[SelectionDAG] Fill in some more cases in `isKnownNeverZero`

[goldstein.w.n]

Summary

This mostly copies cases that already exist in ValueTracking, although
it skips the more complex ones. Those can be filled in as needed.

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5074–5117

I assume we checked operand 1 first to favor the cheap case of a non-zero constant right hand side.

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5120

return after break?

5155

too many blank lines

5158

I don't think curly braces are needed here

5175

Or here

[goldstein.w.n]

Swap operand order. Fix some style nits

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5074–5117

Makes sense, updated that everywhere else.

[goldstein.w.n]

Rebase

[RKSimon]

Are we really hitting all these cases? We should really ensure we have test coverage for them all.

[goldstein.w.n]

In D149199#4298104, @RKSimon wrote:

Are we really hitting all these cases? We should really ensure we have test coverage for them all.

Certainly not all of them. But fair point, will add tests.

[goldstein.w.n]

Rebase with tests

[goldstein.w.n]

In D149199#4298104, @RKSimon wrote:

Are we really hitting all these cases? We should really ensure we have test coverage for them all.

Added tests at D149299. Is there a better way to isolate/test DAGComber codes?

[goldstein.w.n]

Rebase

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5089

This looks like a bad merge?

[goldstein.w.n]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5089

hmm?

[goldstein.w.n]

Rebase

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5089

Sorry, I think I was miscounting brackets :)

[goldstein.w.n]

Rebase

[goldstein.w.n]

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

5089

You where right!

[goldstein.w.n]

Fix broken parens

[goldstein.w.n]

rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

ping. Dropped the assosiated setcc patch, but would like to get this through as an improvement to the helper as is.

[RKSimon]

LGTM

[goldstein.w.n]

Rebase

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

bool SelectionDAG::isKnownNeverZero(SDValue Op, unsigned Depth) const {
   assert(!Op.getValueType().isFloatingPoint() &&
          "Floating point types unsupported - use isKnownNeverZeroFloat");
 
   // If the value is a constant, we can obviously see if it is a zero or not.
   if (ISD::matchUnaryPredicate(Op,
                                [](ConstantSDNode *C) { return !C->isZero(); }))
     return true;
 
-  // TODO: Recognize more cases here.
+  // TODO: Recognize more cases here. Most of the cases are also incomplete to
+  // some degree.
   switch (Op.getOpcode()) {
-  default: break;
+  default:
+    break;
+
   case ISD::OR:
+    return isKnownNeverZero(Op.getOperand(1), Depth + 1) ||
+           isKnownNeverZero(Op.getOperand(0), Depth + 1);
+
+  case ISD::VSELECT:
+  case ISD::SELECT:
+    return isKnownNeverZero(Op.getOperand(1), Depth + 1) &&
+           isKnownNeverZero(Op.getOperand(2), Depth + 1);
+
+  case ISD::SHL:
+    if (Op->getFlags().hasNoSignedWrap() || Op->getFlags().hasNoUnsignedWrap())
+      return isKnownNeverZero(Op.getOperand(0), Depth + 1);
+
+    // 1 << X is never zero. TODO: This can be expanded if we can bound X.
+    // The expression is really !Known.One[BitWidth-MaxLog2(Known):0].isZero()
+    if (computeKnownBits(Op.getOperand(0), Depth + 1).One[0])
+      return true;
+    break;
+
+  case ISD::UADDSAT:
+  case ISD::UMAX:
+    return isKnownNeverZero(Op.getOperand(1), Depth + 1) ||
+           isKnownNeverZero(Op.getOperand(0), Depth + 1);
+
+  case ISD::UMIN:
+    return isKnownNeverZero(Op.getOperand(1), Depth + 1) &&
+           isKnownNeverZero(Op.getOperand(0), Depth + 1);
+
+  case ISD::ROTL:
+  case ISD::ROTR:
+  case ISD::BITREVERSE:
+  case ISD::BSWAP:
+  case ISD::CTPOP:
+  case ISD::ABS:
+    return isKnownNeverZero(Op.getOperand(0), Depth + 1);
+
+  case ISD::SRA:
+  case ISD::SRL:
+    if (Op->getFlags().hasExact())
+      return isKnownNeverZero(Op.getOperand(0), Depth + 1);
+    // Signed >> X is never zero. TODO: This can be expanded if we can bound X.
+    // The expression is really
+    // !Known.One[SignBit:SignBit-(BitWidth-MaxLog2(Known))].isZero()
+    if (computeKnownBits(Op.getOperand(0), Depth + 1).isNegative())
+      return true;
+    break;
+
+  case ISD::UDIV:
+  case ISD::SDIV:
+    // div exact can only produce a zero if the dividend is zero.
+    // TODO: For udiv this is also true if Op1 u<= Op0
+    if (Op->getFlags().hasExact())
+      return isKnownNeverZero(Op.getOperand(0), Depth + 1);
+    break;
+
+  case ISD::ADD:
+    if (Op->getFlags().hasNoUnsignedWrap())
       if (isKnownNeverZero(Op.getOperand(1), Depth + 1) ||
           isKnownNeverZero(Op.getOperand(0), Depth + 1))
         return true;
+    // TODO: There are a lot more cases we can prove for add.
     break;
+
+  case ISD::SUB: {
+    if (isNullConstant(Op.getOperand(0)))
+      return isKnownNeverZero(Op.getOperand(1), Depth + 1);
+
+    std::optional<bool> ne =
+        KnownBits::ne(computeKnownBits(Op.getOperand(0), Depth + 1),
+                      computeKnownBits(Op.getOperand(1), Depth + 1));
+    return ne && *ne;
+  }
+
+  case ISD::MUL:
+    if (Op->getFlags().hasNoSignedWrap() || Op->getFlags().hasNoUnsignedWrap())
+      if (isKnownNeverZero(Op.getOperand(1), Depth + 1) &&
+          isKnownNeverZero(Op.getOperand(0), Depth + 1))
+        return true;
+    break;
+
+  case ISD::ZERO_EXTEND:
+  case ISD::SIGN_EXTEND:
+    return isKnownNeverZero(Op.getOperand(0), Depth + 1);
   }
 
   return computeKnownBits(Op, Depth).isNonZero();
 }
 
 bool SelectionDAG::isEqualTo(SDValue A, SDValue B) const {
   // Check the obvious case.
   if (A == B) return true;
 
   // For for negative and positive zero.
   if (const ConstantFPSDNode *CA = dyn_cast<ConstantFPSDNode>(A))
     if (const ConstantFPSDNode *CB = dyn_cast<ConstantFPSDNode>(B))
       if (CA->isZero() && CB->isZero()) return true;
 
   // Otherwise they may not be equal.
   return false;
 }
 
 // Only bits set in Mask must be negated, other bits may be arbitrary.
 SDValue llvm::getBitwiseNotOperand(SDValue V, SDValue Mask, bool AllowUndefs) {
   if (isBitwiseNot(V, AllowUndefs))
     return V.getOperand(0);
 
   // Handle any_extend (not (truncate X)) pattern, where Mask only sets
   // bits in the non-extended part.
   ConstantSDNode *MaskC = isConstOrConstSplat(Mask);
   if (!MaskC || V.getOpcode() != ISD::ANY_EXTEND)

