summaryrefslogtreecommitdiff
path: root/doc
diff options
context:
space:
mode:
authorAntoine Pitrou <pitrou@free.fr>2017-06-20 16:39:19 +0200
committerCharles Harris <charlesr.harris@gmail.com>2017-06-20 08:39:19 -0600
commitb7050a9732a46c52c3661b5b5869fbfdac1a39bd (patch)
tree3b635e56883ad1b8e31e0bda082203574d1b72af /doc
parent4e6a8f0efa09f3af04b900c1ec796622eaf8141c (diff)
downloadnumpy-b7050a9732a46c52c3661b5b5869fbfdac1a39bd.tar.gz
ENH: Release the GIL in einsum() special-cased loops (#9246)
* ENH: Release the GIL in einsum() special-cased loops This does not only allow parallelization but also fixes issues where a background IO loop would suffer long wait times due to a large concurrent einsum() in another thread. * Use NPY_BEGIN_THREADS_THRESHOLDED * Add release notes entry
Diffstat (limited to 'doc')
-rw-r--r--doc/release/1.14.0-notes.rst7
1 files changed, 7 insertions, 0 deletions
diff --git a/doc/release/1.14.0-notes.rst b/doc/release/1.14.0-notes.rst
index 3390ffd36..8a7492eb7 100644
--- a/doc/release/1.14.0-notes.rst
+++ b/doc/release/1.14.0-notes.rst
@@ -40,6 +40,13 @@ New Features
Improvements
============
+GIL released for all ``np.einsum`` variations
+---------------------------------------------
+
+Some specific loop structures which have an accelerated loop version
+did not release the GIL prior to NumPy 1.14.0. This oversight has been
+fixed.
+
Changes
=======