Update.

theano/gpuarray/reduction.py

@@ -55,6 +55,12 @@ class GpuMaxAndArgmax(Op):
         max_typecode = pygpu.gpuarray.dtype_to_typecode(node.inputs[0].dtype)
         argmax_typecode = pygpu.gpuarray.dtype_to_typecode(self.argmax_dtype)
         ret = """
+        #if PY_MAJOR_VERSION >= 3
+            #ifndef PyInt_AS_LONG
+                #define PyInt_AS_LONG PyLong_AS_LONG
+            #endif
+        #endif
+
         unsigned  %(name)s_redux_len = PyTuple_GET_SIZE(%(axes)s);
         unsigned* %(name)s_axes_to_reduce = (unsigned*)malloc(%(name)s_redux_len * sizeof(unsigned));
         for (unsigned i = 0; i < %(name)s_redux_len; ++i) {

theano/gpuarray/tests/test_reduction.py

@@ -11,7 +11,11 @@ from .config import mode_with_gpu, mode_without_gpu
 from .test_basic_ops import rand_gpuarray
 from .. import GpuArrayType
 
-test_shape = (1000, 100, 10, 5, 2)
+import math
+
+# Number of values to be used in test tensors (except with 0-D tensors!).
+test_size = 10000000
+
 # NB: This order of "unsorted axes" is arbitrary and is here
 # just to have the same informations on profile output
 # from one test to another.
@@ -29,7 +33,7 @@ def numpy_random_array(shapes):
 
 def numpy_maxandargmax(X, axis=None):
     if axis is None:
-        axis = range(X.ndim)
+        axis = list(range(X.ndim))
     elif not isinstance(axis, (tuple, list)):
         axis = [int(axis)]
     axis = list(set(axis))  # remove duplicated values.
@@ -62,14 +66,22 @@ def check_if_gpu_maxandargmax_not_in_graph(theano_function):
 class BaseTest:
     # This attribute must be set in subclasses.
     tensor_size = None
+    shape = None
 
     dtype = theano.config.floatX
 
+    def get_shape(self):
+        if self.tensor_size == 0:
+            return []
+        return [int(math.ceil(math.pow(test_size, 1 / self.tensor_size)))] * self.tensor_size
+
     def setUp(self):
         if not isinstance(self.tensor_size, int):
             raise SkipTest("No tensor ndim defined.")
         if self.tensor_size < 0 or self.tensor_size > 5:
             raise SkipTest("We allow from 0 (included) to 5 (inclued) dimensons for these tests.")
+        if self.shape is None:
+            self.shape = self.get_shape()
 
     def get_host_tensor(self):
         broadcastable = (False,) * self.tensor_size
@@ -80,10 +92,10 @@ class BaseTest:
         return GpuArrayType(self.dtype, broadcastable)()
 
     def get_host_value(self):
-        return numpy_random_array(test_shape[:self.tensor_size])
+        return numpy_random_array(self.shape)
 
     def get_gpu_value(self):
-        return rand_gpuarray(*(test_shape[:self.tensor_size]))
+        return rand_gpuarray(*self.shape)
 
     # NB: In compute_host() and compute_gpu(),
     # the first call of the theano function should be ignored in profiling,
@@ -92,7 +104,7 @@ class BaseTest:
     def compute_host(self, test_tensor, axis):
         M = self.get_host_tensor()
         f = theano.function([M], [T.max(M, axis=axis), T.argmax(M, axis=axis)],
-                            name='HOST-function', mode=mode_without_gpu)
+                            name='HOST/shape:'+str(test_tensor.shape)+'/axis:'+str(axis), mode=mode_without_gpu)
         check_if_gpu_maxandargmax_not_in_graph(f)
         f(test_tensor)
         theano_max, theano_argmax = f(test_tensor)
@@ -103,7 +115,7 @@ class BaseTest:
     def compute_gpu(self, test_gpu_tensor, test_host_tensor, axis):
         M = self.get_gpu_tensor()
         f = theano.function([M], [T.max(M, axis=axis), T.argmax(M, axis=axis)],
-                            name='GPU-function', mode=mode_with_gpu)
+                            name='GPU/shape:'+str(test_gpu_tensor.shape)+'/axis:'+str(axis), mode=mode_with_gpu)
         check_if_gpu_maxandargmax_in_graph(f)
         f(test_gpu_tensor)
         theano_max, theano_argmax = f(test_gpu_tensor)
@@ -119,22 +131,17 @@ class BaseTest:
         self.compute_gpu(test_gpu_tensor, test_host_tensor, axis)
 
