test_elemwise0 passed

__init__.py

+from .type import CudaNdarrayType
+
+from .var import (CudaNdarrayVariable,
+    CudaNdarrayConstant,
+    CudaNdarraySharedVariable,
+    shared_constructor)

tests/test_basic_ops.py

@@ -4,35 +4,53 @@ from theano import tensor
 
 import numpy
 
-import gputensor as gpt
+import theano_cuda_ndarray as tcn
 
 
-def test0():
+def test_elemwise0():
 
-    a = gpt.gpu_tensor_shared_constructor(numpy.random.rand(3,4), 'a')
+    a = tcn.shared_constructor(numpy.random.rand(4,4), 'a')
 
     b = tensor.dmatrix()
 
     f = pfunc([b], [], updates=[(a, a+b)])
 
     a0 = a.value * 1.0
-    f(numpy.ones((3,4)))
+    print 'BEFORE ADD', a.value
+    f(numpy.ones((4,4)))
     print f.maker.env.toposort()
+    print 'AFTER ADD', a.value
 
     assert numpy.all(a0 + 1.0 == a.value)
 
-def test1():
-
-    a = gpt.gpu_tensor_shared_constructor(numpy.random.rand(3,4), 'a')
+def test_elemwise1():
+    """ Several kinds of elemwise expressions with no broadcasting, non power-of-two shape """
 
+    shape = (3,4)
+    a = tcn.shared_constructor(numpy.random.rand(*shape), 'a')
     b = tensor.dmatrix()
-
-    f = pfunc([b], [], updates=[(a, a+b)])
-    for i, node in enumerate( f.maker.env.toposort()):
-        print 'test1 toposort', i, node
-
-    a0 = a.value * 1.0
-    f(numpy.ones((3,4)))
-
-    assert numpy.all(a0 + 1.0 == a.value)
-
+    f = pfunc([b], [], updates=[(a, a+b * tensor.exp(b**a))])
+    #let debugmode catch any mistakes
+    f(numpy.ones(shape))
+
+def test_elemwise2():
+    """ Several kinds of elemwise expressions with dimension permutations """
+    
+    shape = (3,4,5,6)
+    a = tcn.shared_constructor(numpy.random.rand(*shape), 'a')
+    b = tensor.Tensor(dtype='float32', broadcastable=[0]*len(shape))()
+    f = pfunc([b], [], updates=[(a, (a+b).dimshuffle([2,0,3,1]) *
+        tensor.exp(b**a).dimshuffle([2,0,3,1]))])
+    #let debugmode catch errors
+    f(numpy.ones(shape))
+
+def test_elemwise3():
+    """ Several kinds of elemwise expressions with dimension permutations and broadcasting"""
+    
+    shape = (3,4,5,6)
+    a = tcn.shared_constructor(numpy.random.rand(*shape), 'a')
+    b = tensor.dvector()
+    f = pfunc([b], [], updates=[(a, (a+b).dimshuffle([2,0,3,1]) * tensor.exp(1 +
+        b**a).dimshuffle([2,0,3,1]))])
+    #let debugmode catch errors
+    f(numpy.ones(6))

type_support.cu

@@ -7,15 +7,14 @@
 #define DECL(s) static PyObject * s(PyObject * self, PyObject *args)
 
 static PyObject * 
-filter(PyObject* self, PyObject *args) // args = (data, typenum, broadcastable, strict)
+filter(PyObject* self, PyObject *args) // args = (data, broadcastable, strict)
 {
     PyObject *py_data=NULL;
     PyArrayObject * data = NULL;
-    int dtype_typenum=-1;
     int strict = 0;
     PyObject * broadcastable=NULL;
 
-    if (!PyArg_ParseTuple(args, "OiOi", &py_data, &dtype_typenum, &broadcastable, &strict)) return NULL;
+    if (!PyArg_ParseTuple(args, "OOi", &py_data, &broadcastable, &strict)) return NULL;
 
     if (!PyTuple_Check(broadcastable)){
         PyErr_SetString(PyExc_TypeError, "broadcastable arg should be a tuple of int.");
@@ -99,9 +98,9 @@ static PyMethodDef MyMethods[] = {
 
 
 PyMODINIT_FUNC
-init_theano_cuda_ndarray(void)
+inittype_support(void)
 {
-        (void) Py_InitModule("_theano_cuda_ndarray", MyMethods);
+        (void) Py_InitModule("type_support", MyMethods);
         import_array();
 }