diff --git a/doc/source/array.rst b/doc/source/array.rst
index d9d7ccf75eac37a7904dfa8ce982320fc565c424..e3802292ce38630db05e606c3d0df02f01379fdb 100644
--- a/doc/source/array.rst
+++ b/doc/source/array.rst
@@ -125,6 +125,7 @@ Constructing :class:`Array` Instances
.. autofunction:: zeros_like
.. autofunction:: arange
.. autofunction:: take
+.. autofunction:: concatenate
Conditionals
^^^^^^^^^^^^
diff --git a/doc/source/conf.py b/doc/source/conf.py
index 4f46d9a166ce69086f4cf199e1eeb6117f6432cf..8b10c70cbb70453e61b8a1fb6c8a2c7df948f82a 100644
--- a/doc/source/conf.py
+++ b/doc/source/conf.py
@@ -167,8 +167,9 @@ htmlhelp_basename = 'PyCudadoc'
# Grouping the document tree into LaTeX files. List of tuples
# (source start file, target name, title, author, document class [howto/manual]).
latex_documents = [
- ('index', 'pyopencl.tex', 'PyOpenCL Documentation', 'Andreas Kloeckner', 'manual'),
-]
+ ('index', 'pyopencl.tex', 'PyOpenCL Documentation',
+ 'Andreas Kloeckner', 'manual'),
+ ]
# The name of an image file (relative to this directory) to place at the top of
# the title page.
diff --git a/doc/source/misc.rst b/doc/source/misc.rst
index 78a02f05da0db9ec5281cde04e25bb5c297bec51..6d9c7f198107acc63dcd55b856f0d44ea2215bcf 100644
--- a/doc/source/misc.rst
+++ b/doc/source/misc.rst
@@ -105,6 +105,7 @@ Version 2013.1
arrays will fail for now. This will be fixed in a future release.
* :class:`pyopencl.CommandQueue` may be used as a context manager (in a ``with`` statement)
* Add :func:`pyopencl.clmath.atan2`, :func:`pyopencl.clmath.atan2pi`.
+* Add :func:`pyopencl.array.concatenate`.
Version 2012.1
--------------
diff --git a/pyopencl/array.py b/pyopencl/array.py
index a48029c36f37207fb4ec27ce9770e26f69b8621b..9833559de02cf3e3db37d903ffb484d520e7e8e1 100644
--- a/pyopencl/array.py
+++ b/pyopencl/array.py
@@ -519,9 +519,15 @@ class Array(object):
events=self.events)
def with_queue(self, queue):
- """Return a copy of *self* with the default queue set to *queue*."""
+ """Return a copy of *self* with the default queue set to *queue*.
- assert queue.context == self.context
+ *None* is allowed as a value for *queue*.
+
+ .. versionadded:: 2013.1
+ """
+
+ if queue is not None:
+ assert queue.context == self.context
return self._new_with_changes(self.base_data, self.offset,
queue=queue)
@@ -1165,32 +1171,19 @@ class Array(object):
shape=tuple(new_shape),
strides=tuple(new_strides))
- def __setitem__(self, subscript, value):
- """Set the slice of *self* identified *subscript* to *value*.
-
- *value* is allowed to be:
-
- * A :class:`Array` of the same :attr:`shape` and (for now) :attr:`strides`,
- but with potentially different :attr:`dtype`.
- * A :class:`numpy.ndarray` of the same :attr:`shape` and (for now)
- :attr:`strides`, but with potentially different :attr:`dtype`.
- * A scalar.
-
- Non-scalar broadcasting is not currently supported.
