Merge pull request #3875 from abergeron/fix_buildbot

theano/compile/debugmode.py

@@ -1784,8 +1784,7 @@ class _VariableEquivalenceTracker(object):
 # List of default version of make thunk.
 # This is needed to know if the user overrided it.
 # The GpuOp will be added here when theano.sandbox.cuda is imported.
-default_make_thunk = [get_unbound_function(theano.gof.Op.make_thunk),
-                      get_unbound_function(theano.gof.OpenMPOp.make_thunk)]
+default_make_thunk = [get_unbound_function(theano.gof.Op.make_thunk)]
 
 
 # Debug mode cheats and initializes the linker in a different way in
@@ -1879,6 +1878,10 @@ class _Linker(gof.link.LocalLinker):
                 thunk.inputs = [storage_map[v] for v in node.inputs]
                 thunk.outputs = [storage_map[v] for v in node.outputs]
                 thunk_other = thunk
+            else:
+                new_node = node.op.prepare_node(node)
+                if new_node is not None:
+                    node = new_node
 
             try:
                 if not self.maker.mode.check_c_code:

theano/gof/op.py

@@ -817,6 +817,16 @@ class Op(utils.object2, PureOp, CLinkerOp):
         else:
             return NotImplemented
 
+    def prepare_node(self, node):
+        """
+        Make any special modifications that the Op needs before doing
+        make_thunk().
+
+        This can either modify the node inplace or return a new one.
+
+        """
+        pass
+
     def make_c_thunk(self, node, storage_map, compute_map, no_recycling):
         """
         Like make_thunk, but will only try to make a C thunk.
@@ -926,6 +936,10 @@ class Op(utils.object2, PureOp, CLinkerOp):
         """
         logger = logging.getLogger('theano.gof.op.Op')
 
+        new_node = self.prepare_node(node)
+        if new_node is not None:
+            node = new_node
+
         if self._op_use_c_code:
             try:
                 return self.make_c_thunk(node, storage_map, compute_map,
@@ -1162,10 +1176,8 @@ int main( int argc, const char* argv[] )
                 self.openmp = False
                 theano.config.openmp = False
 
-    def make_thunk(self, node, storage_map, compute_map, no_recycling):
+    def prepare_node(self, node):
         self.update_self_openmp()
-        return super(OpenMPOp, self).make_thunk(node, storage_map,
-                                                compute_map, no_recycling)
 
 
 def simple_meth(tag):

theano/sandbox/multinomial.py

@@ -55,7 +55,7 @@ class MultinomialFromUniform(Op):
         return [T.zeros_like(x) for x in ins]
 
     def c_code_cache_version(self):
-        return (7,)
+        return (8,)
 
     def c_code(self, node, name, ins, outs, sub):
         # support old pickled graphs
@@ -122,7 +122,7 @@ class MultinomialFromUniform(Op):
             for (int n = 0; n < nb_multi; ++n)
             {
                 int waiting = 1;
-                dtype_%(pvals)s cummul = 0.;
+                double cummul = 0.;
                 const dtype_%(unis)s* unis_n = (dtype_%(unis)s*)PyArray_GETPTR1(%(unis)s, c*nb_multi + n);
                 for (int m = 0; m < nb_outcomes; ++m)
                 {

theano/tensor/elemwise.py

@@ -7,7 +7,6 @@ import numpy
 import theano
 from theano import gof
 from theano.compat import izip
-from theano.compat import get_unbound_function
 from six import iteritems
 from six.moves import xrange
 from theano.gof import Apply, Op, OpenMPOp
@@ -793,8 +792,7 @@ class Elemwise(OpenMPOp):
 
         return ret
 
-    def make_thunk(self, node, storage_map, compute_map, no_recycling):
-        node_ = node
+    def prepare_node(self, node):
         # Postpone the ufunc building to the last minutes
         # NumPy ufunc support only up to 31 inputs.
         # But our c code support more.
@@ -830,8 +828,6 @@ class Elemwise(OpenMPOp):
             char = numpy.sctype2char(out_dtype)
             sig = char * node.nin + '->' + char * node.nout
             node.tag.sig = sig
-        return super(Elemwise, node_.op).make_thunk(node_, storage_map,
-                                                    compute_map, no_recycling)
 
     def perform(self, node, inputs, output_storage):
         if len(node.inputs) >= 32:
@@ -1273,19 +1269,6 @@ class Elemwise(OpenMPOp):
         """
         return node.outputs[0].ndim == 0
 
-theano.compile.debugmode.default_make_thunk.append(
-    get_unbound_function(Elemwise.make_thunk))
-
-# def elemwise_to_scal(fgraph):
-# TODO: why is this commented out? should it be removed?
-#       it has needed maintenance despite being commented
-#     mapping = {}
-#     inputs = []
-#     outputs = []
-#     for node in fgraph.io_toposort():
-#         if not isinstance(node.op, Elemwise):
-#             raise TypeError('All ops in the graph must be Elemwise.')
-
 
 ################
 #   CAReduce   #

theano/tensor/nnet/tests/test_conv.py

@@ -99,6 +99,8 @@ class TestConv2D(utt.InferShapeTester):
         out_shape2d = numpy.array(N_image_shape[-2:]) +\
                       s * numpy.array(N_filter_shape[-2:]) - s
         out_shape2d = numpy.ceil(out_shape2d / numpy.array(subsample))
+        # avoid numpy deprecation
+        out_shape2d = out_shape2d.astype('int32')
         out_shape = (N_image_shape[0], N_filter_shape[0]) + tuple(out_shape2d)
         ref_output = numpy.zeros(out_shape)
 

theano/tensor/nnet/tests/test_corr.py

@@ -97,6 +97,8 @@ class TestCorr2D(utt.InferShapeTester):
         else:
             raise NotImplementedError('Unsupported border_mode {}'.format(border_mode))
         out_shape2d = numpy.floor((img_shape2d + 2 * (padHW) - fil_shape2d) / subsample2d) + 1
+        # avoid numpy deprecation
+        out_shape2d = out_shape2d.astype('int32')
         out_shape = (N_image_shape[0], N_filter_shape[0]) + tuple(out_shape2d)
         ref_output = numpy.zeros(out_shape)
 

theano/tensor/tests/test_basic.py

@@ -1493,6 +1493,12 @@ CosInplaceTester = makeBroadcastTester(
     grad=_grad_broadcast_unary_wide,
     inplace=True)
 
+def test_py_c_match():
+    a = tensor.TensorType(dtype='int8', broadcastable=(False,))()
+    f = theano.function([a], tensor.arccos(a), mode='DebugMode')
+    # This can fail in DebugMode
+    f(numpy.asarray([1, 0, -1], dtype='int8'))
+
 ArccosTester = makeBroadcastTester(op=tensor.arccos,
                                    expected=upcast_float16_ufunc(numpy.arccos),
                                    good=_good_broadcast_unary_arcsin,