Index: lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- lib/Target/X86/X86ISelDAGToDAG.cpp
+++ lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2827,21 +2827,27 @@
 
   SDLoc DL(Node);
 
-  SDValue OrigNBits = NBits;
-  if (NBits.getValueType() != NVT) {
-    // Truncate the shift amount.
-    NBits = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, NBits);
-    insertDAGNode(*CurDAG, OrigNBits, NBits);
+  auto MakeValBeZeroExtendedI8 = [this, NVT, DL](SDValue Val,
+                                                 SDValue &OrigVal) {
+    OrigVal = Val;
 
-    // Insert 8-bit NBits into lowest 8 bits of NVT-sized (32 or 64-bit)
+    // Truncate the shift amount.
+    Val = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, Val);
+    insertDAGNode(*CurDAG, OrigVal, Val);
+
+    // Insert 8-bit Val into lowest 8 bits of NVT-sized (32 or 64-bit)
     // register. All the other bits are undefined, we do not care about them.
     SDValue ImplDef =
         SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, NVT), 0);
-    insertDAGNode(*CurDAG, OrigNBits, ImplDef);
-    NBits =
-        CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, NVT, ImplDef, NBits);
-    insertDAGNode(*CurDAG, OrigNBits, NBits);
-  }
+    insertDAGNode(*CurDAG, OrigVal, ImplDef);
+    Val = CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, NVT, ImplDef, Val);
+    insertDAGNode(*CurDAG, OrigVal, Val);
+
+    return Val;
+  };
+
+  SDValue OrigNBits;
+  NBits = MakeValBeZeroExtendedI8(NBits, OrigNBits);
 
   if (Subtarget->hasBMI2()) {
     // Great, just emit the the BZHI..
@@ -2861,7 +2867,19 @@
   SDValue C8 = CurDAG->getConstant(8, DL, MVT::i8);
   SDValue Control = CurDAG->getNode(ISD::SHL, DL, NVT, NBits, C8);
   insertDAGNode(*CurDAG, OrigNBits, Control);
-  // NOTE: could also try to extract  start  from  (x >> start)
+
+  // If the 'X' is *logically* shifted, we can fold that shift into 'control'.
+  if (X.getOpcode() == ISD::SRL) {
+    SDValue ShiftAmt = X.getOperand(1);
+    X = X.getOperand(0);
+
+    SDValue OrigShiftAmt;
+    ShiftAmt = MakeValBeZeroExtendedI8(ShiftAmt, OrigShiftAmt);
+
+    // Insert 8-bit ShiftAmt into lowest 8 bits of 'control'.
+    Control = CurDAG->getNode(ISD::OR, DL, NVT, Control, ShiftAmt);
+    insertDAGNode(*CurDAG, OrigNBits, Control);
+  }
 
   // And finally, form the BEXTR itself.
   SDValue Extract = CurDAG->getNode(X86ISD::BEXTR, DL, NVT, X, Control);
Index: test/CodeGen/X86/extract-bits.ll
===================================================================
--- test/CodeGen/X86/extract-bits.ll
+++ test/CodeGen/X86/extract-bits.ll
@@ -50,10 +50,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a0:
@@ -78,10 +77,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -184,10 +181,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a1_indexzext:
@@ -212,10 +208,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -252,13 +246,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_a2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a2_load:
@@ -285,12 +278,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_a2_load:
@@ -325,13 +315,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
@@ -358,12 +347,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
@@ -401,10 +387,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a4_commutative:
@@ -429,10 +414,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -477,11 +460,10 @@
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
 ; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %esi
 ; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
@@ -523,9 +505,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_a5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -677,10 +658,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -967,10 +946,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1118,12 +1096,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; 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:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_a2_load:
@@ -1268,12 +1243,10 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_a3_load_indexzext:
@@ -1417,10 +1390,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1605,9 +1576,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
@@ -1657,10 +1627,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b0:
@@ -1685,10 +1654,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1724,10 +1691,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b1_indexzext:
@@ -1752,10 +1718,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1792,13 +1756,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1825,12 +1788,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1865,13 +1825,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1898,12 +1857,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1941,10 +1897,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b4_commutative:
@@ -1969,10 +1924,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2017,11 +1970,10 @@
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
 ; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %esi
 ; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
@@ -2063,9 +2015,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -2214,10 +2165,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2355,10 +2304,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2503,12 +2451,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; 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:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b2_load:
@@ -2650,12 +2595,10 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b3_load_indexzext:
@@ -2796,10 +2739,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2979,9 +2920,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
@@ -5166,10 +5106,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d0:
@@ -5194,10 +5133,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5230,10 +5167,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d1_indexzext:
@@ -5258,10 +5194,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5296,12 +5230,11 @@
 ; X86-BMI1NOTBM-LABEL: bextr32_d2_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 {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -5327,12 +5260,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -5364,13 +5294,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; 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:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    orl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -5396,12 +5325,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -5445,11 +5371,10 @@
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
 ; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
+; X86-BMI1NOTBM-NEXT:    orl %ecx, %eax
+; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %esi
 ; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
@@ -5491,9 +5416,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    orl %esi, %edx
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -5673,10 +5597,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5845,10 +5767,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -6023,12 +5944,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; 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:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d2_load:
@@ -6200,12 +6118,10 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; 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:    # kill: def $esi killed $esi def $rsi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d3_load_indexzext:
@@ -6419,9 +6335,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    orq %rsi, %rdx
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
Index: test/CodeGen/X86/extract-lowbits.ll
===================================================================
--- test/CodeGen/X86/extract-lowbits.ll
+++ test/CodeGen/X86/extract-lowbits.ll
@@ -2908,7 +2908,7 @@
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1BMI2-NEXT:    bzhil %eax, (%ecx), %eax
+; X86-BMI1BMI2-NEXT:    bzhil %ecx, (%eax), %eax
 ; X86-BMI1BMI2-NEXT:    retl
 ;
 ; X64-NOBMI-LABEL: bzhi32_d2_load: