From 51c736a81c8cc51a089dc7a3d1628c4e4742d61d Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Tue, 27 Nov 2012 16:56:56 -0500
Subject: [PATCH] Default order to "C" for enqueue_map_*.
---
doc/source/runtime.rst | 22 ++++++++++++++++++++--
src/wrapper/wrap_cl_part_2.cpp | 6 ++++--
2 files changed, 24 insertions(+), 4 deletions(-)
diff --git a/doc/source/runtime.rst b/doc/source/runtime.rst
index 4dd2c14f..547ce018 100644
--- a/doc/source/runtime.rst
+++ b/doc/source/runtime.rst
@@ -601,7 +601,7 @@ Mapping Memory into Host Address Space
.. method:: release(queue=None, wait_for=None)
-.. function:: enqueue_map_buffer(queue, buf, flags, offset, shape, dtype, order, wait_for=None, is_blocking=True)
+.. function:: enqueue_map_buffer(queue, buf, flags, offset, shape, dtype, order="C", wait_for=None, is_blocking=True)
|explain-waitfor|
*shape*, *dtype*, and *order* have the same meaning
@@ -616,7 +616,10 @@ Mapping Memory into Host Address Space
.. versionchanged:: 2011.1
*is_blocking* now defaults to True.
-.. function:: enqueue_map_image(queue, buf, flags, origin, region, shape, dtype, order, wait_for=None, is_blocking=True)
+ .. versionchanged:: 2012.2
+ *order* now defaults to "C".
+
+.. function:: enqueue_map_image(queue, buf, flags, origin, region, shape, dtype, order="C", wait_for=None, is_blocking=True)
|explain-waitfor|
*shape*, *dtype*, and *order* have the same meaning
@@ -631,6 +634,8 @@ Mapping Memory into Host Address Space
.. versionchanged:: 2011.1
*is_blocking* now defaults to True.
+ .. versionchanged:: 2012.2
+ *order* now defaults to "C".
Samplers
^^^^^^^^
@@ -830,6 +835,19 @@ Programs and Kernels
*global_size* and *local_size* also do not have to have the same number
of dimensions.
+ .. note::
+
+ :meth:`__call__` is *not* thread-safe. It sets the arguments using :meth:`set_args`
+ and then runs :func:`enqueue_nd_range_kernel`. Another thread could race it
+ in doing the same things, with undefined outcome. This issue is inherited
+ from the C-level OpenCL API. The recommended solution is to make a kernel
+ (i.e. access `prg.kernel_name`, which corresponds to making a new kernel)
+ for every thread that may enqueue calls to the kernel.
+
+ A solution involving implicit locks was discussed and decided against on the
+ mailing list in `October 2012
+ <http://lists.tiker.net/pipermail/pyopencl/2012-October/001311.html>`_.
+
.. versionchanged:: 0.92
*local_size* was promoted to third positional argument from being a
keyword argument. The old keyword argument usage will continue to
diff --git a/src/wrapper/wrap_cl_part_2.cpp b/src/wrapper/wrap_cl_part_2.cpp
index cc7942c8..c48ee7ae 100644
--- a/src/wrapper/wrap_cl_part_2.cpp
+++ b/src/wrapper/wrap_cl_part_2.cpp
@@ -151,13 +151,15 @@ void pyopencl_expose_part_2()
py::def("enqueue_map_buffer", enqueue_map_buffer,
(py::args("queue", "buf", "flags",
"offset",
- "shape", "dtype", "order"),
+ "shape", "dtype"),
+ py::arg("order")="C",
py::arg("wait_for")=py::object(),
py::arg("is_blocking")=true));
py::def("enqueue_map_image", enqueue_map_image,
(py::args("queue", "img", "flags",
"origin", "region",
- "shape", "dtype", "order"),
+ "shape", "dtype"),
+ py::arg("order")="C",
py::arg("wait_for")=py::object(),
py::arg("is_blocking")=true));
--
GitLab