     def compute_axis(self, pos):
-        if 0 <= pos < self.tensor_size:
+        if self.tensor_size != 1 and 0 <= pos < self.tensor_size:
             self.compute(pos)
 
     def compute_some_axes(self, count):
-        if 0 <= count <= self.tensor_size:
+        if 0 <= count < self.tensor_size:
             self.compute([i for i in unsorted_axes if i < self.tensor_size][:count])
 
+    # Equivalent to test reduction on all axes.
     def test_none(self):
         self.compute(None)
 
-    def test_all_axes(self):
-        self.compute(range(self.tensor_size))
-
-    def test_all_axes_unsorted(self):
-        self.compute([i for i in unsorted_axes if i < self.tensor_size])
-
     def test_axis_1(self):
         self.compute_axis(0)
 
@@ -169,6 +176,16 @@ class TestScalar(BaseTest, TestCase):
 class TestVector(BaseTest, TestCase):
     tensor_size = 1
 
+# Special case
+class TestRow(BaseTest, TestCase):
+    tensor_size = 2
+    shape = [1,test_size]
+
+# Special case
+class TestColumn(BaseTest, TestCase):
+    tensor_size = 2
+    shape = [test_size, 1]
+
 
 class TestMatrix(BaseTest, TestCase):
     tensor_size = 2
@@ -176,3 +193,4 @@ class TestMatrix(BaseTest, TestCase):
 
 class TestTensor5(BaseTest, TestCase):
     tensor_size = 5
+

theano/tensor/basic.py

@@ -1237,6 +1237,12 @@ class MaxAndArgmax(Op):
         max, argmax = out
         fail = sub["fail"]
         ret = """
+        #if PY_MAJOR_VERSION >= 3
+            #ifndef PyInt_AS_LONG
+                #define PyInt_AS_LONG PyLong_AS_LONG
+            #endif
+        #endif
+
         int axis;
 
         if (PyTuple_GET_SIZE(%(axis)s) == PyArray_NDIM(%(x)s)) {
@@ -1597,7 +1603,7 @@ def max_and_argmax(a, axis=None, keepdims=False):
     # Check axis and convert it to a Python list of integers.
     # Axis will be used as an op param of MaxAndArgmax.
     if axis is None:
-        axis = range(a.type.ndim)
+        axis = list(range(a.type.ndim))
     elif (isinstance(axis, (integer_types, numpy.integer)) or
             (isinstance(axis, numpy.ndarray) and axis.ndim == 0)):
         axis = [int(axis)]
@@ -1605,7 +1611,7 @@ def max_and_argmax(a, axis=None, keepdims=False):
         axis = [int(i) for i in axis]
     elif isinstance(axis, Variable):
         if NoneConst.equals(axis):
-            axis = range(a.type.ndim)
+            axis = list(range(a.type.ndim))
         elif not isinstance(axis, TensorConstant):
             raise TypeError("max and argmax computation needs a constant axis. Got %s" % axis)
         else:
@@ -1616,7 +1622,7 @@ def max_and_argmax(a, axis=None, keepdims=False):
             elif isinstance(axis.data, (list, numpy.ndarray)):
                 axis = [int(i) for i in axis.data]
     if len(axis) == 0:
-        axis = range(a.type.ndim)
+        axis = list(range(a.type.ndim))
     else:
         for i in range(len(axis)):
             if axis[i] < 0: