DAG: Handle lowering of unordered fcZero|fcSubnormal to fcmp

[arsenm]

Rebase and fix inverted compare type

[arsenm]

Rebase on more tests

[arsenm]

ping

[sepavloff]

LGTM.

[arsenm]

b59022b42e0b2c3f5f183306114b398c7eda7e92

llvm/lib/CodeGen/CodeGenCommonISel.cpp

FPClassTest llvm::invertFPClassTestIfSimpler(FPClassTest Test) {
   case fcPosSubnormal:
   case fcNegSubnormal:
   case fcZero:
   case fcPosZero:
   case fcNegZero:
   case fcFinite:
   case fcPosFinite:
   case fcNegFinite:
+  case fcZero | fcNan:
   case fcSubnormal | fcZero:
+  case fcSubnormal | fcZero | fcNan:
     return InvertedTest;
   default:
     return fcNone;
   }
 
   llvm_unreachable("covered FPClassTest");
 }
 

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

SDValue TargetLowering::expandFMINNUM_FMAXNUM(SDNode *Node,
   }
 
   if (SDValue SelCC = createSelectForFMINNUM_FMAXNUM(Node, DAG))
     return SelCC;
 
   return SDValue();
 }
 
-/// If this FPClassTest can be performed with a fcmp to 0, return the test mask
-/// for the floating-point mode.
-static bool isFCmpEqualZero(FPClassTest Test, const fltSemantics &Semantics,
+/// Returns a true value if if this FPClassTest can be performed with an ordered
+/// fcmp to 0, and a false value if it's an unordered fcmp to 0. Returns
+/// std::nullopt if it cannot be performed as a compare with 0.
+static std::optional<bool> isFCmpEqualZero(FPClassTest Test,
+                                           const fltSemantics &Semantics,
                                            const MachineFunction &MF) {
-  // TODO: Handle unordered compares
-  if (Test == fcZero &&
+  FPClassTest OrderedMask = Test & ~fcNan;
+  FPClassTest NanTest = Test & fcNan;
+  bool IsOrdered = NanTest == fcNone;
+  bool IsUnordered = NanTest == fcNan;
+
+  // Skip cases that are testing for only a qnan or snan.
+  if (!IsOrdered && !IsUnordered)
+    return std::nullopt;
+
+  if (OrderedMask == fcZero &&
       MF.getDenormalMode(Semantics).Input == DenormalMode::IEEE)
-    return true;
-  if (Test == (fcZero | fcSubnormal) &&
+    return IsOrdered;
+  if (OrderedMask == (fcZero | fcSubnormal) &&
       MF.getDenormalMode(Semantics).inputsAreZero())
-    return true;
-  return false;
+    return IsOrdered;
+  return std::nullopt;
 }
 
 SDValue TargetLowering::expandIS_FPCLASS(EVT ResultVT, SDValue Op,
                                          FPClassTest Test, SDNodeFlags Flags,
                                          const SDLoc &DL,
                                          SelectionDAG &DAG) const {
   EVT OperandVT = Op.getValueType();
   assert(OperandVT.isFloatingPoint());
SDValue TargetLowering::expandIS_FPCLASS(EVT ResultVT, SDValue Op,
   const Type *FloatTy = ScalarFloatVT.getTypeForEVT(*DAG.getContext());
   const llvm::fltSemantics &Semantics = FloatTy->getFltSemantics();
   bool IsF80 = (ScalarFloatVT == MVT::f80);
 
   // Some checks can be implemented using float comparisons, if floating point
   // exceptions are ignored.
   if (Flags.hasNoFPExcept() &&
       isOperationLegalOrCustom(ISD::SETCC, OperandVT.getScalarType())) {
-    if (isFCmpEqualZero(Test, Semantics, DAG.getMachineFunction()) &&
-        (isCondCodeLegalOrCustom(IsInverted ? ISD::SETUNE : ISD::SETOEQ,
-                                 OperandVT.getScalarType().getSimpleVT()))) {
+    ISD::CondCode OrderedCmpOpcode = IsInverted ? ISD::SETUNE : ISD::SETOEQ;
+    ISD::CondCode UnorderedCmpOpcode = IsInverted ? ISD::SETONE : ISD::SETUEQ;
+
+    if (std::optional<bool> IsCmp0 =
+            isFCmpEqualZero(Test, Semantics, DAG.getMachineFunction());
+        IsCmp0 && (isCondCodeLegalOrCustom(
+                      *IsCmp0 ? OrderedCmpOpcode : UnorderedCmpOpcode,
+
       // If denormals could be implicitly treated as 0, this is not equivalent
       // to a compare with 0 since it will also be true for denormals.
       return DAG.getSetCC(DL, ResultVT, Op,
                           DAG.getConstantFP(0.0, DL, OperandVT),
-                          IsInverted ? ISD::SETUNE : ISD::SETOEQ);
+                          *IsCmp0 ? OrderedCmpOpcode : UnorderedCmpOpcode);
     }
 
     if (Test == fcNan &&
         isCondCodeLegalOrCustom(IsInverted ? ISD::SETO : ISD::SETUO,
                                 OperandVT.getScalarType().getSimpleVT())) {
       return DAG.getSetCC(DL, ResultVT, Op, Op,
                           IsInverted ? ISD::SETO : ISD::SETUO);
     }

llvm/test/CodeGen/AMDGPU/llvm.is.fpclass.f16.ll

entry:
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f16(half %x) {
 ; GFX7SELDAG-LABEL: not_iszero_or_nan_f16:
 ; GFX7SELDAG:       ; %bb.0: ; %entry
 ; GFX7SELDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7SELDAG-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c00
-; GFX7SELDAG-NEXT:    s_movk_i32 s6, 0x7800
+; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c01
 ; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT:    v_cmp_eq_u32_e32 vcc, s4, v0
-; GFX7SELDAG-NEXT:    v_add_i32_e64 v1, s[4:5], -1, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x3ff
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e64 s[4:5], s4, v1
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
-; GFX7SELDAG-NEXT:    v_add_i32_e32 v0, vcc, 0xfffffc00, v0
-; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e32 vcc, s6, v0
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
+; GFX7SELDAG-NEXT:    v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT:    v_cmp_ne_u32_e64 s[4:5], 0, v0
+; GFX7SELDAG-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GFX7SELDAG-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[4:5]
 ; GFX7SELDAG-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7GLISEL-LABEL: not_iszero_or_nan_f16:
 ; GFX7GLISEL:       ; %bb.0: ; %entry
 ; GFX7GLISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7GLISEL-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
 ; GFX7GLISEL-NEXT:    v_subrev_i32_e32 v1, vcc, 1, v0
entry:
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f_daz(half %x) #0 {
 ; GFX7SELDAG-LABEL: not_iszero_or_nan_f_daz:
 ; GFX7SELDAG:       ; %bb.0: ; %entry
 ; GFX7SELDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7SELDAG-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c00
-; GFX7SELDAG-NEXT:    s_movk_i32 s6, 0x7800
+; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c01
 ; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT:    v_cmp_eq_u32_e32 vcc, s4, v0
-; GFX7SELDAG-NEXT:    v_add_i32_e64 v1, s[4:5], -1, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x3ff
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e64 s[4:5], s4, v1
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
-; GFX7SELDAG-NEXT:    v_add_i32_e32 v0, vcc, 0xfffffc00, v0
-; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e32 vcc, s6, v0
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
+; GFX7SELDAG-NEXT:    v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT:    v_cmp_ne_u32_e64 s[4:5], 0, v0
+; GFX7SELDAG-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GFX7SELDAG-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[4:5]
 ; GFX7SELDAG-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7GLISEL-LABEL: not_iszero_or_nan_f_daz:
 ; GFX7GLISEL:       ; %bb.0: ; %entry
 ; GFX7GLISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7GLISEL-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
 ; GFX7GLISEL-NEXT:    v_subrev_i32_e32 v1, vcc, 1, v0
entry:
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f_maybe_daz(half %x) #1 {
 ; GFX7SELDAG-LABEL: not_iszero_or_nan_f_maybe_daz:
 ; GFX7SELDAG:       ; %bb.0: ; %entry
 ; GFX7SELDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7SELDAG-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c00
-; GFX7SELDAG-NEXT:    s_movk_i32 s6, 0x7800
+; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x7c01
 ; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
-; GFX7SELDAG-NEXT:    v_cmp_eq_u32_e32 vcc, s4, v0
-; GFX7SELDAG-NEXT:    v_add_i32_e64 v1, s[4:5], -1, v0
-; GFX7SELDAG-NEXT:    s_movk_i32 s4, 0x3ff
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e64 s[4:5], s4, v1
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
-; GFX7SELDAG-NEXT:    v_add_i32_e32 v0, vcc, 0xfffffc00, v0
-; GFX7SELDAG-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX7SELDAG-NEXT:    v_cmp_gt_u32_e32 vcc, s6, v0
-; GFX7SELDAG-NEXT:    s_or_b64 s[4:5], s[4:5], vcc
+; GFX7SELDAG-NEXT:    v_cmp_gt_i32_e32 vcc, s4, v0
+; GFX7SELDAG-NEXT:    v_cmp_ne_u32_e64 s[4:5], 0, v0
+; GFX7SELDAG-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GFX7SELDAG-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[4:5]
 ; GFX7SELDAG-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7GLISEL-LABEL: not_iszero_or_nan_f_maybe_daz:
 ; GFX7GLISEL:       ; %bb.0: ; %entry
 ; GFX7GLISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7GLISEL-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
 ; GFX7GLISEL-NEXT:    v_subrev_i32_e32 v1, vcc, 1, v0

llvm/test/CodeGen/X86/is_fpclass.ll

entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1023)
   ret i1 %0
 }
 
 
 define i1 @iszero_or_nan_f(float %x) {
 ; CHECK-32-LABEL: iszero_or_nan_f:
 ; CHECK-32:       # %bb.0: # %entry
-; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
-; CHECK-32-NEXT:    setge %cl
-; CHECK-32-NEXT:    testl %eax, %eax
+; CHECK-32-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-32-NEXT:    fldz
+; CHECK-32-NEXT:    fucompp
+; CHECK-32-NEXT:    fnstsw %ax
+; CHECK-32-NEXT:    # kill: def $ah killed $ah killed $ax
+; CHECK-32-NEXT:    sahf
 ; CHECK-32-NEXT:    sete %al
-; CHECK-32-NEXT:    orb %cl, %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: iszero_or_nan_f:
 ; CHECK-64:       # %bb.0: # %entry
-; CHECK-64-NEXT:    movd %xmm0, %eax
-; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
-; CHECK-64-NEXT:    setge %cl
-; CHECK-64-NEXT:    testl %eax, %eax
+; CHECK-64-NEXT:    xorps %xmm1, %xmm1
+; CHECK-64-NEXT:    ucomiss %xmm1, %xmm0
 ; CHECK-64-NEXT:    sete %al
-; CHECK-64-NEXT:    orb %cl, %al
 ; CHECK-64-NEXT:    retq
 entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 99)  ; 0x60|0x3 = "zero|nan"
   ret i1 %0
 }
 
 define i1 @iszero_or_nan_f_daz(float %x) #0 {
 ; CHECK-32-LABEL: iszero_or_nan_f_daz:

 entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 99)  ; 0x60|0x3 = "zero|nan"
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f(float %x) {
 ; CHECK-32-LABEL: not_iszero_or_nan_f:
 ; CHECK-32:       # %bb.0: # %entry
-; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    leal -1(%eax), %edx
-; CHECK-32-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-32-NEXT:    setb %dl
-; CHECK-32-NEXT:    orb %cl, %dl
-; CHECK-32-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-32-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-32-NEXT:    setb %al
-; CHECK-32-NEXT:    orb %dl, %al
+; CHECK-32-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-32-NEXT:    fldz
+; CHECK-32-NEXT:    fucompp
+; CHECK-32-NEXT:    fnstsw %ax
+; CHECK-32-NEXT:    # kill: def $ah killed $ah killed $ax
+; CHECK-32-NEXT:    sahf
+; CHECK-32-NEXT:    setne %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: not_iszero_or_nan_f:
 ; CHECK-64:       # %bb.0: # %entry
-; CHECK-64-NEXT:    movd %xmm0, %eax
-; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    leal -1(%rax), %edx
-; CHECK-64-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-64-NEXT:    setb %dl
-; CHECK-64-NEXT:    orb %cl, %dl
-; CHECK-64-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-64-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-64-NEXT:    setb %al
-; CHECK-64-NEXT:    orb %dl, %al
+; CHECK-64-NEXT:    xorps %xmm1, %xmm1
+; CHECK-64-NEXT:    ucomiss %xmm1, %xmm0
+; CHECK-64-NEXT:    setne %al
 ; CHECK-64-NEXT:    retq
 entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 924)  ; ~0x60 = "~(zero|nan)"
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f_daz(float %x) #0 {
 ; CHECK-32-LABEL: not_iszero_or_nan_f_daz:
 ; CHECK-32:       # %bb.0: # %entry
 ; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
 ; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    leal -1(%eax), %edx
-; CHECK-32-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-32-NEXT:    setb %dl
-; CHECK-32-NEXT:    orb %cl, %dl
-; CHECK-32-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-32-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-32-NEXT:    setb %al
-; CHECK-32-NEXT:    orb %dl, %al
+; CHECK-32-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-32-NEXT:    setl %cl
+; CHECK-32-NEXT:    testl %eax, %eax
+; CHECK-32-NEXT:    setne %al
+; CHECK-32-NEXT:    andb %cl, %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: not_iszero_or_nan_f_daz:
 ; CHECK-64:       # %bb.0: # %entry
 ; CHECK-64-NEXT:    movd %xmm0, %eax
 ; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    leal -1(%rax), %edx
-; CHECK-64-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-64-NEXT:    setb %dl
-; CHECK-64-NEXT:    orb %cl, %dl
-; CHECK-64-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-64-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-64-NEXT:    setb %al
-; CHECK-64-NEXT:    orb %dl, %al
+; CHECK-64-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-64-NEXT:    setl %cl
+; CHECK-64-NEXT:    testl %eax, %eax
+; CHECK-64-NEXT:    setne %al
+; CHECK-64-NEXT:    andb %cl, %al
 ; CHECK-64-NEXT:    retq
 entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 924)  ; ~(0x60|0x3) = "~(zero|nan)"
   ret i1 %0
 }
 
 define i1 @not_iszero_or_nan_f_maybe_daz(float %x) #1 {
 ; CHECK-32-LABEL: not_iszero_or_nan_f_maybe_daz:
 ; CHECK-32:       # %bb.0: # %entry
 ; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
 ; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    leal -1(%eax), %edx
-; CHECK-32-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-32-NEXT:    setb %dl
-; CHECK-32-NEXT:    orb %cl, %dl
-; CHECK-32-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-32-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-32-NEXT:    setb %al
-; CHECK-32-NEXT:    orb %dl, %al
+; CHECK-32-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-32-NEXT:    setl %cl
+; CHECK-32-NEXT:    testl %eax, %eax
+; CHECK-32-NEXT:    setne %al
+; CHECK-32-NEXT:    andb %cl, %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: not_iszero_or_nan_f_maybe_daz:
 ; CHECK-64:       # %bb.0: # %entry
 ; CHECK-64-NEXT:    movd %xmm0, %eax
 ; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    leal -1(%rax), %edx
-; CHECK-64-NEXT:    cmpl $8388607, %edx # imm = 0x7FFFFF
-; CHECK-64-NEXT:    setb %dl
-; CHECK-64-NEXT:    orb %cl, %dl
-; CHECK-64-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-64-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-64-NEXT:    setb %al
-; CHECK-64-NEXT:    orb %dl, %al
+; CHECK-64-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-64-NEXT:    setl %cl
+; CHECK-64-NEXT:    testl %eax, %eax
+; CHECK-64-NEXT:    setne %al
+; CHECK-64-NEXT:    andb %cl, %al
 ; CHECK-64-NEXT:    retq
 entry:
   %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 924)  ; ~(0x60|0x3) = "~(zero|nan)"
   ret i1 %0
 }
 
 define i1 @iszero_or_qnan_f(float %x) {
 ; CHECK-32-LABEL: iszero_or_qnan_f:

