Index: llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -3412,7 +3412,7 @@
 
       MVT VT;
       int SubRegOp;
-      unsigned Op;
+      unsigned ROpc, MOpc;
 
       // For each of these checks we need to be careful if the sign flag is
       // being used. It is only safe to use the sign flag in two conditions,
@@ -3425,7 +3425,8 @@
         // For example, convert "testl %eax, $8" to "testb %al, $8"
         VT = MVT::i8;
         SubRegOp = X86::sub_8bit;
-        Op = X86::TEST8ri;
+        ROpc = X86::TEST8ri;
+        MOpc = X86::TEST8mi;
       } else if (OptForMinSize && isUInt<16>(Mask) &&
                  (!(Mask & 0x8000) || CmpVT == MVT::i16 ||
                   hasNoSignedComparisonUses(Node))) {
@@ -3435,7 +3436,8 @@
         // changing prefix penalty in the decoders.
         VT = MVT::i16;
         SubRegOp = X86::sub_16bit;
-        Op = X86::TEST16ri;
+        ROpc = X86::TEST16ri;
+        MOpc = X86::TEST16mi;
       } else if (isUInt<32>(Mask) && N0.getValueType() != MVT::i16 &&
                  (!(Mask & 0x80000000) || CmpVT == MVT::i32 ||
                   hasNoSignedComparisonUses(Node))) {
@@ -3446,7 +3448,8 @@
         // they had a good reason not to and do not promote here.
         VT = MVT::i32;
         SubRegOp = X86::sub_32bit;
-        Op = X86::TEST32ri;
+        ROpc = X86::TEST32ri;
+        MOpc = X86::TEST32mi;
       } else {
         // No eligible transformation was found.
         break;
@@ -3457,12 +3460,25 @@
       SDValue Imm = CurDAG->getTargetConstant(Mask, dl, VT);
       SDValue Reg = N0.getOperand(0);
 
-      // Extract the subregister if necessary.
-      if (N0.getValueType() != VT)
-        Reg = CurDAG->getTargetExtractSubreg(SubRegOp, dl, VT, Reg);
-
       // Emit a testl or testw.
-      SDNode *NewNode = CurDAG->getMachineNode(Op, dl, MVT::i32, Reg, Imm);
+      MachineSDNode *NewNode;
+      SDValue Tmp0, Tmp1, Tmp2, Tmp3, Tmp4;
+      if (tryFoldLoad(Node, N0.getNode(), Reg, Tmp0, Tmp1, Tmp2, Tmp3, Tmp4)) {
+        SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Imm,
+                          Reg.getOperand(0) };
+        NewNode = CurDAG->getMachineNode(MOpc, dl, MVT::i32, MVT::Other, Ops);
+        // Update the chain.
+        ReplaceUses(Reg.getValue(1), SDValue(NewNode, 1));
+        // Record the mem-refs
+        CurDAG->setNodeMemRefs(NewNode,
+                               {cast<LoadSDNode>(Reg)->getMemOperand()});
+      } else {
+        // Extract the subregister if necessary.
+        if (N0.getValueType() != VT)
+          Reg = CurDAG->getTargetExtractSubreg(SubRegOp, dl, VT, Reg);
+
+        NewNode = CurDAG->getMachineNode(ROpc, dl, MVT::i32, Reg, Imm);
+      }
       // Replace CMP with TEST.
       ReplaceNode(Node, NewNode);
       return;
Index: llvm/trunk/test/CodeGen/X86/test-shrink.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/test-shrink.ll
+++ llvm/trunk/test/CodeGen/X86/test-shrink.ll
@@ -645,8 +645,7 @@
 ;
 ; CHECK-X86-LABEL: and32_trunc_8_sign:
 ; CHECK-X86:       # %bb.0:
-; CHECK-X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-X86-NEXT:    testb $-128, %al
+; CHECK-X86-NEXT:    testb $-128, {{[0-9]+}}(%esp)
 ; CHECK-X86-NEXT:    jg .LBB14_2
 ; CHECK-X86-NEXT:  # %bb.1: # %yes
 ; CHECK-X86-NEXT:    calll bar
@@ -689,8 +688,7 @@
 ;
 ; CHECK-X86-LABEL: and64_trunc_8_sign:
 ; CHECK-X86:       # %bb.0:
-; CHECK-X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-X86-NEXT:    testb $-128, %al
+; CHECK-X86-NEXT:    testb $-128, {{[0-9]+}}(%esp)
 ; CHECK-X86-NEXT:    jg .LBB15_2
 ; CHECK-X86-NEXT:  # %bb.1: # %yes
 ; CHECK-X86-NEXT:    calll bar
@@ -733,8 +731,7 @@
 ;
 ; CHECK-X86-LABEL: and32_trunc_16_sign:
 ; CHECK-X86:       # %bb.0:
-; CHECK-X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-X86-NEXT:    testw $-32768, %ax # imm = 0x8000
+; CHECK-X86-NEXT:    testw $-32768, {{[0-9]+}}(%esp) # imm = 0x8000
 ; CHECK-X86-NEXT:    jg .LBB16_2
 ; CHECK-X86-NEXT:  # %bb.1: # %yes
 ; CHECK-X86-NEXT:    calll bar
@@ -777,8 +774,7 @@
 ;
 ; CHECK-X86-LABEL: and64_trunc_32_sign:
 ; CHECK-X86:       # %bb.0:
-; CHECK-X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-X86-NEXT:    testw $-32768, %ax # imm = 0x8000
+; CHECK-X86-NEXT:    testw $-32768, {{[0-9]+}}(%esp) # imm = 0x8000
 ; CHECK-X86-NEXT:    jg .LBB17_2
 ; CHECK-X86-NEXT:  # %bb.1: # %yes
 ; CHECK-X86-NEXT:    calll bar