Our copy propagation tends to be bad at handling the later array
accesses of the matrix argument we moved to a temporary. Generally we
don't need to move it to a temporary, though, so this avoids needing
more copy propagation complexity.
Reduces instruction count of some Unigine Tropics and Sanctuary
fragment shaders that do operations on uniform matrix arrays by 5.9%
on gen6.
---
src/glsl/lower_mat_op_to_vec.cpp | 11 +++++++++++
1 files changed, 11 insertions(+), 0 deletions(-)
diff --git a/src/glsl/lower_mat_op_to_vec.cpp b/src/glsl/lower_mat_op_to_vec.cpp
index fe89802..9d593f9 100644
--- a/src/glsl/lower_mat_op_to_vec.cpp
+++ b/src/glsl/lower_mat_op_to_vec.cpp
@@ -329,7 +329,18 @@ ir_mat_op_to_vec_visitor::visit_leave(ir_assignment
*orig_assign)
*/
for (i = 0; i < orig_expr->get_num_operands(); i++) {
ir_assignment *assign;
+ ir_dereference *deref = orig_expr->operands[i]->as_dereference();
+ /* Avoid making a temporary if we don't need to to avoid aliasing. */
+ if (deref &&
+ deref->variable_referenced() != result->variable_referenced()) {
+ op[i] = deref;
+ continue;
+ }
+
+ /* Otherwise, store the operand in a temporary generally if it's
+ * not a dereference.
+ */
ir_variable *var = new(mem_ctx) ir_variable(orig_expr->operands[i]->type,
"mat_op_to_vec",
ir_var_temporary);
--
1.7.5.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
