ def local_dnn_conv_output_merge(node, *inputs):
 @register_opt('cudnn')
 @output_merge(GpuDnnConvGradW, alpha_in=4, beta_in=5, out_in=2)
 def local_dnn_convw_output_merge(node, *inputs):
-    with inherit_stack_trace(node.outputs):
     inputs = inputs[0:2] + (gpu_contiguous(inputs[2]),) + inputs[3:]
     return [GpuDnnConvGradW(algo=node.op.algo)(*inputs)]
 
@@ -3440,7 +3438,6 @@ def local_dnn_convw_output_merge(node, *inputs):
 @register_opt('cudnn')
 @output_merge(GpuDnnConvGradI, alpha_in=4, beta_in=5, out_in=2)
 def local_dnn_convi_output_merge(node, *inputs):
-    with inherit_stack_trace(node.outputs):
     inputs = inputs[0:2] + (gpu_contiguous(inputs[2]),) + inputs[3:]
     return [GpuDnnConvGradI(algo=node.op.algo)(*inputs)]
 

theano/gpuarray/opt_util.py

@@ -274,6 +274,7 @@ def output_merge(cls, alpha_in, beta_in, out_in):
                 inputs = list(targ.inputs)
                 inputs[out_in] = W
                 inputs[beta_in] = _one.clone()
+                with inherit_stack_trace(node.outputs):
                     return maker(targ, *inputs)
         return opt
     return wrapper