Index: lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- lib/Target/X86/X86ISelDAGToDAG.cpp
+++ lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -457,7 +457,8 @@
     }
 
     bool foldLoadStoreIntoMemOperand(SDNode *Node);
-    bool matchBEXTRFromAnd(SDNode *Node);
+    bool matchBEXTRWithImmFromAnd(SDNode *Node);
+    bool matchBEXTR(SDNode *Node);
     bool shrinkAndImmediate(SDNode *N);
     bool isMaskZeroExtended(SDNode *N) const;
     bool tryShiftAmountMod(SDNode *N);
@@ -2583,7 +2584,7 @@
 }
 
 // See if this is an (X >> C1) & C2 that we can match to BEXTR/BEXTRI.
-bool X86DAGToDAGISel::matchBEXTRFromAnd(SDNode *Node) {
+bool X86DAGToDAGISel::matchBEXTRWithImmFromAnd(SDNode *Node) {
   MVT NVT = Node->getSimpleValueType(0);
   SDLoc dl(Node);
 
@@ -2638,6 +2639,87 @@
   return true;
 }
 
+// See if this is an  X & Mask  that we can match to BEXTR.
+// Where Mask is one of the following patterns:
+//   a) x &  (1 << nbits) - 1
+//   b) x & ~(-1 << nbits)
+//   c) x &  (-1 >> (32 - y))
+//   d) x << (32 - y) >> (32 - y)
+bool X86DAGToDAGISel::matchBEXTR(SDNode *Node) {
+  // BEXTR is BMI instruction. However, if we have BMI2, we prefer BZHI.
+  if (!Subtarget->hasBMI() || Subtarget->hasBMI2())
+    return false;
+
+  MVT NVT = Node->getSimpleValueType(0);
+
+  // Only supported for 32 and 64 bits.
+  if (NVT != MVT::i32 && NVT != MVT::i64)
+    return false;
+
+  SDValue NBits;
+
+  // b) x & ~(-1 << nbits)
+  auto matchPatternB = [&NBits](SDValue Mask) -> bool {
+    // Match `~()`. Must only have one use!
+    if (!isBitwiseNot(Mask) || !Mask->hasOneUse())
+      return false;
+    // Match `-1 << nbits`. Must only have one use!
+    SDValue M0 = Mask->getOperand(0);
+    if (M0->getOpcode() != ISD::SHL || !M0->hasOneUse())
+      return false;
+    ConstantSDNode *ShlCst = dyn_cast<ConstantSDNode>(M0->getOperand(0));
+    if (!ShlCst || !ShlCst->isAllOnesValue())
+      return false;
+    NBits = M0->getOperand(1);
+    return true;
+  };
+
+  auto matchLowBitMask = [&matchPatternB](SDValue Mask) -> bool {
+    // FIXME: patterns a, c, d.
+    return matchPatternB(Mask);
+  };
+
+  SDValue X = Node->getOperand(0);
+  SDValue Mask = Node->getOperand(1);
+
+  if (matchLowBitMask(Mask)) {
+    // Great.
+  } else {
+    std::swap(X, Mask);
+    if (!matchLowBitMask(Mask))
+      return false;
+  }
+
+  SDLoc DL(Node);
+
+  // Make NBits to be the same type as the Node.
+  if (NBits->getSimpleValueType(0) != NVT) {
+    SDValue NBitsTrunc = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, NBits);
+    insertDAGNode(*CurDAG, NBits, NBitsTrunc);
+    // We do not care about the new high bits above the exisitng i8.
+    NBits = CurDAG->getNode(ISD::ANY_EXTEND, DL, NVT, NBitsTrunc);
+    insertDAGNode(*CurDAG, NBitsTrunc, NBits);
+  }
+
+  // The 'control' of BEXTR has the pattern of:
+  // [    AH    ][    AX    ] sub-reg
+  // [15...8 bit][ 7...0 bit] location
+  // [ bit count][     shift] name
+  // I.e. 0b000000011'00000001 means  (x >> 0b1) & 0b11
+
+  // Produce the 'control' with NBits in AH subreg.
+  SDValue C8 = CurDAG->getConstant(8, DL, MVT::i8);
+  SDValue Control = CurDAG->getNode(ISD::SHL, DL, NVT, NBits, C8);
+  insertDAGNode(*CurDAG, SDValue(Node, 0), Control);
+
+  // And finally, the BEXTR itself.
+  SDValue Extract = CurDAG->getNode(X86ISD::BEXTR, DL, NVT, X, Control);
+  ReplaceNode(Node, Extract.getNode());
+  SelectCode(Extract.getNode());
+
+  return true;
+}
+
 // Emit a PCMISTR(I/M) instruction.
 MachineSDNode *X86DAGToDAGISel::emitPCMPISTR(unsigned ROpc, unsigned MOpc,
                                              bool MayFoldLoad, const SDLoc &dl,
@@ -2945,7 +3027,9 @@
     break;
 
