Merged,
please check.
Maxim.
On 01/26/2016 15:41, Savolainen, Petri (Nokia - FI/Espoo) wrote:
Thanks. Maxim could you first try to merge and resolve conflicts.
Those should be quite simple to resolve: additional atomic validation
tests into validation/atomic/atomic.c instead of old
validation/synchronizers/synchronizers.c.
-Petri
*From:*EXT Ola Liljedahl [mailto:ola.liljed...@linaro.org]
*Sent:* Tuesday, January 26, 2016 2:16 PM
*To:* Savolainen, Petri (Nokia - FI/Espoo)
*Cc:* LNG ODP Mailman List
*Subject:* Re: [lng-odp] [API-NEXT PATCH v3 00/13] atomi api completion
On 20 January 2016 at 10:19, Petri Savolainen
<petri.savolai...@nokia.com <mailto:petri.savolai...@nokia.com>> wrote:
Completed additions to atomic API. Added exchange operation in
relaxed memory
model. Added CAS operations in rel and acq_rel memory order. Added
64 bit
versions of all non-relaxed operations defined so far. Added all
missing
validation tests. Replaced internal atomics usage in lock, etc code
(all except spinlock and timer) with API calls.
This set of atomic calls seem to be sufficient to current
linux-generic use
cases. Timer and spinlock use atomic flag and 128 bit types, but
does not
require additional operation types (e.g. load_acq, store_rel, cas,
cas_rel, etc
are supported by the API).
Additional operations can be added later if needed. Only commonly
needed
combinations of non-relaxed operations will be supported (added
based on use
case as any other ODP API). This set should already enable major
part of
lock-free algorithms.
They enable a lot of *lock-less* implementations but not all lock-less
designs are properly "lock-free". Lock-free implies that no thread can
block other other threads from doing progress. E.g. the DPDK ring
buffer we copied is lock-less (based on CAS) but not lock-free, a
thread which has entered the critical section (between CAS and the
subsequent releasing store) must exit it before other threads can exit
the critical section. The lock-less design enables some parallelism in
operating on the data structure.
v3:
* Corrected race condition in xchg validation test
v2:
* rebased on latest api-next
Failed to actually build and run validation tests since some patches
require some manual effort to merge. We agreed Petri & Maxim would do
that.
Whole patch set
Reviewed-by: Ola Liljedahl <ola.liljed...@linaro.org
<mailto:ola.liljed...@linaro.org>>
Petri Savolainen (13):
api: atomic: rename release ordering
api: atomic: added 32bit cas_rel and cas_acq_rel
linux-generic: atomic: 32bit cas_rel and cas_acq_rel
api: atomic: add non-relaxed 64bit operations
linux-generic: atomic: non-relaxed 64bit operations
api: atomic: added relaxed exchange operation
linux-generic: atomic: implemented exchange
validation: atomic: added max and min tests
validation: atomic: added cas test
validation: atomic: added xchg test
validation: atomic: added non-relaxed test
linux-generic: locks: replace internal atomics
linux-generic: barrier: use API memory barrier
include/odp/api/atomic.h | 220 +++++++++++--
platform/linux-generic/include/odp/atomic.h | 142 ++++++++-
.../linux-generic/include/odp_atomic_internal.h | 6 -
platform/linux-generic/include/odp_pool_internal.h | 1 -
platform/linux-generic/odp_barrier.c | 8 +-
platform/linux-generic/odp_pool.c | 1 -
platform/linux-generic/odp_rwlock.c | 23 +-
platform/linux-generic/odp_ticketlock.c | 21 +-
test/validation/synchronizers/synchronizers.c | 347
++++++++++++++++++++-
test/validation/synchronizers/synchronizers.h | 4 +
10 files changed, 689 insertions(+), 84 deletions(-)
--
2.6.3
_______________________________________________
lng-odp mailing list
lng-odp@lists.linaro.org <mailto:lng-odp@lists.linaro.org>
https://lists.linaro.org/mailman/listinfo/lng-odp
_______________________________________________
lng-odp mailing list
lng-odp@lists.linaro.org
https://lists.linaro.org/mailman/listinfo/lng-odp
