Merge pull request #2265 from nouiz/warning

theano/compile/debugmode.py

@@ -2098,7 +2098,7 @@ class _Linker(gof.link.LocalLinker):
             return deco
 
         f = run_with_tensortype_filter_check(f)
-
+        f.storage_map = storage_map
         f.allow_gc = True
         assert len(fgraph.inputs) == len(input_storage)
         assert len(fgraph.outputs) == len(output_storage)

theano/sandbox/cuda/dnn.py

@@ -1136,8 +1136,8 @@ if cuda_available:
             subsample = node.op.subsample
             return [dnn_conv(gpu_contiguous(img), gpu_contiguous(kern),
                              border_mode=border_mode, subsample=subsample)]
-
-    @register_opt('cudnn')
+# DISABLED as there is problems in the handling of borders
+#    @register_opt('cudnn')
     @local_optimizer([GpuDownsampleFactorMax])
     def local_pool_dnn(node):
         if not dnn_available():
@@ -1149,7 +1149,8 @@ if cuda_available:
             ds = node.op.ds
             return [dnn_pool(gpu_contiguous(img), ds, ds)]
 
-    @register_opt('cudnn')
+# DISABLED as there is problems in the handling of borders
+#    @register_opt('cudnn')
     @local_optimizer([GpuDownsampleFactorMaxGrad])
     def local_pool_dnn_grad(node):
         if not dnn_available():

theano/sandbox/cuda/opt.py

@@ -1337,11 +1337,13 @@ conv_groupopt.register('conv_fft_full', local_conv_fft_full, 10,
 # It can be disabled by excluding 'conv_dnn' or 'cudnn'.
 from . import dnn
 if dnn.dnn_available():
-    conv_groupopt.register('conv_dnn', dnn.local_conv_dnn, 20,
+    conv_groupopt.register('local_conv_dnn', dnn.local_conv_dnn, 20,
+                           'conv_dnn',
                            'fast_compile', 'fast_run', 'cudnn')
 # The GEMM-based convolution comes last to catch all remaining cases.
 # It can be disabled by excluding 'conv_gemm'.
-conv_groupopt.register('conv_gemm', local_conv_gemm, 30,
+conv_groupopt.register('local_conv_gemm', local_conv_gemm, 30,
+                       'conv_gemm',
                        'fast_compile', 'fast_run')
 
 

theano/sandbox/cuda/tests/test_conv_cuda_ndarray.py

@@ -616,7 +616,13 @@ def test_default_conv():
         assert any([isinstance(a.op, cuda.blas.GpuCorrMM)
                     for a in f.maker.fgraph.apply_nodes])
 
-    mode = theano_mode.excluding('local_gpu_conv', 'local_conv_gemm')
+    mode = theano_mode.excluding('local_conv_dnn', 'local_conv_gemm')
+    f = theano.function([img, fil], c, mode=mode)
+
+    assert any([isinstance(a.op, cuda.blas.GpuConv)
+                for a in f.maker.fgraph.apply_nodes])
+
+    mode = theano_mode.excluding('conv_dnn', 'conv_gemm')
     f = theano.function([img, fil], c, mode=mode)
 
     assert any([isinstance(a.op, cuda.blas.GpuConv)

theano/sandbox/cuda/tests/test_dnn.py

@@ -121,7 +121,7 @@ def test_dnn_tag():
             [x],
             max_pool_2d(x, ds=(2, 2)),
             mode=mode_with_gpu.including("cudnn"))
-    except RuntimeError, e:
+    except (AssertionError, RuntimeError), e:
         assert not cuda.dnn.dnn_available()
         raised = True
     finally:

theano/sandbox/cuda/tests/test_nnet.py

@@ -301,6 +301,12 @@ class test_SoftMax(unittest.TestCase):
         self._cmp(2 << 15, 5, f, f_gpu)
         self._cmp(0, 10, f, f_gpu)
 
+    def test_softmax_cudnn(self):
+        if not cuda.dnn.dnn_available():
+            raise SkipTest(cuda.dnn.dnn_available.msg)
+        x = T.fmatrix('x')
+        z = T.nnet.softmax
+
         def check_types_with_cudnn(graph, graph_gpu):
             self._check_types(
                 graph,
@@ -320,7 +326,7 @@ class test_SoftMax(unittest.TestCase):
             check_types_with_cudnn
         )
 
-    def test_cudnn_softmax(self):
+    def test_cudnn_softmax_grad(self):
         if not cuda.dnn.dnn_available():
             raise SkipTest(cuda.dnn.dnn_available.msg)
 

theano/tensor/subtensor.py

@@ -1893,7 +1893,9 @@ class AdvancedSubtensor(Op):
         # Really special case
         if len(ishapes) == 3:
             xshp, ind1shp, ind2shp = ishapes
-            if len(xshp) == 2 and len(ind1shp) == 1 and len(ind2shp) == 1:
+            if (len(xshp) == 2 and
+                    ind1shp is not None and len(ind1shp) == 1 and
+                    ind2shp is not None and len(ind2shp) == 1):
                 # if the graph is correct, we can assume ind1shp[0] and
                 # ind2shp[0] will have the same value.
                 # Try to return the one closest to the graph input.

theano/tensor/tests/test_basic.py

@@ -3481,7 +3481,7 @@ class T_Join_and_Split(unittest.TestCase):
 
     def test_join_matrixV(self):
         """variable join axis"""
-        v = numpy.array([[1., 2., 3.], [4., 5., 6.]], dtype=self.floatX)
+        v = numpy.array([[.1, .2, .3], [.4, .5, .6]], dtype=self.floatX)
         a = self.shared(v.copy())
         b = as_tensor_variable(v.copy())
         ax = lscalar()
@@ -3491,13 +3491,15 @@ class T_Join_and_Split(unittest.TestCase):
         topo = f.maker.fgraph.toposort()
         assert [True for node in topo if isinstance(node.op, self.join_op)]
 
-        want = numpy.array([[1, 2, 3], [4, 5, 6], [1, 2, 3], [4, 5, 6]])
+        want = numpy.array([[.1, .2, .3], [.4, .5, .6],
+                            [.1, .2, .3], [.4, .5, .6]])
         got = f(0)
-        self.assertTrue((got == want).all(), (got, want))
+        assert numpy.allclose(got, want)
 
-        want = numpy.array([[1, 2, 3, 1, 2, 3], [4, 5, 6, 4, 5, 6]])
+        want = numpy.array([[.1, .2, .3, .1, .2, .3],
+                            [.4, .5, .6, .4, .5, .6]])
         got = f(1)
-        self.assertTrue((got == want).all(), (got, want))
+        assert numpy.allclose(got, want)
 
         utt.verify_grad(lambda a, b: join(0, a, b), [v, 2 * v], mode=self.mode)
         utt.verify_grad(lambda a, b: join(1, a, b), [v, 2 * v], mode=self.mode)

theano/tensor/tests/test_gc.py

@@ -94,7 +94,7 @@ def test_gc_never_pickles_temporaries():
         # that weren't collected shouldn't be pickled anyway
         # Allow for a couple of bytes of difference, since timing info,
         # for instance, can be represented as text of varying size.
-        assert abs(len_post_f - len_post_g) < 4, (
+        assert abs(len_post_f - len_post_g) < 16, (
             f_linker, len_post_f, len_post_g)
 
 

theano/tensor/tests/test_subtensor.py

@@ -1455,3 +1455,24 @@ class TestInferShape(utt.InferShapeTester):
         self._compile_and_check([admat, advec],
                     [set_subtensor(admat[aivec_val, bivec_val], advec)],
                     [admat_val, advec_val], AdvancedIncSubtensor)
+
+    def test_adv_sub(self):
+        admat = dmatrix()
+        aivec = lvector()
+        bivec = lvector()
+
+        admat_val = rand(5, 4)
+        aivec_val = [1, 3, 2]
+        bivec_val = [0, 3, 3]
+        self._compile_and_check([admat, aivec, bivec],
+                                [admat[aivec, bivec]],
+                                [admat_val, aivec_val, bivec_val], AdvancedSubtensor)
+        # Test case that aren't implemented, but make sure they do not crash.
+        self._compile_and_check([admat, aivec],
+                                [admat[aivec, 1:3]],
+                                [admat_val, aivec_val], AdvancedSubtensor,
+                                check_topo=False)
+        self._compile_and_check([admat, aivec],
+                                [admat[1:3, aivec]],
+                                [admat_val, aivec_val], AdvancedSubtensor,
+                                check_topo=False)

theano/tests/unittest_tools.py

@@ -191,7 +191,7 @@ class InferShapeTester(unittest.TestCase):
         self.mode = mode.including("canonicalize")
 
     def _compile_and_check(self, inputs, outputs, numeric_inputs, cls,
-                           excluding=None, warn=True):
+                           excluding=None, warn=True, check_topo=True):
         """This tests the infer_shape method only
 
         When testing with input values with shapes that take the same
@@ -204,6 +204,9 @@ class InferShapeTester(unittest.TestCase):
         matrices will not detect the problem. If warn=True, we emit a
         warning when testing with such values.
 
+        :param check_topo: If True, we check that the Op where removed
+            from the graph. False is useful to test not implemented case.
+
         """
         mode = self.mode
         if excluding:
@@ -236,8 +239,9 @@ class InferShapeTester(unittest.TestCase):
                                           mode=mode)
         #theano.printing.debugprint(shapes_function)
         # Check that the Op is removed from the compiled function.
-        topo_shape = shapes_function.maker.fgraph.toposort()
-        assert not any(isinstance(t.op, cls) for t in topo_shape)
+        if check_topo:
+            topo_shape = shapes_function.maker.fgraph.toposort()
+            assert not any(isinstance(t.op, cls) for t in topo_shape)
         topo_out = outputs_function.maker.fgraph.toposort()
         assert any(isinstance(t.op, cls) for t in topo_out)
         # Check that the shape produced agrees with the actual shape.