diff --git a/lib/Target/R600/SILowerControlFlow.cpp b/lib/Target/R600/SILowerControlFlow.cpp
index c2e8f02..09cf25b 100644
--- a/lib/Target/R600/SILowerControlFlow.cpp
+++ b/lib/Target/R600/SILowerControlFlow.cpp
@@ -409,6 +409,7 @@
 bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
   TII = MF.getTarget().getInstrInfo();
   TRI = MF.getTarget().getRegisterInfo();
+  SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
 
   bool HaveKill = false;
   bool NeedM0 = false;
@@ -508,7 +509,7 @@
             AMDGPU::M0).addImm(0xffffffff);
   }
 
-  if (NeedWQM) {
+  if (NeedWQM && MFI->ShaderType != ShaderType::COMPUTE) {
     MachineBasicBlock &MBB = MF.front();
     BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
             AMDGPU::EXEC).addReg(AMDGPU::EXEC);
diff --git a/test/CodeGen/R600/load.ll b/test/CodeGen/R600/load.ll
index c7fe611..8829ff5 100644
--- a/test/CodeGen/R600/load.ll
+++ b/test/CodeGen/R600/load.ll
@@ -446,6 +446,7 @@
 ; R600-CHECK: @load_i8_local
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; SI-CHECK: @load_i8_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U8
 define void @load_i8_local(i32 addrspace(1)* %out, i8 addrspace(3)* %in) {
   %1 = load i8 addrspace(3)* %in
@@ -458,6 +459,7 @@
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; R600-CHECK: ASHR
 ; SI-CHECK: @load_i8_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I8
 define void @load_i8_sext_local(i32 addrspace(1)* %out, i8 addrspace(3)* %in) {
 entry:
@@ -471,6 +473,7 @@
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; SI-CHECK: @load_v2i8_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U8
 ; SI-CHECK: DS_READ_U8
 define void @load_v2i8_local(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(3)* %in) {
@@ -487,6 +490,7 @@
 ; R600-CHECK-DAG: ASHR
 ; R600-CHECK-DAG: ASHR
 ; SI-CHECK: @load_v2i8_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I8
 ; SI-CHECK: DS_READ_I8
 define void @load_v2i8_sext_local(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(3)* %in) {
@@ -503,6 +507,7 @@
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; R600-CHECK: LDS_UBYTE_READ_RET
 ; SI-CHECK: @load_v4i8_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U8
 ; SI-CHECK: DS_READ_U8
 ; SI-CHECK: DS_READ_U8
@@ -525,6 +530,7 @@
 ; R600-CHECK-DAG: ASHR
 ; R600-CHECK-DAG: ASHR
 ; SI-CHECK: @load_v4i8_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I8
 ; SI-CHECK: DS_READ_I8
 ; SI-CHECK: DS_READ_I8
@@ -541,6 +547,7 @@
 ; R600-CHECK: @load_i16_local
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; SI-CHECK: @load_i16_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U16
 define void @load_i16_local(i32 addrspace(1)* %out, i16 addrspace(3)* %in) {
 entry:
@@ -554,6 +561,7 @@
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; R600-CHECK: ASHR
 ; SI-CHECK: @load_i16_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I16
 define void @load_i16_sext_local(i32 addrspace(1)* %out, i16 addrspace(3)* %in) {
 entry:
@@ -567,6 +575,7 @@
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; SI-CHECK: @load_v2i16_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U16
 ; SI-CHECK: DS_READ_U16
 define void @load_v2i16_local(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(3)* %in) {
@@ -583,6 +592,7 @@
 ; R600-CHECK-DAG: ASHR
 ; R600-CHECK-DAG: ASHR
 ; SI-CHECK: @load_v2i16_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I16
 ; SI-CHECK: DS_READ_I16
 define void @load_v2i16_sext_local(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(3)* %in) {
@@ -599,6 +609,7 @@
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; R600-CHECK: LDS_USHORT_READ_RET
 ; SI-CHECK: @load_v4i16_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_U16
 ; SI-CHECK: DS_READ_U16
 ; SI-CHECK: DS_READ_U16
@@ -621,6 +632,7 @@
 ; R600-CHECK-DAG: ASHR
 ; R600-CHECK-DAG: ASHR
 ; SI-CHECK: @load_v4i16_sext_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_I16
 ; SI-CHECK: DS_READ_I16
 ; SI-CHECK: DS_READ_I16
@@ -637,6 +649,7 @@
 ; R600-CHECK: @load_i32_local
 ; R600-CHECK: LDS_READ_RET
 ; SI-CHECK: @load_i32_local
+; SI-CHECK-NOT: S_WQM_B64
 ; SI-CHECK: DS_READ_B32
 define void @load_i32_local(i32 addrspace(1)* %out, i32 addrspace(3)* %in) {
 entry:
diff --git a/test/CodeGen/R600/local-memory.ll b/test/CodeGen/R600/local-memory.ll
index 9ebb769..bd0d59c 100644
--- a/test/CodeGen/R600/local-memory.ll
+++ b/test/CodeGen/R600/local-memory.ll
@@ -13,6 +13,7 @@
 ; SI-CHECK-NEXT: .long 32768
 
 ; EG-CHECK: LDS_WRITE
+; SI-CHECK_NOT: S_WQM_B64
 ; SI-CHECK: DS_WRITE_B32 0
 
 ; GROUP_BARRIER must be the last instruction in a clause