diff --git a/llvm/lib/CodeGen/MachineScheduler.cpp b/llvm/lib/CodeGen/MachineScheduler.cpp
--- a/llvm/lib/CodeGen/MachineScheduler.cpp
+++ b/llvm/lib/CodeGen/MachineScheduler.cpp
@@ -1590,16 +1590,32 @@
       if (DAG->addEdge(SUb, SDep(SUa, SDep::Cluster))) {
         LLVM_DEBUG(dbgs() << "Cluster ld/st SU(" << SUa->NodeNum << ") - SU("
                           << SUb->NodeNum << ")\n");
-        // Copy successor edges from SUa to SUb. Interleaving computation
-        // dependent on SUa can prevent load combining due to register reuse.
-        // Predecessor edges do not need to be copied from SUb to SUa since
-        // nearby loads should have effectively the same inputs.
-        for (const SDep &Succ : SUa->Succs) {
-          if (Succ.getSUnit() == SUb)
-            continue;
-          LLVM_DEBUG(dbgs()
-                     << "  Copy Succ SU(" << Succ.getSUnit()->NodeNum << ")\n");
-          DAG->addEdge(Succ.getSUnit(), SDep(SUb, SDep::Artificial));
+        if (IsLoad) {
+          // Copy successor edges from SUa to SUb. Interleaving computation
+          // dependent on SUa can prevent load combining due to register reuse.
+          // Predecessor edges do not need to be copied from SUb to SUa since
+          // nearby loads should have effectively the same inputs.
+          for (const SDep &Succ : SUa->Succs) {
+            if (Succ.getSUnit() == SUb)
+              continue;
+            LLVM_DEBUG(dbgs() << "  Copy Succ SU(" << Succ.getSUnit()->NodeNum
+                              << ")\n");
+            DAG->addEdge(Succ.getSUnit(), SDep(SUb, SDep::Artificial));
+          }
+        } else {
+          // Copy predecessor edges from SUb to SUa to avoid the SUnits that
+          // SUb dependent on scheduled in-between SUb and SUa. Successor edges
+          // do not need to be copied from SUa to SUb since no one will depend
+          // on stores.
+          // Notice that, we don't need to care about the memory dependency as
+          // we won't try to cluster them if they have any memory dependency.
+          for (const SDep &Pred : SUb->Preds) {
+            if (Pred.getSUnit() == SUa)
+              continue;
+            LLVM_DEBUG(dbgs() << "  Copy Pred SU(" << Pred.getSUnit()->NodeNum
+                              << ")\n");
+            DAG->addEdge(SUa, SDep(Pred.getSUnit(), SDep::Artificial));
+          }
         }
         ++ClusterLength;
       } else
diff --git a/llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll b/llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll
--- a/llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll
+++ b/llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll
@@ -365,26 +365,26 @@
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[0:3], 0 addr64 offset:5
-; SI-NEXT:    buffer_load_ubyte v3, v[0:1], s[0:3], 0 addr64 offset:6
-; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[0:3], 0 addr64
-; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[0:3], 0 addr64 offset:1
-; SI-NEXT:    buffer_load_ubyte v6, v[0:1], s[0:3], 0 addr64 offset:2
-; SI-NEXT:    buffer_load_ubyte v7, v[0:1], s[0:3], 0 addr64 offset:3
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[0:3], 0 addr64 offset:4
+; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[0:3], 0 addr64
+; SI-NEXT:    buffer_load_ubyte v3, v[0:1], s[0:3], 0 addr64 offset:1
+; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[0:3], 0 addr64 offset:2
+; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[0:3], 0 addr64 offset:3
+; SI-NEXT:    buffer_load_ubyte v6, v[0:1], s[0:3], 0 addr64 offset:4
+; SI-NEXT:    buffer_load_ubyte v7, v[0:1], s[0:3], 0 addr64 offset:5
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[0:3], 0 addr64 offset:6
 ; SI-NEXT:    s_mov_b32 s6, -1
-; SI-NEXT:    s_waitcnt vmcnt(6)
-; SI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; SI-NEXT:    s_waitcnt vmcnt(5)
+; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v3
+; SI-NEXT:    v_or_b32_e32 v1, v1, v2
 ; SI-NEXT:    s_waitcnt vmcnt(3)
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v5
-; SI-NEXT:    v_or_b32_e32 v1, v1, v4
+; SI-NEXT:    v_lshlrev_b32_e32 v2, 8, v5
+; SI-NEXT:    v_or_b32_e32 v2, v2, v4
 ; SI-NEXT:    s_waitcnt vmcnt(1)
-; SI-NEXT:    v_lshlrev_b32_e32 v5, 8, v7
-; SI-NEXT:    v_cvt_f32_ubyte0_e32 v7, v3
-; SI-NEXT:    v_or_b32_e32 v3, v5, v6
+; SI-NEXT:    v_lshlrev_b32_e32 v3, 8, v7
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_or_b32_e32 v0, v2, v0
-; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v3
+; SI-NEXT:    v_cvt_f32_ubyte0_e32 v7, v0
+; SI-NEXT:    v_or_b32_e32 v0, v3, v6
+; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v5, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v0
 ; SI-NEXT:    v_or_b32_e32 v0, v2, v1
diff --git a/llvm/test/CodeGen/AMDGPU/half.ll b/llvm/test/CodeGen/AMDGPU/half.ll
--- a/llvm/test/CodeGen/AMDGPU/half.ll
+++ b/llvm/test/CodeGen/AMDGPU/half.ll
@@ -307,6 +307,7 @@
 ; GCN: flat_load_dwordx4
 ; GCN: flat_load_dwordx4
 
+; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
@@ -315,7 +316,6 @@
 
 ; GCN: flat_store_dwordx4
 
-; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
diff --git a/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll b/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
--- a/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
+++ b/llvm/test/CodeGen/AMDGPU/salu-to-valu.ll
@@ -173,9 +173,9 @@
 ; GCN-NOHSA-DAG: s_mov_b32 [[OFFSET0:s[0-9]+]], 0x9a40{{$}}
 ; CI-NOHSA-DAG: s_mov_b32 [[OFFSET1:s[0-9]+]], 0x9a50{{$}}
 ; CI-NOHSA-NOT: v_add
-; SI: buffer_load_dwordx4 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+:[0-9]+}}], 0 addr64 offset:16
 ; CI-NOHSA: buffer_load_dwordx4 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+:[0-9]+}}], [[OFFSET1]] addr64{{$}}
 ; GCN-NOHSA: buffer_load_dwordx4 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+:[0-9]+}}], [[OFFSET0]] addr64{{$}}
+; SI: buffer_load_dwordx4 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, s[{{[0-9]+:[0-9]+}}], 0 addr64 offset:16
 
 ; GCN-NOHSA: v_or_b32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
 ; GCN-NOHSA: v_or_b32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
diff --git a/llvm/test/CodeGen/AMDGPU/widen-smrd-loads.ll b/llvm/test/CodeGen/AMDGPU/widen-smrd-loads.ll
--- a/llvm/test/CodeGen/AMDGPU/widen-smrd-loads.ll
+++ b/llvm/test/CodeGen/AMDGPU/widen-smrd-loads.ll
@@ -135,10 +135,9 @@
 ; SI-NEXT:    v_mov_b32_e32 v0, s0
 ; SI-NEXT:    s_bfe_u32 s0, s0, 0x10010
 ; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
+; SI-NEXT:    v_mov_b32_e32 v1, s0
 ; SI-NEXT:    s_mov_b32 s4, 2
-; SI-NEXT:    s_waitcnt expcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s0
-; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_byte v1, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: widen_i17_constant_load: