minor changes

theano/tensor/nnet/conv2d.py

@@ -110,7 +110,7 @@ def conv2d(img,
         filters = filters[:, :, ::-1, ::-1]
     ### FIXME input shape/kernel shape
     conv_op = Conv2d(imshp=input_shape, kshp=filter_shape, bsize=batch_size,
-                     border_mode="valid", subsample=subsample)
+                     border_mode=border_mode, subsample=subsample)
     return conv_op(img, filters)
 
 
@@ -354,15 +354,15 @@ def replace_conv_with_cudnn(convop, inputs):
                         direction_hint='forward')
         return rval
     if (isinstance(convop, Conv2d_gradWeights)):
-        rval = dnn_conv(inp1, inp2,
-                        border_mode=node.op.border_mode,
-                        subsample=node.op.subsample,
+        rval = dnn_conv(inp1.dimshuffle(1, 0, 2, 3), inp2,
+                        border_mode=convop.border_mode,
+                        subsample=convop.subsample,
                         direction_hint='bprop weights')
         return rval
     if (isinstance(convop, Conv2d_gradInputs)):
         rval = dnn_conv(inp1, inp2,
-                        border_mode=node.op.border_mode,
-                        subsample=node.op.subsample,
+                        border_mode=convop.border_mode,
+                        subsample=convop.subsample,
                         direction_hint='bprop inputs')
         return rval
 
@@ -428,7 +428,7 @@ def replace_convgradinputs_withcorrmm(convop, inputs):
     kern, topgrad, shape = inputs
     rval =  GpuCorrMM_gradInputs(border_mode=convop.border_mode,
     subsample=convop.subsample)(
-    gpu_contiguous(kern), gpu_contiguous(topgrad))
+        gpu_contiguous(kern), gpu_contiguous(topgrad), shape)
     return rval
 
 

theano/tensor/nnet/tests/test_conv2d.py

@@ -32,7 +32,8 @@ class TestConv2d(unittest.TestCase):
         inputs = shared(inputs_val)
         filters = shared(filters_val)
         c_ref = conv_ref.conv2d(inputs, filters,
-                                border_mode="valid", subsample=subsample)
+                                border_mode="valid",
+                                subsample=subsample)
 
         c = conv.conv2d(inputs, filters,
                         border_mode="valid", subsample=subsample)
@@ -43,6 +44,7 @@ class TestConv2d(unittest.TestCase):
 
         res_ref = f_ref()
         res = f()
+        print res_ref.shape, res.shape
         utt.assert_allclose(res_ref, res)
         if verify_grad:
             utt.verify_grad(conv.Conv2d(border_mode="valid",
@@ -52,29 +54,35 @@ class TestConv2d(unittest.TestCase):
 
     def test_valid(self):
         mode = mode_with_gpu
-        if dnn_available():
-            self.run_conv(inputs_shape=(16, 1, 2, 2),
-                          filters_shape=(10, 1, 2, 2),
-                          verify_grad=False)
-            # self.run_conv(inputs_shape=(16, 1, 8, 8),
+       # if dnn_available():
+            # self.run_conv(inputs_shape=(16, 1, 2, 2),
             #               filters_shape=(10, 1, 2, 2),
-            #               subsample=(2, 2),
             #               verify_grad=False)
+            # # self.run_conv(inputs_shape=(16, 1, 8, 8),
+            # #               filters_shape=(10, 1, 2, 2),
+            # #               subsample=(2, 2),
+            # #               verify_grad=False)
             # self.run_conv(inputs_shape=(16, 1, 2, 2),
             #               filters_shape=(10, 1, 2, 2),
             #               verify_grad=True)
-            # self.run_conv(inputs_shape=(16, 1, 8, 8),
-            #               filters_shape=(10, 1, 2, 2),
-            #               subsample=(2, 2),
-            #               verify_grad=True)
+            # # self.run_conv(inputs_shape=(16, 1, 8, 8),
+            # #               filters_shape=(10, 1, 2, 2),
+            # #               subsample=(2, 2),
+            # #               verify_grad=True)
 
         mode = mode.excluding('cudnn')
         self.run_conv(inputs_shape=(16, 1, 2, 2),
                       filters_shape=(10, 1, 2, 2),
                       verify_grad=False, mode=mode)
+        self.run_conv(inputs_shape=(16, 1, 2, 2),
+                      filters_shape=(10, 1, 2, 2),
+                      subsample=(1, 1),
+                      verify_grad=True,mode=mode)
+
+
         # self.run_conv(inputs_shape=(16, 1, 8, 8),
-        #               filters_shape=(10, 1, 2, 2),
-        #               subsample=(2, 2),
+        #               filters_shape=(10, 1, 4, 4),
+        #                subsample=(2, 2),
         #               verify_grad=False,mode=mode)
         # self.run_conv(inputs_shape=(16, 1, 2, 2),
         #               filters_shape=(10, 1, 2, 2),