   case ISD::AND:
-    if (matchBEXTRFromAnd(Node))
+    if (matchBEXTRWithImmFromAnd(Node))
+      return;
+    if (matchBEXTR(Node))
       return;
     if (AndImmShrink && shrinkAndImmediate(Node))
       return;
Index: test/CodeGen/X86/extract-bits.ll
===================================================================
--- test/CodeGen/X86/extract-bits.ll
+++ test/CodeGen/X86/extract-bits.ll
@@ -1507,16 +1507,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b0:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
-; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, %eax, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b0:
@@ -1544,10 +1540,8 @@
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b0:
@@ -1580,16 +1574,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
-; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, %eax, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b1_indexzext:
@@ -1617,10 +1607,8 @@
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b1_indexzext:
@@ -1656,17 +1644,13 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
-; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, %eax, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1697,10 +1681,8 @@
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X64-BMI1NOTBM-NEXT:    andnl %eax, %esi, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1735,17 +1717,13 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
-; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, %eax, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1776,10 +1754,8 @@
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X64-BMI1NOTBM-NEXT:    andnl %eax, %esi, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1815,16 +1791,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b4_commutative:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
-; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
-; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, %eax, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b4_commutative:
@@ -1852,10 +1824,8 @@
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b4_commutative:
@@ -1896,24 +1866,19 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    pushl %edi
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
-; X86-BMI1NOTBM-NEXT:    pushl %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
-; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
-; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
-; X86-BMI1NOTBM-NEXT:    andnl %esi, %edi, %esi
-; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
-; X86-BMI1NOTBM-NEXT:    addl $4, %esp
+; X86-BMI1NOTBM-NEXT:    addl $8, %esp
 ; X86-BMI1NOTBM-NEXT:    popl %esi
-; X86-BMI1NOTBM-NEXT:    popl %edi
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b5_skipextrauses:
@@ -1952,10 +1917,8 @@
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %ebx
+; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
 ; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
@@ -2106,10 +2069,8 @@
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b0:
@@ -2248,10 +2209,9 @@
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movzbl %dl, %eax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b1_indexzext:
@@ -2399,10 +2359,8 @@
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rsi
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rsi
-; X64-BMI1NOTBM-NEXT:    andnq %rax, %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b2_load:
@@ -2547,10 +2505,9 @@
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rsi
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rsi
-; X64-BMI1NOTBM-NEXT:    andnq %rax, %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    movzbl %dl, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rcx
+; X64-BMI1NOTBM-NEXT:    bextrq %rcx, %rax, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b3_load_indexzext:
@@ -2694,10 +2651,8 @@
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b4_commutative:
@@ -2876,12 +2831,10 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rbx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
 ; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
Index: test/CodeGen/X86/extract-lowbits.ll
===================================================================
--- test/CodeGen/X86/extract-lowbits.ll
+++ test/CodeGen/X86/extract-lowbits.ll
@@ -794,10 +794,9 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bzhi32_b0:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_b0:
@@ -818,11 +817,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_b0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %esi
+; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_b0:
@@ -847,10 +843,9 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bzhi32_b1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_b1_indexzext:
@@ -871,11 +866,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %esi
+; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_b1_indexzext:
@@ -903,10 +895,9 @@
 ; X86-BMI1NOTBM-LABEL: bzhi32_b2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %edx
-; X86-BMI1NOTBM-NEXT:    andnl (%eax), %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_b2_load:
@@ -928,11 +919,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl (%rdi), %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %esi
+; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_b2_load:
@@ -960,10 +948,9 @@
 ; X86-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
-; X86-BMI1NOTBM-NEXT:    shll %cl, %edx
-; X86-BMI1NOTBM-NEXT:    andnl (%eax), %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext:
@@ -985,11 +972,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl (%rdi), %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %esi
+; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext:
@@ -1016,10 +1000,9 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bzhi32_b4_commutative:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_b4_commutative:
@@ -1040,11 +1023,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_b4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
-; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    shll $8, %esi
+; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_b4_commutative:
@@ -1128,11 +1108,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_b0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
+; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_b0:
@@ -1214,11 +1191,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_b1_indexzext:
@@ -1307,11 +1282,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq (%rdi), %rax, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
+; X64-BMI1NOTBM-NEXT:    bextrq %rsi, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_b2_load:
@@ -1399,11 +1371,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq (%rdi), %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext:
@@ -1488,11 +1458,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_b4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
+; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_b4_commutative: