Add changelog.

diff --git a/doc/news/changes/minor/20171227MartinKronbichler b/doc/news/changes/minor/20171227MartinKronbichler
new file mode 100644 (file)
index 0000000..75b4080
--- /dev/null
+++ b/doc/news/changes/minor/20171227MartinKronbichler
@@ -0,0 +1,12 @@
+New: There is now a function VectorizedArray::streaming_store() that
+stores the content of a vectorized array into another array which bypasses the
+CPU's caches on supported architectures using _mm_stream_pd() intrinsics. This
+operation is useful because it can avoid the read-for-ownership memory access
+on large arrays and directly write to the destination for streaming
+stores. However, it can also be detrimental to performance in case the store
+would fit into caches. This new function is not used within the library
+because only the final user code knows the properties of hardware and whether
+a particular store pattern is so large that it will not fit into caches and
+thus benefit from this optimization.
+<br>
+(Martin Kronbichler, 2017/12/01)