Merge pull request #364 from SYSTRAN/master

test on inputGpu emptiness symmetric in backward and forward
torch · Nov 2, 2016 · aa256bc · aa256bc
2 parents 64224a6 + 98b777b
commit aa256bc
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/DataParallelTable.lua b/DataParallelTable.lua
@@ -231,8 +231,8 @@ function DataParallelTable:__backward(method, input, gradOutput, scale)
       self:_distribute(self.gradOutputGpu, gradOutput)
 
       self.gradInputGpu = self.impl:exec(function(m, i)
-         if torch.isTensor(inputGpu[i]) and inputGpu[i]:numel() == 0 then
-            return torch.CudaTensor()
+         if not _hasData(inputGpu[i]) then
+            return inputGpu[i]
          else
             return m[method](m, inputGpu[i], gradOutputGpu[i], scale)
          end
@@ -246,8 +246,8 @@ function DataParallelTable:__backward(method, input, gradOutput, scale)
 
    if method == 'accGradParameters' then
       self.impl:exec(function(m, i)
-         if torch.isTensor(inputGpu[i]) and inputGpu[i]:numel() == 0 then
-            return torch.CudaTensor()
+         if not _hasData(inputGpu[i]) then
+            return inputGpu[i]
          else
             return m:accGradParameters(inputGpu[i], gradOutputGpu[i], scale)
          end