glsl/lower_ubo_reference: split struct copies into element copies

Improves register pressure, since otherwise we end up emitting
loads for all the elements in the RHS and them emitting
stores for all elements in the LHS.

Reviewed-by: Jordan Justen <[email protected]>

1 files changed, 54 insertions, 2 deletions

diff --git a/src/glsl/lower_ubo_reference.cpp b/src/glsl/lower_ubo_reference.cpp
index b82d800eefb..8ec83465b69 100644
--- a/src/glsl/lower_ubo_reference.cpp
+++ b/src/glsl/lower_ubo_reference.cpp
@@ -155,6 +155,7 @@ public:
                       ir_rvalue *offset);
 
    bool check_for_buffer_array_copy(ir_assignment *ir);
+   bool check_for_buffer_struct_copy(ir_assignment *ir);
    void check_for_ssbo_store(ir_assignment *ir);
    void write_to_memory(ir_dereference *deref,
                         ir_variable *var,
@@ -1190,12 +1191,60 @@ lower_ubo_reference_visitor::check_for_buffer_array_copy(ir_assignment *ir)
    return true;
 }
 
+bool
+lower_ubo_reference_visitor::check_for_buffer_struct_copy(ir_assignment *ir)
+{
+   if (!ir || !ir->lhs || !ir->rhs)
+      return false;
+
+   /* LHS and RHS must be records */
+   if (!ir->lhs->type->is_record() || !ir->rhs->type->is_record())
+      return false;
+
+   /* RHS must be a buffer-backed variable. This is what can cause the problem
+    * since it would lead to a series of loads that need to live until we
+    * see the writes to the LHS.
+    */
+   ir_variable *rhs_var = ir->rhs->variable_referenced();
+   if (!rhs_var || !is_buffer_backed_variable(rhs_var))
+      return false;
+
+   /* Split the struct copy into individual element copies to reduce
+    * register pressure
+    */
+   ir_dereference *rhs_deref = ir->rhs->as_dereference();
+   if (!rhs_deref)
+      return false;
+
+   ir_dereference *lhs_deref = ir->lhs->as_dereference();
+   if (!lhs_deref)
+      return false;
+
+   assert(lhs_deref->type->record_compare(rhs_deref->type));
+   mem_ctx = ralloc_parent(shader->ir);
+
+   for (unsigned i = 0; i < lhs_deref->type->length; i++) {
+      const char *field_name = lhs_deref->type->fields.structure[i].name;
+      ir_dereference *lhs_field =
+         new(mem_ctx) ir_dereference_record(lhs_deref->clone(mem_ctx, NULL),
+                                            field_name);
+      ir_dereference *rhs_field =
+         new(mem_ctx) ir_dereference_record(rhs_deref->clone(mem_ctx, NULL),
+                                            field_name);
+      ir->insert_after(assign(lhs_field, rhs_field));
+   }
+
+   ir->remove();
+   progress = true;
+   return true;
+}
+
 ir_visitor_status
 lower_ubo_reference_visitor::visit_enter(ir_assignment *ir)
 {
-   /* Array copies could involve large amounts of load/store
+   /* Array and struct copies could involve large amounts of load/store
     * operations. To improve register pressure we want to special-case
-    * these and split array copies into individual element copies.
+    * these and split them into individual element copies.
     * This way we avoid emitting all the loads for the RHS first and
     * all the writes for the LHS second and register usage is more
     * efficient.
@@ -1203,6 +1252,9 @@ lower_ubo_reference_visitor::visit_enter(ir_assignment *ir)
    if (check_for_buffer_array_copy(ir))
       return visit_continue_with_parent;
 
+   if (check_for_buffer_struct_copy(ir))
+      return visit_continue_with_parent;
+
    check_ssbo_unsized_array_length_assignment(ir);
    check_for_ssbo_store(ir);
    return rvalue_visit(ir);