[PATCH 2/2] drm/i915/execlists: Cache the last priolist lookup

[Chris Wilson]

From: Michał Winiarski <michal.winiarski at intel.com>

Avoid the repeated rbtree lookup for each request as we unwind them by
tracking the last priolist.

v2: Fix up my unhelpful suggestion of using default_priolist.

Signed-off-by: Michał Winiarski <michal.winiarski at intel.com>
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_request.c |  2 +-
 drivers/gpu/drm/i915/i915_gem_request.h |  1 +
 drivers/gpu/drm/i915/intel_lrc.c        | 24 ++++++++++++++++--------
 3 files changed, 18 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 4eb1a76731b2..14956d899911 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -186,7 +186,7 @@ i915_priotree_init(struct i915_priotree *pt)
 	INIT_LIST_HEAD(&pt->signalers_list);
 	INIT_LIST_HEAD(&pt->waiters_list);
 	INIT_LIST_HEAD(&pt->link);
-	pt->priority = INT_MIN;
+	pt->priority = I915_PRIORITY_INVALID;
 }
 
 static int reset_all_global_seqno(struct drm_i915_private *i915, u32 seqno)
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index 96eb52471dad..f520114833b1 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -69,6 +69,7 @@ struct i915_priotree {
 	struct list_head waiters_list; /* those after us, they depend upon us */
 	struct list_head link;
 	int priority;
+#define I915_PRIORITY_INVALID INT_MIN
 #define I915_PRIORITY_MAX 1024
 #define I915_PRIORITY_NORMAL 0
 #define I915_PRIORITY_MIN (-I915_PRIORITY_MAX)
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index cbac2fff8e4d..7d6da130b184 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -358,25 +358,31 @@ static void unwind_wa_tail(struct drm_i915_gem_request *rq)
 static void unwind_incomplete_requests(struct intel_engine_cs *engine)
 {
 	struct drm_i915_gem_request *rq, *rn;
+	struct i915_priolist *uninitialized_var(p);
+	int last_prio = I915_PRIORITY_INVALID;
 
 	lockdep_assert_held(&engine->timeline->lock);
 
 	list_for_each_entry_safe_reverse(rq, rn,
 					 &engine->timeline->requests,
 					 link) {
-		struct i915_priolist *p;
-
 		if (i915_gem_request_completed(rq))
 			return;
 
 		__i915_gem_request_unsubmit(rq);
 		unwind_wa_tail(rq);
 
-		p = lookup_priolist(engine,
-				    &rq->priotree,
-				    rq->priotree.priority);
-		list_add(&rq->priotree.link,
-			 &ptr_mask_bits(p, 1)->requests);
+		GEM_BUG_ON(rq->priotree.priority == I915_PRIORITY_INVALID);
+		if (rq->priotree.priority != last_prio) {
+			p = lookup_priolist(engine,
+					    &rq->priotree,
+					    rq->priotree.priority);
+			p = ptr_mask_bits(p, 1);
+
+			last_prio = rq->priotree.priority;
+		}
+
+		list_add(&rq->priotree.link, &p->requests);
 	}
 }
 
@@ -863,6 +869,8 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio)
 	struct i915_dependency stack;
 	LIST_HEAD(dfs);
 
+	GEM_BUG_ON(prio == I915_PRIORITY_INVALID);
+
 	if (prio <= READ_ONCE(request->priotree.priority))
 		return;
 
@@ -911,7 +919,7 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio)
 	 * execlists_submit_request()), we can set our own priority and skip
 	 * acquiring the engine locks.
 	 */
-	if (request->priotree.priority == INT_MIN) {
+	if (request->priotree.priority == I915_PRIORITY_INVALID) {
 		GEM_BUG_ON(!list_empty(&request->priotree.link));
 		request->priotree.priority = prio;
 		if (stack.dfs_link.next == stack.dfs_link.prev)
-- 
2.14.1