r600/sb: adding lds oq tracking to the scheduler

This adds support for tracking the lds oq read/writes
so can avoid scheduling other things in between.

This patch just adds the tracking and assert to show
problems.

Acked-By: Roland Scheidegger <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

2 files changed, 15 insertions, 3 deletions

diff --git a/src/gallium/drivers/r600/sb/sb_sched.cpp b/src/gallium/drivers/r600/sb/sb_sched.cpp
index 6d7ab671ff5..26e4811b1ce 100644
--- a/src/gallium/drivers/r600/sb/sb_sched.cpp
+++ b/src/gallium/drivers/r600/sb/sb_sched.cpp
@@ -312,7 +312,7 @@ alu_group_tracker::alu_group_tracker(shader &sh)
 	  gpr(), lt(), slots(),
 	  max_slots(sh.get_ctx().is_cayman() ? 4 : 5),
 	  has_mova(), uses_ar(), has_predset(), has_kill(),
-	  updates_exec_mask(), chan_count(), interp_param(), next_id() {
+	  updates_exec_mask(), consumes_lds_oqa(), produces_lds_oqa(), chan_count(), interp_param(), next_id() {
 
 	available_slots = sh.get_ctx().has_trans ? 0x1F : 0x0F;
 }
@@ -680,6 +680,8 @@ void alu_group_tracker::reset(bool keep_packed) {
 	memset(slots, 0, sizeof(slots));
 	vmap.clear();
 	next_id = 0;
+	produces_lds_oqa = 0;
+	consumes_lds_oqa = 0;
 	has_mova = false;
 	uses_ar = false;
 	has_predset = false;
@@ -703,7 +705,8 @@ void alu_group_tracker::update_flags(alu_node* n) {
 	has_mova |= (flags & AF_MOVA);
 	has_predset |= (flags & AF_ANY_PRED);
 	uses_ar |= n->uses_ar();
-
+	consumes_lds_oqa |= n->consumes_lds_oq();
+	produces_lds_oqa |= n->produces_lds_oq();
 	if (flags & AF_ANY_PRED) {
 		if (n->dst[2] != NULL)
 			updates_exec_mask = true;
@@ -1958,6 +1961,7 @@ void alu_kcache_tracker::reset() {
 void alu_clause_tracker::reset() {
 	group = 0;
 	slot_count = 0;
+	outstanding_lds_oqa_reads = 0;
 	grp0.reset();
 	grp1.reset();
 }
@@ -1966,7 +1970,7 @@ alu_clause_tracker::alu_clause_tracker(shader &sh)
 	: sh(sh), kt(sh.get_ctx().hw_class), slot_count(),
 	  grp0(sh), grp1(sh),
 	  group(), clause(),
-	  push_exec_mask(),
+	  push_exec_mask(), outstanding_lds_oqa_reads(),
 	  current_ar(), current_pr(), current_idx() {}
 
 void alu_clause_tracker::emit_group() {
@@ -1988,6 +1992,8 @@ void alu_clause_tracker::emit_group() {
 
 	clause->push_front(g);
 
+	outstanding_lds_oqa_reads += grp().get_consumes_lds_oqa();
+	outstanding_lds_oqa_reads -= grp().get_produces_lds_oqa();
 	slot_count += grp().slot_count();
 
 	new_group();
@@ -2000,6 +2006,7 @@ void alu_clause_tracker::emit_clause(container_node *c) {
 
 	kt.init_clause(clause->bc);
 
+	assert(!outstanding_lds_oqa_reads);
 	assert(!current_ar);
 	assert(!current_pr);
 
diff --git a/src/gallium/drivers/r600/sb/sb_sched.h b/src/gallium/drivers/r600/sb/sb_sched.h
index 5a2663442bc..91a34e078d9 100644
--- a/src/gallium/drivers/r600/sb/sb_sched.h
+++ b/src/gallium/drivers/r600/sb/sb_sched.h
@@ -127,6 +127,8 @@ class alu_group_tracker {
 	bool has_kill;
 	bool updates_exec_mask;
 
+	bool consumes_lds_oqa;
+	bool produces_lds_oqa;
 	unsigned chan_count[4];
 
 	// param index + 1 (0 means that group doesn't refer to Params)
@@ -166,6 +168,8 @@ public:
 	unsigned literal_slot_count() { return (literal_count() + 1) >> 1; };
 	unsigned slot_count() { return inst_count() + literal_slot_count(); }
 
+	bool get_consumes_lds_oqa() { return consumes_lds_oqa; }
+	bool get_produces_lds_oqa() { return produces_lds_oqa; }
 	alu_group_node* emit();
 
 	rp_kcache_tracker& kcache() { return kc; }
@@ -212,6 +216,7 @@ class alu_clause_tracker {
 
 	bool push_exec_mask;
 
+	unsigned outstanding_lds_oqa_reads;
 public:
 	container_node conflict_nodes;