diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -3991,17 +3991,35 @@
   // so we are not afraid that we might mess up BZHI/BEXTR pattern.
 
   SDValue NewShiftAmt;
-  if (ShiftAmt->getOpcode() == ISD::ADD || ShiftAmt->getOpcode() == ISD::SUB) {
+  if (ShiftAmt->getOpcode() == ISD::ADD || ShiftAmt->getOpcode() == ISD::SUB ||
+      ShiftAmt->getOpcode() == ISD::XOR) {
     SDValue Add0 = ShiftAmt->getOperand(0);
     SDValue Add1 = ShiftAmt->getOperand(1);
     auto *Add0C = dyn_cast<ConstantSDNode>(Add0);
     auto *Add1C = dyn_cast<ConstantSDNode>(Add1);
-    // If we are shifting by X+/-N where N == 0 mod Size, then just shift by X
-    // to avoid the ADD/SUB.
+    // If we are shifting by X+/-/^N where N == 0 mod Size, then just shift by X
+    // to avoid the ADD/SUB/XOR.
     if (Add1C && Add1C->getAPIntValue().urem(Size) == 0) {
       NewShiftAmt = Add0;
-      // If we are shifting by N-X where N == 0 mod Size, then just shift by -X
-      // to generate a NEG instead of a SUB of a constant.
+
+    } else if ((Add0C && Add0C->getAPIntValue().urem(Size) == Size - 1) ||
+               (Add1C && Add1C->getAPIntValue().urem(Size) == Size - 1)) {
+      // If we are doing a NOT on just the lower bits with (Size*N-1) -/^ X
+      // we can replace it with a NOT. In the XOR case it may save some code
+      // size, in the SUB case it also may save a move.
+      assert(Add0C == nullptr || Add1C == nullptr);
+
+      auto *ConstValOp = Add0C == nullptr ? Add1C : Add0C;
+      EVT OpVT = ShiftAmt.getValueType();
+      // ISelLowering will convert this to NOT already.
+      if (ConstValOp->isAllOnes())
+        return false;
+
+      NewShiftAmt = CurDAG->getNOT(DL, Add0C == nullptr ? Add0 : Add1, OpVT);
+
+      insertDAGNode(*CurDAG, OrigShiftAmt, NewShiftAmt);
+      // If we are shifting by N-X where N == 0 mod Size, then just shift by
+      // -X to generate a NEG instead of a SUB of a constant.
     } else if (ShiftAmt->getOpcode() == ISD::SUB && Add0C &&
                Add0C->getZExtValue() != 0) {
       EVT SubVT = ShiftAmt.getValueType();
diff --git a/llvm/test/CodeGen/X86/legalize-shift-64.ll b/llvm/test/CodeGen/X86/legalize-shift-64.ll
--- a/llvm/test/CodeGen/X86/legalize-shift-64.ll
+++ b/llvm/test/CodeGen/X86/legalize-shift-64.ll
@@ -10,7 +10,7 @@
 ; CHECK-NEXT:    movl %edx, %eax
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    shrl %edx
-; CHECK-NEXT:    xorb $31, %cl
+; CHECK-NEXT:    notb %cl
 ; CHECK-NEXT:    shrl %cl, %edx
 ; CHECK-NEXT:    retl
   %conv = zext i32 %xx to i64
diff --git a/llvm/test/CodeGen/X86/not-shift.ll b/llvm/test/CodeGen/X86/not-shift.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/X86/not-shift.ll
@@ -0,0 +1,637 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X86-NOBMI2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X86-NOBMI2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X86-NOBMI2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+bmi2 < %s | FileCheck %s --check-prefixes=X86-BMI2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+bmi2 < %s | FileCheck %s --check-prefixes=X86-BMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X64-NOBMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X64-NOBMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-bmi2 < %s | FileCheck %s --check-prefixes=X64-NOBMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+bmi2 < %s | FileCheck %s --check-prefixes=X64-BMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+bmi2 < %s | FileCheck %s --check-prefixes=X64-BMI2
+
+
+
+define i64 @sub63_shiftl64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub63_shiftl64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb $63, %cl
+; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    shldl %cl, %esi, %edx
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB0_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %eax, %edx
+; X86-NOBMI2-NEXT:    xorl %eax, %eax
+; X86-NOBMI2-NEXT:  .LBB0_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub63_shiftl64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movb $63, %cl
+; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB0_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %eax, %edx
+; X86-BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI2-NEXT:  .LBB0_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub63_shiftl64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub63_shiftl64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = sub i64 63, %cnt
+  %result = shl i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @xor63_shiftr64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor63_shiftr64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    xorb $63, %cl
+; X86-NOBMI2-NEXT:    movl %esi, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB1_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; X86-NOBMI2-NEXT:  .LBB1_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor63_shiftr64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    xorb $63, %cl
+; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB1_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI2-NEXT:  .LBB1_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor63_shiftr64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor63_shiftr64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i64 %cnt, 63
+  %result = lshr i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @sub127_shiftl64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub127_shiftl64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    xorb $127, %cl
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    shldl %cl, %esi, %edx
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB2_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %eax, %edx
+; X86-NOBMI2-NEXT:    xorl %eax, %eax
+; X86-NOBMI2-NEXT:  .LBB2_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub127_shiftl64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    xorb $127, %cl
+; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB2_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %eax, %edx
+; X86-BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI2-NEXT:  .LBB2_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub127_shiftl64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub127_shiftl64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = sub i64 127, %cnt
+  %result = shl i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @xor127_shiftr64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor127_shiftr64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    xorb $127, %cl
+; X86-NOBMI2-NEXT:    movl %esi, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB3_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; X86-NOBMI2-NEXT:  .LBB3_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor127_shiftr64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    xorb $127, %cl
+; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB3_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI2-NEXT:  .LBB3_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor127_shiftr64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor127_shiftr64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i64 %cnt, 127
+  %result = lshr i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @xor64_shiftl64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor64_shiftl64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    xorb $64, %cl
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    shldl %cl, %esi, %edx
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB4_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %eax, %edx
+; X86-NOBMI2-NEXT:    xorl %eax, %eax
+; X86-NOBMI2-NEXT:  .LBB4_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor64_shiftl64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    xorb $64, %cl
+; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB4_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %eax, %edx
+; X86-BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI2-NEXT:  .LBB4_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor64_shiftl64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor64_shiftl64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    shlxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i64 %cnt, 64
+  %result = shl i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @sub1s_shiftr64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub1s_shiftr64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    movl %esi, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB5_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; X86-NOBMI2-NEXT:  .LBB5_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub1s_shiftr64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    notb %cl
+; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB5_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI2-NEXT:  .LBB5_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub1s_shiftr64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub1s_shiftr64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i64 %cnt, -1
+  %result = lshr i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i64 @xor1s_shiftl64(i64 %val, i64 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor1s_shiftl64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    shldl %cl, %esi, %edx
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB6_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %eax, %edx
+; X86-NOBMI2-NEXT:    xorl %eax, %eax
+; X86-NOBMI2-NEXT:  .LBB6_2:
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor1s_shiftl64:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    notb %cl
+; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB6_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %eax, %edx
+; X86-BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI2-NEXT:  .LBB6_2:
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor1s_shiftl64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq %rsi, %rcx
+; X64-NOBMI2-NEXT:    movq %rdi, %rax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor1s_shiftl64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxq %rsi, %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i64 %cnt, -1
+  %result = shl i64 %val, %adjcnt
+  ret i64 %result
+}
+
+define i32 @sub31_shiftr32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub31_shiftr32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub31_shiftr32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub31_shiftr32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub31_shiftr32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = sub i32 31, %cnt
+  %result = lshr i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @xor31_shiftl32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor31_shiftl32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor31_shiftl32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shlxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor31_shiftl32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor31_shiftl32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i32 %cnt, 31
+  %result = shl i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @sub63_shiftr32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub63_shiftr32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub63_shiftr32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub63_shiftr32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub63_shiftr32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = sub i32 63, %cnt
+  %result = lshr i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @xor63_shiftl32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor63_shiftl32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor63_shiftl32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shlxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor63_shiftl32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor63_shiftl32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i32 %cnt, 63
+  %result = shl i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @xor32_shiftr32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor32_shiftr32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor32_shiftr32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor32_shiftr32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor32_shiftr32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i32 %cnt, 32
+  %result = lshr i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @sub1s_shiftl32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: sub1s_shiftl32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: sub1s_shiftl32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shlxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: sub1s_shiftl32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: sub1s_shiftl32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i32 %cnt, -1
+  %result = shl i32 %val, %adjcnt
+  ret i32 %result
+}
+
+define i32 @xor1s_shiftr32(i32 %val, i32 %cnt) nounwind {
+; X86-NOBMI2-LABEL: xor1s_shiftr32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    notb %cl
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: xor1s_shiftr32:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    notb %al
+; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: xor1s_shiftr32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    movl %edi, %eax
+; X64-NOBMI2-NEXT:    notb %cl
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: xor1s_shiftr32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    notb %sil
+; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %adjcnt = xor i32 %cnt, -1
+  %result = lshr i32 %val, %adjcnt
+  ret i32 %result
+}