-
- .. versionadded:: 2013.1
- """
+ def _setitem(self, subscript, value, queue=None):
+ queue = queue or self.queue or value.queue
subarray = self[subscript]
if isinstance(value, np.ndarray):
if subarray.shape == value.shape and subarray.strides == value.strides:
self.events.append(
- cl.enqueue_copy(self.queue, subarray.base_data,
+ cl.enqueue_copy(queue, subarray.base_data,
value, device_offset=subarray.offset))
return
else:
- value = to_device(self.queue, value, self.allocator)
+ value = to_device(queue, value, self.allocator)
if isinstance(value, Array):
if len(subarray.shape) != len(value.shape):
@@ -1203,11 +1196,28 @@ class Array(object):
raise ValueError("cannot assign between arrays of "
"differing strides")
- self._copy(subarray, value)
+ self._copy(subarray, value, queue=queue)
else:
# Let's assume it's a scalar
- subarray.fill(value)
+ subarray.fill(value, queue=queue)
+
+ def __setitem__(self, subscript, value):
+ """Set the slice of *self* identified *subscript* to *value*.
+
+ *value* is allowed to be:
+
+ * A :class:`Array` of the same :attr:`shape` and (for now) :attr:`strides`,
+ but with potentially different :attr:`dtype`.
+ * A :class:`numpy.ndarray` of the same :attr:`shape` and (for now)
+ :attr:`strides`, but with potentially different :attr:`dtype`.
+ * A scalar.
+
+ Non-scalar broadcasting is not currently supported.
+
+ .. versionadded:: 2013.1
+ """
+ self._setitem(subscript, value)
# }}}
@@ -1605,6 +1615,54 @@ def multi_put(arrays, dest_indices, dest_shape=None, out=None, queue=None):
return out
+
+def concatenate(arrays, axis=0, queue=None, allocator=None):
+ # {{{ find properties of result array
+
+ shape = None
+
+ for i_ary, ary in enumerate(arrays):
+ queue = queue or ary.queue
+ allocator = allocator or ary.allocator
+
+ if shape is None:
+ # first array
+ shape = list(ary.shape)
+ else:
+ if len(ary.shape) != len(shape):
+ raise ValueError("%d'th array has different number of axes "
+ "(shold have %d, has %d)"
+ % (i_ary, len(ary.shape), len(shape)))
+
+ ary_shape_list = list(ary.shape)
+ if (ary_shape_list[:axis] != shape[:axis]
+ or ary_shape_list[axis+1:] != shape[axis+1:]):
+ raise ValueError("%d'th array has residual not matching "
+ "other arrays" % i_ary)
+
+ shape[axis] += ary.shape[axis]
+
+ # }}}
+
+ shape = tuple(shape)
+ dtype = np.find_common_type([ary.dtype for ary in arrays], [])
+ result = empty(queue, shape, dtype, allocator=allocator)
+
+ full_slice = (slice(None),) * len(shape)
+
+ base_idx = 0
+ for ary in arrays:
+ my_len = ary.shape[axis]
+ result._setitem(
+ full_slice[:axis]
+ + (slice(base_idx, base_idx+my_len),)
+ + full_slice[axis+1:],
+ ary)
+
+ base_idx += my_len
+
+ return result
+
# }}}
diff --git a/test/test_array.py b/test/test_array.py
index eb72bf4c901b95706ac106d2675eb960a6b86d9c..a2f4739175d80a49fb34d17db2cd03ee28f114b7 100644
--- a/test/test_array.py
+++ b/test/test_array.py
@@ -582,6 +582,26 @@ def test_slice(ctx_factory):
assert la.norm(a_gpu.get() - a) == 0
+@pytools.test.mark_test.opencl
+def test_concatenate(ctx_factory):
+ context = ctx_factory()
+ queue = cl.CommandQueue(context)
+
+ from pyopencl.clrandom import rand as clrand
+
+ a_dev = clrand(queue, (5, 15, 20), dtype=np.float32)
+ b_dev = clrand(queue, (4, 15, 20), dtype=np.float32)
+ c_dev = clrand(queue, (3, 15, 20), dtype=np.float32)
+ a = a_dev.get()
+ b = b_dev.get()
+ c = c_dev.get()
+
+ cat_dev = cl.array.concatenate((a_dev, b_dev, c_dev))
+ cat = np.concatenate((a, b, c))
+
+ assert la.norm(cat - cat_dev.get()) == 0
+
+
if __name__ == "__main__":
# make sure that import failures get reported, instead of skipping the
# tests.