[PATCH] event/dsw: optimize serving port logic

["Mattias Rönnblom" <mattias.ronnblom@ericsson.com>]

To reduce flow migration overhead, replace the array-based
representation of which set of ports are bound to a particular queue
by a bitmask-based one.

The maximum number of DSW event ports remains 64, but after this
change can no longer easily be increased by modifying DSW_MAX_PORTS
and recompile.

Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
---
 drivers/event/dsw/dsw_evdev.c | 39 +++++++++++++++++++++--------------
 drivers/event/dsw/dsw_evdev.h |  5 ++++-
 drivers/event/dsw/dsw_event.c | 10 +++------
 3 files changed, 31 insertions(+), 23 deletions(-)

diff --git a/drivers/event/dsw/dsw_evdev.c b/drivers/event/dsw/dsw_evdev.c
index 1209e73a9d..629c929cb2 100644
--- a/drivers/event/dsw/dsw_evdev.c
+++ b/drivers/event/dsw/dsw_evdev.c
@@ -144,24 +144,23 @@ dsw_queue_release(struct rte_eventdev *dev __rte_unused,
 static void
 queue_add_port(struct dsw_queue *queue, uint16_t port_id)
 {
-	queue->serving_ports[queue->num_serving_ports] = port_id;
+	uint64_t port_mask = UINT64_C(1) << port_id;
+
+	queue->serving_ports |=	port_mask;
 	queue->num_serving_ports++;
 }
 
 static bool
 queue_remove_port(struct dsw_queue *queue, uint16_t port_id)
 {
-	uint16_t i;
+	uint64_t port_mask = UINT64_C(1) << port_id;
+
+	if (queue->serving_ports & port_mask) {
+		queue->num_serving_ports--;
+		queue->serving_ports ^= port_mask;
+		return true;
+	}
 
-	for (i = 0; i < queue->num_serving_ports; i++)
-		if (queue->serving_ports[i] == port_id) {
-			uint16_t last_idx = queue->num_serving_ports - 1;
-			if (i != last_idx)
-				queue->serving_ports[i] =
-					queue->serving_ports[last_idx];
-			queue->num_serving_ports--;
-			return true;
-		}
 	return false;
 }
 
@@ -256,10 +255,20 @@ initial_flow_to_port_assignment(struct dsw_evdev *dsw)
 		struct dsw_queue *queue = &dsw->queues[queue_id];
 		uint16_t flow_hash;
 		for (flow_hash = 0; flow_hash < DSW_MAX_FLOWS; flow_hash++) {
-			uint8_t port_idx =
-				rte_rand() % queue->num_serving_ports;
-			uint8_t port_id =
-				queue->serving_ports[port_idx];
+			uint8_t skip =
+				rte_rand_max(queue->num_serving_ports);
+			uint8_t port_id;
+
+			for (port_id = 0;; port_id++) {
+				uint64_t port_mask = UINT64_C(1) << port_id;
+
+				if (queue->serving_ports & port_mask) {
+					if (skip == 0)
+						break;
+					skip--;
+				}
+			}
+
 			dsw->queues[queue_id].flow_to_port_map[flow_hash] =
 				port_id;
 		}
diff --git a/drivers/event/dsw/dsw_evdev.h b/drivers/event/dsw/dsw_evdev.h
index 6416a8a898..8166340e1e 100644
--- a/drivers/event/dsw/dsw_evdev.h
+++ b/drivers/event/dsw/dsw_evdev.h
@@ -234,12 +234,15 @@ struct dsw_port {
 
 struct dsw_queue {
 	uint8_t schedule_type;
-	uint8_t serving_ports[DSW_MAX_PORTS];
+	uint64_t serving_ports;
 	uint16_t num_serving_ports;
 
 	uint8_t flow_to_port_map[DSW_MAX_FLOWS] __rte_cache_aligned;
 };
 
+/* Limited by the size of the 'serving_ports' bitmask */
+static_assert(DSW_MAX_PORTS <= 64);
+
 struct dsw_evdev {
 	struct rte_eventdev_data *data;
 
diff --git a/drivers/event/dsw/dsw_event.c b/drivers/event/dsw/dsw_event.c
index 93bbeead2e..23488d9030 100644
--- a/drivers/event/dsw/dsw_event.c
+++ b/drivers/event/dsw/dsw_event.c
@@ -447,13 +447,9 @@ static bool
 dsw_is_serving_port(struct dsw_evdev *dsw, uint8_t port_id, uint8_t queue_id)
 {
 	struct dsw_queue *queue = &dsw->queues[queue_id];
-	uint16_t i;
-
-	for (i = 0; i < queue->num_serving_ports; i++)
-		if (queue->serving_ports[i] == port_id)
-			return true;
+	uint64_t port_mask = UINT64_C(1) << port_id;
 
-	return false;
+	return queue->serving_ports & port_mask;
 }
 
 static bool
@@ -575,7 +571,7 @@ dsw_schedule(struct dsw_evdev *dsw, uint8_t queue_id, uint16_t flow_hash)
 		/* A single-link queue, or atomic/ordered/parallel but
 		 * with just a single serving port.
 		 */
-		port_id = queue->serving_ports[0];
+		port_id = rte_bsf64(queue->serving_ports);
 
 	DSW_LOG_DP(DEBUG, "Event with queue_id %d flow_hash %d is scheduled "
 		   "to port %d.\n", queue_id, flow_hash, port_id);
-- 
2.34.1