llvm/test/CodeGen/X86/divrem-by-select.ll

 ; CHECK-X64-V3-NEXT:    vmovq %rax, %xmm1
 ; CHECK-X64-V3-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
 ; CHECK-X64-V3-NEXT:    retq
 ;
 ; CHECK-X64-V4-LABEL: udiv_identity_const_todo_getter_nonzero:
 ; CHECK-X64-V4:       # %bb.0:
 ; CHECK-X64-V4-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; CHECK-X64-V4-NEXT:    vpmovq2m %xmm0, %k1
-; CHECK-X64-V4-NEXT:    vpbroadcastq {{.*#+}} xmm0 = [1,1]
-; CHECK-X64-V4-NEXT:    vpbroadcastq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 {%k1}
-; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm0, %rcx
-; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm1, %rax
-; CHECK-X64-V4-NEXT:    xorl %edx, %edx
-; CHECK-X64-V4-NEXT:    divq %rcx
-; CHECK-X64-V4-NEXT:    movq %rax, %rcx
-; CHECK-X64-V4-NEXT:    vmovq %xmm0, %rsi
-; CHECK-X64-V4-NEXT:    vmovq %xmm1, %rax
-; CHECK-X64-V4-NEXT:    xorl %edx, %edx
-; CHECK-X64-V4-NEXT:    divq %rsi
+; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm1, %rdx
+; CHECK-X64-V4-NEXT:    movabsq $-3689348814741910323, %rax # imm = 0xCCCCCCCCCCCCCCCD
+; CHECK-X64-V4-NEXT:    mulxq %rax, %rcx, %rcx
 ; CHECK-X64-V4-NEXT:    vmovq %rcx, %xmm0
-; CHECK-X64-V4-NEXT:    vmovq %rax, %xmm1
-; CHECK-X64-V4-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; CHECK-X64-V4-NEXT:    vmovq %xmm1, %rdx
+; CHECK-X64-V4-NEXT:    mulxq %rax, %rax, %rax
+; CHECK-X64-V4-NEXT:    vmovq %rax, %xmm2
+; CHECK-X64-V4-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0]
+; CHECK-X64-V4-NEXT:    vpsrlq $3, %xmm0, %xmm1 {%k1}
+; CHECK-X64-V4-NEXT:    vmovdqa %xmm1, %xmm0
 ; CHECK-X64-V4-NEXT:    retq
 
   ;; Fails at the moment because `10` is even so there is no common
   ;; bit between the arms of the select so the default case of
   ;; isKnownNeverZero has no guaranteed 1 bits.
   %d = select <2 x i1> %c, <2 x i64> <i64 10, i64 10>, <2 x i64> <i64 1, i64 1>
   %r = udiv <2 x i64> %x, %d
   ret <2 x i64> %r

 ; CHECK-X64-V3-NEXT:    divq %rsi
 ; CHECK-X64-V3-NEXT:    vmovq %rcx, %xmm0
 ; CHECK-X64-V3-NEXT:    vmovq %rax, %xmm1
 ; CHECK-X64-V3-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
 ; CHECK-X64-V3-NEXT:    retq
 ;
 ; CHECK-X64-V4-LABEL: udiv_indentity_non_zero:
 ; CHECK-X64-V4:       # %bb.0:
-; CHECK-X64-V4-NEXT:    vpsllq $63, %xmm0, %xmm0
-; CHECK-X64-V4-NEXT:    vpmovq2m %xmm0, %k1
-; CHECK-X64-V4-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; CHECK-X64-V4-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [1,1]
-; CHECK-X64-V4-NEXT:    vpsubq %xmm0, %xmm2, %xmm3 {%k1}
-; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm3, %rcx
+; CHECK-X64-V4-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; CHECK-X64-V4-NEXT:    vpsubq %xmm3, %xmm2, %xmm2
+; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm2, %rcx
 ; CHECK-X64-V4-NEXT:    vpextrq $1, %xmm1, %rax
 ; CHECK-X64-V4-NEXT:    xorl %edx, %edx
 ; CHECK-X64-V4-NEXT:    divq %rcx
 ; CHECK-X64-V4-NEXT:    movq %rax, %rcx
-; CHECK-X64-V4-NEXT:    vmovq %xmm3, %rsi
+; CHECK-X64-V4-NEXT:    vmovq %xmm2, %rsi
 ; CHECK-X64-V4-NEXT:    vmovq %xmm1, %rax
 ; CHECK-X64-V4-NEXT:    xorl %edx, %edx
 ; CHECK-X64-V4-NEXT:    divq %rsi
+; CHECK-X64-V4-NEXT:    vpsllq $63, %xmm0, %xmm0
+; CHECK-X64-V4-NEXT:    vpmovq2m %xmm0, %k1
 ; CHECK-X64-V4-NEXT:    vmovq %rcx, %xmm0
-; CHECK-X64-V4-NEXT:    vmovq %rax, %xmm1
-; CHECK-X64-V4-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; CHECK-X64-V4-NEXT:    vmovq %rax, %xmm2
+; CHECK-X64-V4-NEXT:    vpunpcklqdq {{.*#+}} xmm1 {%k1} = xmm2[0],xmm0[0]
+; CHECK-X64-V4-NEXT:    vmovdqa %xmm1, %xmm0
 ; CHECK-X64-V4-NEXT:    retq
   %non_zero = add nsw nuw <2 x i64> %y, <i64 1, i64 1>
   %d = select <2 x i1> %c, <2 x i64> %non_zero, <2 x i64> <i64 1, i64 1>
   %r = udiv <2 x i64> %x, %d
   ret <2 x i64> %r
 }
 
