Speeds up regular MCS locks
authorBarret Rhoden <brho@cs.berkeley.edu>
Tue, 7 May 2013 18:28:43 +0000 (11:28 -0700)
committerBarret Rhoden <brho@cs.berkeley.edu>
Tue, 7 May 2013 18:48:39 +0000 (11:48 -0700)
commit24ee9d8e07d08cce5911a2a820ba785ef793b5c3
treec20e22a75ca9dbccd311873a17f77b5a0ef3efe2
parentce34f872750251e062743efada30b741576c84b3
Speeds up regular MCS locks

Even when the qnode is allocated on the stack (perhaps especially then),
CL aligning it reduces cache-line contention (false sharing of the CL
with the main thread using its stack).

Also this exposes some lower-level MCS locking functions.
user/parlib/include/mcs.h