[Python-checkins] cpython: Bump the blocksize up from 62 to 64 to speed up the modulo calculation.

raymond.hettinger python-checkins at python.org
Fri Feb 27 08:21:41 CET 2015


https://hg.python.org/cpython/rev/430cab0c5cbf
changeset:   94768:430cab0c5cbf
user:        Raymond Hettinger <python at rcn.com>
date:        Thu Feb 26 23:21:29 2015 -0800
summary:
  Bump the blocksize up from 62 to 64 to speed up the modulo calculation.

Remove the old comment suggesting that it was desireable to have
blocksize+2 as a multiple of the cache line length.  That would
have made sense only if the block structure start point was always
aligned to a cache line boundary.  However, the memory allocations
are 16 byte aligned, so we don't really have control over whether
the struct spills across cache line boundaries.

files:
  Lib/test/test_deque.py       |  2 +-
  Modules/_collectionsmodule.c |  9 +++------
  2 files changed, 4 insertions(+), 7 deletions(-)


diff --git a/Lib/test/test_deque.py b/Lib/test/test_deque.py
--- a/Lib/test/test_deque.py
+++ b/Lib/test/test_deque.py
@@ -542,7 +542,7 @@
 
     @support.cpython_only
     def test_sizeof(self):
-        BLOCKLEN = 62
+        BLOCKLEN = 64
         basesize = support.calcobjsize('2P4nlP')
         blocksize = struct.calcsize('2P%dP' % BLOCKLEN)
         self.assertEqual(object.__sizeof__(deque()), basesize)
diff --git a/Modules/_collectionsmodule.c b/Modules/_collectionsmodule.c
--- a/Modules/_collectionsmodule.c
+++ b/Modules/_collectionsmodule.c
@@ -10,14 +10,11 @@
 /* The block length may be set to any number over 1.  Larger numbers
  * reduce the number of calls to the memory allocator, give faster
  * indexing and rotation, and reduce the link::data overhead ratio.
- *
- * Ideally, the block length will be set to two less than some
- * multiple of the cache-line length (so that the full block
- * including the leftlink and rightlink will fit neatly into
- * cache lines).
+ * Making the block length a power of two speeds-up the modulo
+ * calculation in deque_item().
  */
 
-#define BLOCKLEN 62
+#define BLOCKLEN 64
 #define CENTER ((BLOCKLEN - 1) / 2)
 
 /* A `dequeobject` is composed of a doubly-linked list of `block` nodes.

-- 
Repository URL: https://hg.python.org/cpython


More information about the Python-checkins mailing list