Merge pull request #178 from nouiz/important_fix2

theano/sandbox/cuda/elemwise.py

@@ -815,10 +815,13 @@ nd_collapse_[i]=0;
         """
         kernels = "".join(
             [self.c_src_kernel(node, nodename, x) for x in xrange(1, nd + 1)]
-            + [self.c_src_kernel_Ccontiguous(node, nodename)],
+            + [self.c_src_kernel_Ccontiguous(node, nodename)]
             + [self.c_src_callkernel(node, nodename)])
         return defines + kernels
 
+    def c_support_code(self):
+        raise gof.utils.MethodNotDefined()
+
     def c_code(self, node, nodename, inputs, outputs, sub):
         d = dict(sub)
         nd = node.outputs[0].type.ndim

theano/sandbox/cuda/opt.py

@@ -77,7 +77,7 @@ class InputToGpuOptimizer(Optimizer):
 
                     if new_input.type==input.type:
                         env.replace_validate(input, new_input, "InputToGpuOptimizer")
-                except Exception, e:
+                except TypeError, e:
                     #as we currently only support float32, this can fail.
                     #Using try except make that we won't need
                     pass
@@ -243,7 +243,12 @@ def local_gpu_dot_to_dot22(node):
 
     A more suitable solution would be to use the right cublas call
     """
+
+    # In case the got do input upcast, we much check that we can
+    # make it run on the gpu.
     if node.op == gpu_from_host:
+        if node.outputs[0].type.dtype != 'float32':
+            return False
         host_input = node.inputs[0]
         if host_input.owner and host_input.owner.op == tensor.basic.dot:
             x, y = host_input.owner.inputs
@@ -264,6 +269,8 @@ def local_gpu_dot_to_dot22(node):
 
             return [GpuReshape(1)(gpu_dot22(gpu_x, gpu_y), shape_out)]
     if node.op == tensor.basic.dot:
+        if node.outputs[0].type.dtype != 'float32':
+            return False
         if numpy.any([(i.owner and i.owner.op == host_from_gpu) for i in node.inputs]):
             x, y = node.inputs
             if _is_real_vector(x) and _is_real_matrix(y):
@@ -1267,7 +1274,7 @@ def gpu_scan_make_inplace(node):
         n_outs = len(ls)
         for idx in xrange(n_outs):
             if ls[idx] in ls[:idx]:
-                ls[idx] = deep_copy_op(ls[idx])
+                ls[idx] = compile.function_module.deep_copy_op(ls[idx])
 
         inputs = ls_begin + ls + ls_end
 

theano/tensor/opt.py

@@ -922,9 +922,17 @@ class ShapeFeature(object):
             self.set_shape(r, s)
 
     def on_change_input(self, env, node, i, r, new_r):
+        if new_r not in self.shape_of:
+            # It happen that the env didn't called on_import for some
+            # new_r.  This happen when new_r don't have an
+            # owner(i.e. it is a constant or an input of the graph)
+            # update_shape suppose that r and new_r are in shape_of.
+            self.init_r(new_r)
+
         # This tells us that r and new_r must have the same shape
         # if we didn't know that the shapes are related, now we do.
-        self.update_shape(new_r, r)
+        # We should give priority to new_r, so we put it last.
+        self.update_shape(r, new_r)
 
         # change_input happens in two cases:
         # 1) we are trying to get rid of r, or