r600g/sb: fix handling of new multislot instructions on cayman

Ex-scalar instructions that became multislot on cayman do replicate result
to all channels - handle them similar to DOT4.

Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>

3 files changed, 6 insertions, 5 deletions

diff --git a/src/gallium/drivers/r600/sb/sb_ir.cpp b/src/gallium/drivers/r600/sb/sb_ir.cpp
index 382fceed08..5226893de7 100644
--- a/src/gallium/drivers/r600/sb/sb_ir.cpp
+++ b/src/gallium/drivers/r600/sb/sb_ir.cpp
@@ -51,7 +51,7 @@ bool alu_packed_node::accept(vpass& p, bool enter) {
 	return p.visit(*this, enter);
 }
 
-void alu_packed_node::init_args() {
+void alu_packed_node::init_args(bool repl) {
 	alu_node *p = static_cast<alu_node*>(first);
 	assert(p->is_valid());
 	while (p) {
@@ -60,8 +60,6 @@ void alu_packed_node::init_args() {
 		p = static_cast<alu_node*>(p->next);
 	}
 
-	// if it's packed then it's always multislot, no need to check slot flags
-	bool repl = (op_ptr()->flags & AF_REPL);
 	value *replicated_value = NULL;
 
 	for (vvec::iterator I = dst.begin(), E = dst.end(); I != E; ++I) {
diff --git a/src/gallium/drivers/r600/sb/sb_ir.h b/src/gallium/drivers/r600/sb/sb_ir.h
index d55f4de67b..c838f62b56 100644
--- a/src/gallium/drivers/r600/sb/sb_ir.h
+++ b/src/gallium/drivers/r600/sb/sb_ir.h
@@ -1014,7 +1014,7 @@ public:
 		return static_cast<alu_node*>(first)->bc.op_ptr;
 	}
 	unsigned op() { return static_cast<alu_node*>(first)->bc.op; }
-	void init_args();
+	void init_args(bool repl);
 
 	virtual bool is_valid() { return subtype == NST_ALU_PACKED_INST; }
 	virtual bool accept(vpass &p, bool enter);
diff --git a/src/gallium/drivers/r600/sb/sb_ssa_builder.cpp b/src/gallium/drivers/r600/sb/sb_ssa_builder.cpp
index 87eafaee24..3ad628bb68 100644
--- a/src/gallium/drivers/r600/sb/sb_ssa_builder.cpp
+++ b/src/gallium/drivers/r600/sb/sb_ssa_builder.cpp
@@ -218,7 +218,10 @@ bool ssa_rename::visit(alu_packed_node& n, bool enter) {
 			I->accept(*this, false);
 		}
 
-		n.init_args();
+		bool repl = (n.op_ptr()->flags & AF_REPL) ||
+				(ctx.is_cayman() && (n.first->alu_op_slot_flags() & AF_S));
+
+		n.init_args(repl);
 	}
 	return false;
 }