summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMatt Turner <[email protected]>2016-08-18 16:47:05 -0700
committerMatt Turner <[email protected]>2016-08-19 16:52:25 -0700
commita73116ecc60414ade89802150b707b3336d8d50f (patch)
tree8ff8f65d3552fa4bc728ca8917e4a6e9b87af86d
parente7c376adfdecd4c1333997c8be8bb066a87c67b4 (diff)
i965/sched: Simplify work done by add_barrier_deps().
Scheduling barriers are implemented by placing a dependence on every node before and after the barrier. This is unnecessary as we can limit the number of nodes we place dependencies on to those between us and the next barrier in each direction. Runtime of dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer.23 is reduced from ~25 minutes to a little more than three. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94681 Reviewed-by: Kenneth Graunke <[email protected]>
-rw-r--r--src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp9
1 files changed, 9 insertions, 0 deletions
diff --git a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
index dfcaa80b58b..a8987065f18 100644
--- a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
+++ b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
@@ -94,6 +94,8 @@ public:
* successors is an exit node.
*/
schedule_node *exit;
+
+ bool is_barrier;
};
/**
@@ -800,6 +802,7 @@ schedule_node::schedule_node(backend_instruction *inst,
this->cand_generation = 0;
this->delay = 0;
this->exit = NULL;
+ this->is_barrier = false;
/* We can't measure Gen6 timings directly but expect them to be much
* closer to Gen7 than Gen4.
@@ -931,9 +934,13 @@ instruction_scheduler::add_barrier_deps(schedule_node *n)
schedule_node *prev = (schedule_node *)n->prev;
schedule_node *next = (schedule_node *)n->next;
+ n->is_barrier = true;
+
if (prev) {
while (!prev->is_head_sentinel()) {
add_dep(prev, n, 0);
+ if (prev->is_barrier)
+ break;
prev = (schedule_node *)prev->prev;
}
}
@@ -941,6 +948,8 @@ instruction_scheduler::add_barrier_deps(schedule_node *n)
if (next) {
while (!next->is_tail_sentinel()) {
add_dep(n, next, 0);
+ if (next->is_barrier)
+ break;
next = (schedule_node *)next->next;
}
}