Fix subsampling in GpuCorr3dMM_gradWeigth

theano/sandbox/cuda/blas.py

@@ -9,6 +9,7 @@ from theano.sandbox.cuda.type import CudaNdarrayType
 from theano.sandbox.cuda import GpuOp
 from theano.sandbox.cuda.basic_ops import (as_cuda_ndarray_variable,
                                            gpu_contiguous)
+from theano.tensor import as_tensor_variable
 
 
 class GpuDot22(GpuOp):
@@ -1307,8 +1308,8 @@ class GpuCorr3dMM(BaseGpuCorr3dMM):
         batchsize or number of filters) may also work around the CUBLAS bug.
     """
     def __init__(self, border_mode="valid",
-            subsample=(1, 1, 1),
-            pad=(0, 0, 0)):
+                 subsample=(1, 1, 1),
+                 pad=(0, 0, 0)):
         """
         :param border_mode: currently supports "valid" only; "full" can be
             simulated by setting `pad="full"` (at the cost of performance), or
@@ -1375,6 +1376,9 @@ class GpuCorr3dMM_gradWeights(BaseGpuCorr3dMM):
     def make_node(self, img, topgrad, shape=None):
         img = as_cuda_ndarray_variable(img)
         topgrad = as_cuda_ndarray_variable(topgrad)
+        if shape is not None:
+            shape = as_tensor_variable(shape)
+
         if img.type.ndim != 5:
             raise TypeError('img must be 5D tensor')
         if topgrad.type.ndim != 5:

theano/sandbox/cuda/opt.py

@@ -1380,7 +1380,7 @@ def local_convgrad3d_gemm(node):
         f = gpu_contiguous(f.dimshuffle(0, 4, 1, 2, 3))
 
         rval = GpuCorr3dMM_gradWeights(subsample=(sx, sy, sz))(x, f,
-                                                               shape=node.inputs[2])
+                                                               shape=node.inputs[2][1:4])
         # Shuffle from (ic, oc, 0, 1, t) to (oc, 0, 1, t, ic)
         return [rval.dimshuffle(0, 2, 3, 4, 1)]
 

theano/sandbox/cuda/tests/test_gemmcorr3d.py

@@ -69,10 +69,8 @@ class TestCorr3DMM(unittest.TestCase):
                         subsample=(1, 1, 1)):
         inputs_val = numpy.random.random(inputs_shape).astype('float32')
         dCdH_val = numpy.random.random(dCdH_shape).astype('float32')
-        filters_val = numpy.random.random(filters_shape).astype('float32')
         inputs = shared(inputs_val)
         dCdH = shared(dCdH_val)
-        filters = shared(filters_val)
 
         conv = theano.tensor.nnet.convGrad3D(V=inputs, dCdH=dCdH,
                                               WShape=filters_shape,
@@ -85,7 +83,7 @@ class TestCorr3DMM(unittest.TestCase):
         else:
             conv_gemm = GpuCorr3dMM_gradWeights(subsample=subsample)(img,
                                                                      topgrad,
-                                                                     shape=filters.shape[1:4])
+                                                                     shape=filters_shape[1:4])
         conv_gemm = conv_gemm.dimshuffle(0, 2, 3, 4, 1)
         f_ref = theano.function([], conv)
         f = theano.function([], conv_gemm)