I changed the release state to use an enum instead of an unsigned int with #defines in order to make debugging easier. I also removed unnecessary atomic operations in the initializer function, and made some formatting fixes.
I added an extra argument to the CK_COHORT_INIT macro to allow users to specify custom pass limits
when using it. I also added a reference to the paper on which the cohort implementation was based.
The constants and macros in ck_cohort.h didn't conform to CK naming conventions. Additionally, I was able to remove a lot of unnecessary atomic operations and memory fences, which reduced latency by 10-20% and increased throughput using ticket locks by nearly an order of magnitude.
CK_LIST_INSERT_HEAD was incorrectly managing prev
pointer on insertion to non-empty list. This bug
would cause erroneous behavior on CK_LIST_REMOVE
to non-head elements. Unit test will be updated
for this regression.
An off-by-one was introduced in downgrade path from writer.
This can cause deadlock if a writer downgrades from a write lock.
Pointed out by Jeffrey Birnbaum <jmb...@...>.
Both LLVM-backed compilers and GCC incorrectly treat
a barrier-sandwiched load as a loop invariant in dequeue_spmc.
Forcing volatile atomic load semantics generates the right
thing.
Thanks to Devon O'Dell and Abel Mathew for help in catching
this issue.
The distinction between additive/exponential implementation
and geometric implementation does little but confuse users.
The terminology used in ck_backoff now reflects terminology
used in literature.
ck_backoff_gb has been removed.
This operation is of format:
CK_S*LIST_MOVE(a, b, linkage) and is equivalent to intializing
a with the contents of b. This is done in a manner that is atomic
with respect to readers. Read-only operations are still valid in
b, but behavior is undefined for write-side operations on b after
a MOVE operation.
Brendon Scheinman
Joao Fernandes
Samy Al Bahra
Abel Mathew
Samy Al Bahra