 define <2 x i64> @udiv_indentity_zero(<2 x i1> %c, <2 x i64> %x) {

llvm/test/CodeGen/X86/known-never-zero.ll

 
 define i32 @select_known_nonzero(i1 %c, i32 %x) {
 ; CHECK-LABEL: select_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    orl $1, %esi
 ; CHECK-NEXT:    testb $1, %dil
 ; CHECK-NEXT:    movl $122, %eax
 ; CHECK-NEXT:    cmovnel %esi, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %y = or i32 %x, 1
   %z = select i1 %c, i32 %y, i32 122
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @select_maybe_zero(i1 %c, i32 %x) {

 
 define i32 @shl_known_nonzero_1s_bit_set(i32 %x) {
 ; CHECK-LABEL: shl_known_nonzero_1s_bit_set:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $123, %eax
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %z = shl i32 123, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @shl_known_nonzero_nsw(i32 %x, i32 %yy) {
 ; CHECK-LABEL: shl_known_nonzero_nsw:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    orl $256, %esi # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %esi
-; CHECK-NEXT:    bsfl %esi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %esi, %eax
 ; CHECK-NEXT:    retq
   %y = or i32 %yy, 256
   %z = shl nsw i32 %y, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @shl_known_nonzero_nuw(i32 %x, i32 %yy) {
 ; CHECK-LABEL: shl_known_nonzero_nuw:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    orl $256, %esi # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %esi
-; CHECK-NEXT:    bsfl %esi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %esi, %eax
 ; CHECK-NEXT:    retq
   %y = or i32 %yy, 256
   %z = shl nuw i32 %y, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @shl_maybe_zero(i32 %x, i32 %y) {

 }
 
 define i32 @uaddsat_known_nonzero(i32 %x) {
 ; CHECK-LABEL: uaddsat_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    incl %edi
 ; CHECK-NEXT:    movl $-1, %eax
 ; CHECK-NEXT:    cmovnel %edi, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %z = call i32 @llvm.uadd.sat.i32(i32 %x, i32 1)
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @uaddsat_maybe_zero(i32 %x, i32 %y) {
 ; CHECK-LABEL: uaddsat_maybe_zero:

 ; CHECK-LABEL: umax_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %esi, %ecx
 ; CHECK-NEXT:    movl $4, %eax
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    cmpl %eax, %edi
 ; CHECK-NEXT:    cmoval %edi, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %yy = shl nuw i32 4, %y
   %z = call i32 @llvm.umax.i32(i32 %x, i32 %yy)
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @umax_maybe_zero(i32 %x, i32 %y) {

 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $4, %eax
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    addl $4, %esi
 ; CHECK-NEXT:    cmpl %esi, %eax
 ; CHECK-NEXT:    cmovbl %eax, %esi
-; CHECK-NEXT:    bsfl %esi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %esi, %eax
 ; CHECK-NEXT:    retq
   %x = shl nuw i32 4, %xx
   %y = add nuw nsw i32 %yy, 4
   %z = call i32 @llvm.umin.i32(i32 %x, i32 %y)
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 

 
 define i32 @sra_known_nonzero_exact(i32 %x, i32 %yy) {
 ; CHECK-LABEL: sra_known_nonzero_exact:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    orl $256, %esi # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    sarl %cl, %esi
-; CHECK-NEXT:    bsfl %esi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %esi, %eax
 ; CHECK-NEXT:    retq
   %y = or i32 %yy, 256
   %z = ashr exact i32 %y, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @sra_maybe_zero(i32 %x, i32 %y) {

