modify grad functions in cudnn to include num_groups

theano/gpuarray/dnn.py

@@ -582,8 +582,8 @@ class GpuDnnConv(DnnBase):
 
         top = gpu_contiguous(top)
 
-        d_img = GpuDnnConvGradI()(kerns, top, empty_like(img), desc)
-        d_kerns = GpuDnnConvGradW()(img, top, empty_like(kerns), desc)
+        d_img = GpuDnnConvGradI(num_groups=self.num_groups)(kerns, top, empty_like(img), desc)
+        d_kerns = GpuDnnConvGradW(num_groups=self.num_groups)(img, top, empty_like(kerns), desc)
         d_alpha = grad_not_implemented(self, 4, alpha)
         d_beta = grad_not_implemented(self, 5, beta)
 
@@ -680,8 +680,8 @@ class GpuDnnConvGradW(DnnBase):
 
         kerns = gpu_contiguous(kerns)
 
-        d_img = GpuDnnConvGradI()(kerns, top, empty_like(img), desc)
-        d_top = GpuDnnConv()(img, kerns, empty_like(top), desc)
+        d_img = GpuDnnConvGradI(num_groups=self.num_groups)(kerns, top, empty_like(img), desc)
+        d_top = GpuDnnConv(num_groups=self.num_groups)(img, kerns, empty_like(top), desc)
         d_alpha = grad_not_implemented(self, 4, alpha)
         d_beta = grad_not_implemented(self, 5, beta)
 
@@ -811,8 +811,8 @@ class GpuDnnConvGradI(DnnBase):
 
         img = gpu_contiguous(img)
 
-        d_kerns = GpuDnnConvGradW()(img, top, empty_like(kerns), desc)
-        d_top = GpuDnnConv()(img, kerns, empty_like(top), desc)
+        d_kerns = GpuDnnConvGradW(num_groups=self.num_groups)(img, top, empty_like(kerns), desc)
+        d_top = GpuDnnConv(num_groups=self.num_groups)(img, kerns, empty_like(top), desc)
         d_alpha = grad_not_implemented(self, 4, alpha)
         d_beta = grad_not_implemented(self, 5, beta)