diff --git a/src/amd/compiler/aco_optimizer.cpp b/src/amd/compiler/aco_optimizer.cpp
index af7dc3a3f68..8e4daaa8167 100644
--- a/src/amd/compiler/aco_optimizer.cpp
+++ b/src/amd/compiler/aco_optimizer.cpp
@@ -1642,6 +1642,13 @@ label_instruction(opt_ctx& ctx, aco_ptr<Instruction>& instr)
             ctx.info[instr->definitions[0].tempId()].set_uniform_bool(
                ctx.info[instr->operands[0].tempId()].instr->definitions[1].getTemp());
             break;
+         } else if ((ctx.program->stage.num_sw_stages() > 1 ||
+                     ctx.program->stage.hw == HWStage::NGG) &&
+                    instr->pass_flags == 1) {
+            /* In case of merged shaders, pass_flags=1 means that all lanes are active (exec=-1), so
+             * s_and is unnecessary. */
+            ctx.info[instr->definitions[0].tempId()].set_temp(instr->operands[0].getTemp());
+            break;
          } else if (ctx.info[instr->operands[0].tempId()].is_vopc()) {
             Instruction* vopc_instr = ctx.info[instr->operands[0].tempId()].instr;
             /* Remove superfluous s_and when the VOPC instruction uses the same exec and thus