[RFC] ring: remove unnecessary fences in C11 ring for performance

Message ID 20230612194716.1050379-2-wathsala.vithanage@arm.com (mailing list archive)
State Changes Requested
Delegated to: Thomas Monjalon
Headers
Series [RFC] ring: remove unnecessary fences in C11 ring for performance |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/loongarch-compilation success Compilation OK
ci/loongarch-unit-testing success Unit Testing PASS
ci/Intel-compilation success Compilation OK
ci/intel-Testing success Testing PASS
ci/intel-Functional success Functional PASS

Commit Message

Wathsala Wathawana Vithanage June 12, 2023, 7:47 p.m. UTC
  For improved performance over the current C11 based ring
implementation, two ACQUIRE fences between the load of the old_head
and the load of the const_tail in __rte_ring_move_prod_head and the
load of old_head and load of the prod_tail in
__rte_ring_move_cons_head functions have been removed. These two
fences are not required for the safety of the ring library.

Signed-off-by: Wathsala Vithanage <wathsala.vithanage@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
---
 .mailmap                    | 1 +
 lib/ring/rte_ring_c11_pvt.h | 6 ------
 2 files changed, 1 insertion(+), 6 deletions(-)
  

Comments

Stephen Hemminger June 13, 2024, 3:34 p.m. UTC | #1
On Mon, 12 Jun 2023 19:47:16 +0000
Wathsala Vithanage <wathsala.vithanage@arm.com> wrote:

> For improved performance over the current C11 based ring
> implementation, two ACQUIRE fences between the load of the old_head
> and the load of the const_tail in __rte_ring_move_prod_head and the
> load of old_head and load of the prod_tail in
> __rte_ring_move_cons_head functions have been removed. These two
> fences are not required for the safety of the ring library.
> 
> Signed-off-by: Wathsala Vithanage <wathsala.vithanage@arm.com>
> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>

Makes sense, but needs to be rebased with current rte_thread changes.
  

Patch

diff --git a/.mailmap b/.mailmap
index 4018f0fc47..367115d134 100644
--- a/.mailmap
+++ b/.mailmap
@@ -1430,6 +1430,7 @@  Walter Heymans <walter.heymans@corigine.com>
 Wang Sheng-Hui <shhuiw@gmail.com>
 Wangyu (Eric) <seven.wangyu@huawei.com>
 Waterman Cao <waterman.cao@intel.com>
+Wathsala Vithanage <wathsala.vithanage@arm.com>
 Weichun Chen <weichunx.chen@intel.com>
 Wei Dai <wei.dai@intel.com>
 Weifeng Li <liweifeng96@126.com>
diff --git a/lib/ring/rte_ring_c11_pvt.h b/lib/ring/rte_ring_c11_pvt.h
index f895950df4..eb59c61b6a 100644
--- a/lib/ring/rte_ring_c11_pvt.h
+++ b/lib/ring/rte_ring_c11_pvt.h
@@ -66,9 +66,6 @@  __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
 		/* Reset n to the initial burst count */
 		n = max;
 
-		/* Ensure the head is read before tail */
-		__atomic_thread_fence(__ATOMIC_ACQUIRE);
-
 		/* load-acquire synchronize with store-release of ht->tail
 		 * in update_tail.
 		 */
@@ -142,9 +139,6 @@  __rte_ring_move_cons_head(struct rte_ring *r, int is_sc,
 		/* Restore n as it may change every loop */
 		n = max;
 
-		/* Ensure the head is read before tail */
-		__atomic_thread_fence(__ATOMIC_ACQUIRE);
-
 		/* this load-acquire synchronize with store-release of ht->tail
 		 * in update_tail.
 		 */