Index: lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -253,6 +253,10 @@
   for (unsigned i = 1, e = Offsets.size(); i != e; ++i) {
     int64_t Offset = Offsets[i];
     SDNode *Load = O2SMap[Offset];
+    // If BASELOAD depends on LOAD, we could not schedule BASELOAD ahead of LOAD.
+    // We continue to search for other candidates.
+    if (Load->isOperandOf(BaseLoad))
+      continue;
     if (!TII->shouldScheduleLoadsNear(BaseLoad, Load, BaseOff, Offset,NumLoads))
       break; // Stop right here. Ignore loads that are further away.
     Loads.push_back(Load);
Index: test/CodeGen/AMDGPU/cluster-neighboring-loads.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/cluster-neighboring-loads.ll
@@ -0,0 +1,17 @@
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
+
+; GCN-LABEL: {{^}}cluster_neighboring_loads:
+; GCN: buffer_load_ushort [[DST:v[0-9]+]], off, [[RSRC:s\[[0-9]+:[0-9]+\]]], [[SOFF:s[0-9]+]] offset:2
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_load_short_d16_hi [[DST]], off, [[RSRC]], [[SOFF]]
+
+define amdgpu_kernel void @cluster_neighboring_loads() {
+bb:
+  %loads = load <2 x half>, <2 x half> addrspace(5)* null, align 2
+  %shuffled = shufflevector <2 x half> %loads, <2 x half> undef, <2 x i32> <i32 1, i32 0>
+  br label %bb1
+
+bb1:
+  call void asm sideeffect "; use $0", "v"(<2 x half> %shuffled)
+  br label %bb1
+}