Index: llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
===================================================================
--- llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
+++ llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
@@ -276,8 +276,6 @@
   addPass(createNVPTXLowerArgsPass(&getNVPTXTargetMachine()));
   if (getOptLevel() != CodeGenOpt::None) {
     addAddressSpaceInferencePasses();
-    if (!DisableLoadStoreVectorizer)
-      addPass(createLoadStoreVectorizerPass());
     addStraightLineScalarOptimizationPasses();
   }
 
@@ -295,8 +293,11 @@
   //   %1 = shl %a, 2
   //
   // but EarlyCSE can do neither of them.
-  if (getOptLevel() != CodeGenOpt::None)
+  if (getOptLevel() != CodeGenOpt::None) {
     addEarlyCSEOrGVNPass();
+    if (!DisableLoadStoreVectorizer)
+      addPass(createLoadStoreVectorizerPass());
+  }
 }
 
 bool NVPTXPassConfig::addInstSelector() {
Index: llvm/test/CodeGen/NVPTX/vector-loads.ll
===================================================================
--- llvm/test/CodeGen/NVPTX/vector-loads.ll
+++ llvm/test/CodeGen/NVPTX/vector-loads.ll
@@ -64,3 +64,39 @@
   store <8 x i32> %t2, <8 x i32>* %a
   ret void
 }
+
+; The following test wasn't passing previously as the address
+; computation was still too complex when LSV was called.
+declare i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #0
+declare i32 @llvm.nvvm.read.ptx.sreg.tid.x() #0
+define void @foo_complex(i8* nocapture readonly align 16 dereferenceable(134217728) %alloc0) {
+; CHECK: .func foo_complex
+; CHECK: ld.v2.u8
+; CHECK-NOT: ld
+; CHECK: ret
+  %targ0.1.typed = bitcast i8* %alloc0 to [1024 x [131072 x i8]]*
+  %t0 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !range !1
+  %t1 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x()
+  %t2 = lshr i32 %t1, 8
+  %t3 = shl nuw nsw i32 %t1, 9
+  %ttile_origin.2 = and i32 %t3, 130560
+  %tstart_offset_x_mul = shl nuw nsw i32 %t0, 1
+  %t4 = or i32 %ttile_origin.2, %tstart_offset_x_mul
+  %t6 = or i32 %t4, 1
+  %t8 = or i32 %t4, 128
+  %t9 = zext i32 %t8 to i64
+  %t10 = or i32 %t4, 129
+  %t11 = zext i32 %t10 to i64
+  %t20 = zext i32 %t2 to i64
+  %t27 = getelementptr inbounds [1024 x [131072 x i8]], [1024 x [131072 x i8]]* %targ0.1.typed, i64 0, i64 %t20, i64 %t9
+  %t28 = load i8, i8* %t27, align 2
+  %t31 = getelementptr inbounds [1024 x [131072 x i8]], [1024 x [131072 x i8]]* %targ0.1.typed, i64 0, i64 %t20, i64 %t11
+  %t32 = load i8, i8* %t31, align 1
+  %t33 = icmp ult i8 %t28, %t32
+  %t34 = select i1 %t33, i8 %t32, i8 %t28
+  store i8 %t34, i8* %t31
+  ret void
+}
+
+
+!1 = !{i32 0, i32 64}