diff --git a/src/amd/compiler/aco_insert_waitcnt.cpp b/src/amd/compiler/aco_insert_waitcnt.cpp
index a8bba344c6d..396373e23d8 100644
--- a/src/amd/compiler/aco_insert_waitcnt.cpp
+++ b/src/amd/compiler/aco_insert_waitcnt.cpp
@@ -1049,6 +1049,11 @@ insert_wait_states(Program* program)
    std::stack<unsigned, std::vector<unsigned>> loop_header_indices;
    unsigned loop_progress = 0;
 
+   if (program->pending_lds_access) {
+      update_barrier_imm(in_ctx[0], get_counters_for_event(event_lds), event_lds,
+                         memory_sync_info(storage_shared));
+   }
+
    for (Definition def : program->args_pending_vmem) {
       update_counters(in_ctx[0], event_vmem);
       insert_wait_entry(in_ctx[0], def, event_vmem);
diff --git a/src/amd/compiler/aco_ir.h b/src/amd/compiler/aco_ir.h
index 716d019432f..09b1da18503 100644
--- a/src/amd/compiler/aco_ir.h
+++ b/src/amd/compiler/aco_ir.h
@@ -2145,6 +2145,9 @@ public:
 
    std::vector<Definition> args_pending_vmem;
 
+   /* For shader part with previous shader part that has lds access. */
+   bool pending_lds_access = false;
+
    struct {
       FILE* output = stderr;
       bool shorten_messages = false;