Index: lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- lib/Target/X86/X86ISelDAGToDAG.cpp
+++ lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -3073,30 +3073,6 @@
         return;
       }
 
-      // For example, "testl %eax, $2048" to "testb %ah, $8".
-      if (isShiftedUInt<8, 8>(Mask) &&
-          (!(Mask & 0x8000) || hasNoSignedComparisonUses(Node))) {
-        // Shift the immediate right by 8 bits.
-        SDValue ShiftedImm = CurDAG->getTargetConstant(Mask >> 8, dl, MVT::i8);
-        SDValue Reg = N0.getOperand(0);
-
-        // Extract the h-register.
-        SDValue Subreg = CurDAG->getTargetExtractSubreg(X86::sub_8bit_hi, dl,
-                                                        MVT::i8, Reg);
-
-        // Emit a testb.  The EXTRACT_SUBREG becomes a COPY that can only
-        // target GR8_NOREX registers, so make sure the register class is
-        // forced.
-        SDNode *NewNode = CurDAG->getMachineNode(X86::TEST8ri_NOREX, dl,
-                                                 MVT::i32, Subreg, ShiftedImm);
-        // Replace SUB|CMP with TEST, since SUB has two outputs while TEST has
-        // one, do not call ReplaceAllUsesWith.
-        ReplaceUses(SDValue(Node, (Opcode == X86ISD::SUB ? 1 : 0)),
-                    SDValue(NewNode, 0));
-        CurDAG->RemoveDeadNode(Node);
-        return;
-      }
-
       // For example, "testl %eax, $32776" to "testw %ax, $32776".
       // NOTE: We only want to form TESTW instructions if optimizing for
       // min size. Otherwise we only save one byte and possibly get a length
@@ -3122,18 +3098,21 @@
       }
 
       // For example, "testq %rax, $268468232" to "testl %eax, $268468232".
-      if (isUInt<32>(Mask) && N0.getValueType() == MVT::i64 &&
+      if (isUInt<32>(Mask) &&
+          (N0.getValueType() == MVT::i32 || N0.getValueType() == MVT::i64) &&
           (!(Mask & 0x80000000) || hasNoSignedComparisonUses(Node))) {
         SDValue Imm = CurDAG->getTargetConstant(Mask, dl, MVT::i32);
         SDValue Reg = N0.getOperand(0);
 
         // Extract the 32-bit subregister.
-        SDValue Subreg = CurDAG->getTargetExtractSubreg(X86::sub_32bit, dl,
-                                                        MVT::i32, Reg);
+        if (N0.getValueType() == MVT::i64) {
+          Reg = CurDAG->getTargetExtractSubreg(X86::sub_32bit, dl, MVT::i32,
+                                               Reg);
+        }
 
         // Emit a testl.
         SDNode *NewNode = CurDAG->getMachineNode(X86::TEST32ri, dl, MVT::i32,
-                                                 Subreg, Imm);
+                                                 Reg, Imm);
         // Replace SUB|CMP with TEST, since SUB has two outputs while TEST has
         // one, do not call ReplaceAllUsesWith.
         ReplaceUses(SDValue(Node, (Opcode == X86ISD::SUB ? 1 : 0)),
Index: test/CodeGen/X86/test-shrink.ll
===================================================================
--- test/CodeGen/X86/test-shrink.ll
+++ test/CodeGen/X86/test-shrink.ll
@@ -484,8 +484,7 @@
 define void @truncand32(i16 inreg %x) nounwind {
 ; CHECK-LINUX64-LABEL: truncand32:
 ; CHECK-LINUX64:       # %bb.0:
-; CHECK-LINUX64-NEXT:    andl $2049, %edi # imm = 0x801
-; CHECK-LINUX64-NEXT:    testw %di, %di
+; CHECK-LINUX64-NEXT:    testl $2049, %edi # imm = 0x801
 ; CHECK-LINUX64-NEXT:    je .LBB11_1
 ; CHECK-LINUX64-NEXT:  # %bb.2: # %no
 ; CHECK-LINUX64-NEXT:    retq
@@ -498,8 +497,7 @@
 ; CHECK-WIN32-64-LABEL: truncand32:
 ; CHECK-WIN32-64:       # %bb.0:
 ; CHECK-WIN32-64-NEXT:    subq $40, %rsp
-; CHECK-WIN32-64-NEXT:    andl $2049, %ecx # imm = 0x801
-; CHECK-WIN32-64-NEXT:    testw %cx, %cx
+; CHECK-WIN32-64-NEXT:    testl $2049, %ecx # imm = 0x801
 ; CHECK-WIN32-64-NEXT:    je .LBB11_1
 ; CHECK-WIN32-64-NEXT:  # %bb.2: # %no
 ; CHECK-WIN32-64-NEXT:    addq $40, %rsp
@@ -511,8 +509,7 @@
 ;
 ; CHECK-X86-LABEL: truncand32:
 ; CHECK-X86:       # %bb.0:
-; CHECK-X86-NEXT:    andl $2049, %eax # imm = 0x801
-; CHECK-X86-NEXT:    testw %ax, %ax
+; CHECK-X86-NEXT:    testl $2049, %eax # imm = 0x801
 ; CHECK-X86-NEXT:    je .LBB11_1
 ; CHECK-X86-NEXT:  # %bb.2: # %no
 ; CHECK-X86-NEXT:    retl
Index: test/CodeGen/X86/testb-je-fusion.ll
===================================================================
--- test/CodeGen/X86/testb-je-fusion.ll
+++ test/CodeGen/X86/testb-je-fusion.ll
@@ -6,9 +6,8 @@
 define i32 @check_flag(i32 %flags, ...) nounwind {
 ; CHECK-LABEL: check_flag:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    testb $2, %ch
+; CHECK-NEXT:    testl $512, %edi # imm = 0x200
 ; CHECK-NEXT:    je .LBB0_2
 ; CHECK-NEXT:  # %bb.1: # %if.then
 ; CHECK-NEXT:    movl $1, %eax
Index: test/CodeGen/X86/vastart-defs-eflags.ll
===================================================================
--- test/CodeGen/X86/vastart-defs-eflags.ll
+++ test/CodeGen/X86/vastart-defs-eflags.ll
@@ -8,9 +8,7 @@
 define i32 @check_flag(i32 %flags, ...) nounwind {
 ; CHECK-LABEL: check_flag:
 ; CHECK:       ## %bb.0: ## %entry
-; CHECK-NEXT:    pushq %rbx
-; CHECK-NEXT:    subq $48, %rsp
-; CHECK-NEXT:    movl %edi, %ebx
+; CHECK-NEXT:    subq $56, %rsp
 ; CHECK-NEXT:    testb %al, %al
 ; CHECK-NEXT:    je LBB0_2
 ; CHECK-NEXT:  ## %bb.1: ## %entry
@@ -29,7 +27,7 @@
 ; CHECK-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    testb $2, %bh
+; CHECK-NEXT:    testl $512, %edi ## imm = 0x200
 ; CHECK-NEXT:    je LBB0_4
 ; CHECK-NEXT:  ## %bb.3: ## %if.then
 ; CHECK-NEXT:    leaq -{{[0-9]+}}(%rsp), %rax
@@ -40,8 +38,7 @@
 ; CHECK-NEXT:    movl $8, 0
 ; CHECK-NEXT:    movl $1, %eax
 ; CHECK-NEXT:  LBB0_4: ## %if.end
-; CHECK-NEXT:    addq $48, %rsp
-; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    addq $56, %rsp
 ; CHECK-NEXT:    retq
 entry:
   %and = and i32 %flags, 512