提交 3c090b70 authored 作者: abalkin's avatar abalkin

Merge pull request #3 from lamblin/take-op-c-code-clean

Only perform expensive check if the cast is unsafe
......@@ -6552,9 +6552,11 @@ class AdvancedSubtensor1(Op):
# many elements on a 32-bit arch).
if i.dtype != numpy.intp:
i_ = theano._asarray(i, dtype=numpy.intp)
if numpy.any(i != i_):
raise IndexError('index contains values that are bigger than '
'the maximum array size on this system.', i)
if not numpy.can_cast(i.dtype, numpy.intp):
# Check if there was actually an incorrect conversion
if numpy.any(i != i_):
raise IndexError('index contains values that are bigger '
'than the maximum array size on this system.', i)
i = i_
out[0] = x.take(i, axis=0, out=o)
......@@ -6603,21 +6605,24 @@ class AdvancedSubtensor1(Op):
if (PyArray_TYPE(%(i_name)s) != NPY_INTP) {
// Cast %(i_name)s to NPY_INTP (expected by PyArray_TakeFrom),
// if all values fit.
PyObject* py_min_val, py_max_val;
npy_int64 min_val, max_val;
py_min_val = PyArray_Min(%(i_name)s, NPY_MAXDIMS, min_val);
py_max_val = PyArray_Max(%(i_name)s, NPY_MAXDIMS, max_val);
min_val = PyLong_AsLongLong(py_min_val);
max_val = PyLong_AsLongLong(py_max_val);
Py_CLEAR(py_min_val);
Py_CLEAR(py_max_val);
if ((min_val < NPY_MIN_INTP) || (max_val > NPY_MAX_INTP))
if (!PyArray_CanCastSafely(PyArray_TYPE(%(i_name)s), NPY_INTP))
{
PyExc_SetErr(PyExc_IndexError, "Index contains values "
"that are bigger than the maximum array "
"size on this system.");
%(fail)s;
PyObject* py_min_val, py_max_val;
npy_int64 min_val, max_val;
py_min_val = PyArray_Min(%(i_name)s, NPY_MAXDIMS, min_val);
py_max_val = PyArray_Max(%(i_name)s, NPY_MAXDIMS, max_val);
min_val = PyLong_AsLongLong(py_min_val);
max_val = PyLong_AsLongLong(py_max_val);
Py_CLEAR(py_min_val);
Py_CLEAR(py_max_val);
if ((min_val < NPY_MIN_INTP) || (max_val > NPY_MAX_INTP))
{
PyExc_SetErr(PyExc_IndexError, "Index contains values "
"that are bigger than the maximum array "
"size on this system.");
%(fail)s;
}
}
indices = PyArray_Cast(%(i_name)s, NPY_INTP);
if (indices == NULL) {
......@@ -6659,7 +6664,7 @@ class AdvancedSubtensor1(Op):
""" % locals()
def c_code_cache_version(self):
return (0, 0, 4)
return (0, 0, 5)
advanced_subtensor1 = AdvancedSubtensor1()
......
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论