nir: Fix overlapping vars in nir_assign_io_var_locations()
authorConnor Abbott <cwabbott0@gmail.com>
Tue, 24 Sep 2019 15:29:53 +0000 (17:29 +0200)
committerConnor Abbott <cwabbott0@gmail.com>
Wed, 25 Sep 2019 13:53:50 +0000 (15:53 +0200)
When handling two variables with overlapping locations, we process the
one with lower location first, and then extend the location ->
driver_location map to guarantee that it's contiguous for the second
variable too. But the loop had the wrong bound, so we weren't extending
the map 100%, which could lead to problems later such as an incorrect
num_inputs. The loop index i is an index into the slots of the variable,
so we need to stop at the final slot of the variable (var_size) instead
of the number of unassigned slots.

This fixes
spec@arb_enhanced_layouts@execution@component-layout@vs-fs-array-interleave-range
on radeonsi NIR.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
src/compiler/nir/nir_linking_helpers.c

index 28d2774cae205d7cf08c8f02b403c39a7d1423e3..8d7302c93c15e4b8ab88610b5318ac797662fb93 100644 (file)
@@ -1079,7 +1079,7 @@ nir_assign_io_var_locations(struct exec_list *var_list, unsigned *size,
          if (last_slot_location > location) {
             unsigned num_unallocated_slots = last_slot_location - location;
             unsigned first_unallocated_slot = var_size - num_unallocated_slots;
-            for (unsigned i = first_unallocated_slot; i < num_unallocated_slots; i++) {
+            for (unsigned i = first_unallocated_slot; i < var_size; i++) {
                assigned_locations[var->data.location + i] = location;
                location++;
             }