aco: be more careful combining additions that could wrap into loads/stores
[mesa.git] / src / amd / compiler / aco_spill.cpp
index c00f6466a7dedd39a1edf279828cd2c9412aa79e..d472f63ec57b543a05419b2bbedc048ca8c7e30b 100644 (file)
@@ -1566,9 +1566,9 @@ void assign_spill_slots(spill_ctx& ctx, unsigned spills_to_vgpr) {
                      split->definitions[i] = bld.def(v1);
                   bld.insert(split);
                   for (unsigned i = 0; i < temp.size(); i++)
-                     bld.mubuf(opcode, scratch_rsrc, Operand(v1), scratch_offset, split->definitions[i].getTemp(), offset + i * 4, false);
+                     bld.mubuf(opcode, scratch_rsrc, Operand(v1), scratch_offset, split->definitions[i].getTemp(), offset + i * 4, false, true);
                } else {
-                  bld.mubuf(opcode, scratch_rsrc, Operand(v1), scratch_offset, temp, offset, false);
+                  bld.mubuf(opcode, scratch_rsrc, Operand(v1), scratch_offset, temp, offset, false, true);
                }
             } else {
                ctx.program->config->spilled_sgprs += (*it)->operands[0].size();
@@ -1632,11 +1632,11 @@ void assign_spill_slots(spill_ctx& ctx, unsigned spills_to_vgpr) {
                   for (unsigned i = 0; i < def.size(); i++) {
                      Temp tmp = bld.tmp(v1);
                      vec->operands[i] = Operand(tmp);
-                     bld.mubuf(opcode, Definition(tmp), scratch_rsrc, Operand(v1), scratch_offset, offset + i * 4, false);
+                     bld.mubuf(opcode, Definition(tmp), scratch_rsrc, Operand(v1), scratch_offset, offset + i * 4, false, true);
                   }
                   bld.insert(vec);
                } else {
-                  bld.mubuf(opcode, def, scratch_rsrc, Operand(v1), scratch_offset, offset, false);
+                  bld.mubuf(opcode, def, scratch_rsrc, Operand(v1), scratch_offset, offset, false, true);
                }
             } else {
                uint32_t spill_slot = slots[spill_id];