 
 define i32 @srl_known_nonzero_sign_bit_set(i32 %x) {
 ; CHECK-LABEL: srl_known_nonzero_sign_bit_set:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $-2147360405, %eax # imm = 0x8001E16B
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shrl %cl, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %z = lshr i32 2147606891, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @srl_known_nonzero_exact(i32 %x, i32 %yy) {
 ; CHECK-LABEL: srl_known_nonzero_exact:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    orl $256, %esi # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shrl %cl, %esi
-; CHECK-NEXT:    bsfl %esi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %esi, %eax
 ; CHECK-NEXT:    retq
   %y = or i32 %yy, 256
   %z = lshr exact i32 %y, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @srl_maybe_zero(i32 %x, i32 %y) {

 
 define i32 @udiv_known_nonzero(i32 %xx, i32 %y) {
 ; CHECK-LABEL: udiv_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    orl $64, %eax
 ; CHECK-NEXT:    xorl %edx, %edx
 ; CHECK-NEXT:    divl %esi
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %x = or i32 %xx, 64
   %z = udiv exact i32 %x, %y
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @udiv_maybe_zero(i32 %x, i32 %y) {

 
 define i32 @sdiv_known_nonzero(i32 %xx, i32 %y) {
 ; CHECK-LABEL: sdiv_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    orl $64, %eax
 ; CHECK-NEXT:    cltd
 ; CHECK-NEXT:    idivl %esi
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %x = or i32 %xx, 64
   %z = sdiv exact i32 %x, %y
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @sdiv_maybe_zero(i32 %x, i32 %y) {
; CHECK-NEXT:    retq
   ret i32 %r
 }
 
 define i32 @add_known_nonzero(i32 %xx, i32 %y) {
 ; CHECK-LABEL: add_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    orl $1, %edi
 ; CHECK-NEXT:    addl %esi, %edi
-; CHECK-NEXT:    bsfl %edi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %edi, %eax
 ; CHECK-NEXT:    retq
   %x = or i32 %xx, 1
   %z = add nuw i32 %x, %y
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @add_maybe_zero(i32 %xx, i32 %y) {

 define i32 @sub_known_nonzero_neg_case(i32 %xx) {
 ; CHECK-LABEL: sub_known_nonzero_neg_case:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $256, %eax # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    negl %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %x = shl nuw nsw i32 256, %xx
   %z = sub i32 0, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @sub_known_nonzero_ne_case(i32 %xx, i32 %yy) {
 ; CHECK-LABEL: sub_known_nonzero_ne_case:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    orl $64, %eax
 ; CHECK-NEXT:    andl $-65, %edi
 ; CHECK-NEXT:    subl %eax, %edi
-; CHECK-NEXT:    bsfl %edi, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %edi, %eax
 ; CHECK-NEXT:    retq
   %x = or i32 %xx, 64
   %y = and i32 %xx, -65
   %z = sub i32 %y, %x
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 

 define i32 @zext_known_nonzero(i16 %xx) {
 ; CHECK-LABEL: zext_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $256, %eax # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    movzwl %ax, %eax
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %x = shl nuw nsw i16 256, %xx
   %z = zext i16 %x to i32
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @zext_maybe_zero(i16 %x) {

 define i32 @sext_known_nonzero(i16 %xx) {
 ; CHECK-LABEL: sext_known_nonzero:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl %edi, %ecx
 ; CHECK-NEXT:    movl $256, %eax # imm = 0x100
 ; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    cwtl
-; CHECK-NEXT:    bsfl %eax, %ecx
-; CHECK-NEXT:    movl $32, %eax
-; CHECK-NEXT:    cmovnel %ecx, %eax
+; CHECK-NEXT:    rep bsfl %eax, %eax
 ; CHECK-NEXT:    retq
   %x = shl nuw nsw i16 256, %xx
   %z = sext i16 %x to i32
   %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
   ret i32 %r
 }
 
 define i32 @sext_maybe_zero(i16 %x) {