 ; CHECK-64-NEXT:    retq
   %class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 243)  ; 0xf0|0x3 = "subnormal|zero|nan"
   ret i1 %class
 }
 
 define i1 @issubnormal_or_zero_or_nan_f_daz(float %x) #0 {
 ; CHECK-32-LABEL: issubnormal_or_zero_or_nan_f_daz:
 ; CHECK-32:       # %bb.0:
-; CHECK-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    testl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-32-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
-; CHECK-32-NEXT:    setge %al
-; CHECK-32-NEXT:    orb %cl, %al
+; CHECK-32-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-32-NEXT:    fldz
+; CHECK-32-NEXT:    fucompp
+; CHECK-32-NEXT:    fnstsw %ax
+; CHECK-32-NEXT:    # kill: def $ah killed $ah killed $ax
+; CHECK-32-NEXT:    sahf
+; CHECK-32-NEXT:    sete %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: issubnormal_or_zero_or_nan_f_daz:
 ; CHECK-64:       # %bb.0:
-; CHECK-64-NEXT:    movd %xmm0, %eax
-; CHECK-64-NEXT:    testl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
-; CHECK-64-NEXT:    setge %al
-; CHECK-64-NEXT:    orb %cl, %al
+; CHECK-64-NEXT:    xorps %xmm1, %xmm1
+; CHECK-64-NEXT:    ucomiss %xmm1, %xmm0
+; CHECK-64-NEXT:    sete %al
 ; CHECK-64-NEXT:    retq
   %class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 243)  ; 0xf0|0x3 = "subnormal|zero|nan"
   ret i1 %class
 }
 
 define i1 @issubnormal_or_zero_or_snan_f(float %x) {
 ; CHECK-32-LABEL: issubnormal_or_zero_or_snan_f:
 ; CHECK-32:       # %bb.0:

 ; CHECK-64-NEXT:    retq
   %class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 876)  ; ~(0x90|0x3) = ~"subnormal|nan"
   ret i1 %class
 }
 
 define i1 @not_issubnormal_or_zero_or_nan_f(float %x) {
 ; CHECK-32-LABEL: not_issubnormal_or_zero_or_nan_f:
 ; CHECK-32:       # %bb.0:
-; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-32-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-32-NEXT:    setb %al
-; CHECK-32-NEXT:    orb %cl, %al
+; CHECK-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; CHECK-32-NEXT:    testl $2139095040, %eax # imm = 0x7F800000
+; CHECK-32-NEXT:    setne %cl
+; CHECK-32-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
+; CHECK-32-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-32-NEXT:    setl %al
+; CHECK-32-NEXT:    andb %cl, %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: not_issubnormal_or_zero_or_nan_f:
 ; CHECK-64:       # %bb.0:
 ; CHECK-64-NEXT:    movd %xmm0, %eax
+; CHECK-64-NEXT:    testl $2139095040, %eax # imm = 0x7F800000
+; CHECK-64-NEXT:    setne %cl
 ; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-64-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-64-NEXT:    setb %al
-; CHECK-64-NEXT:    orb %cl, %al
+; CHECK-64-NEXT:    cmpl $2139095041, %eax # imm = 0x7F800001
+; CHECK-64-NEXT:    setl %al
+; CHECK-64-NEXT:    andb %cl, %al
 ; CHECK-64-NEXT:    retq
   %class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 780)  ; ~(0xf0|0x3) = ~"subnormal|zero|nan"
   ret i1 %class
 }
 
 define i1 @not_issubnormal_or_zero_or_nan_f_daz(float %x) #0 {
 ; CHECK-32-LABEL: not_issubnormal_or_zero_or_nan_f_daz:
 ; CHECK-32:       # %bb.0:
-; CHECK-32-NEXT:    movl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-32-NEXT:    andl {{[0-9]+}}(%esp), %eax
-; CHECK-32-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-32-NEXT:    sete %cl
-; CHECK-32-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-32-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-32-NEXT:    setb %al
-; CHECK-32-NEXT:    orb %cl, %al
+; CHECK-32-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-32-NEXT:    fldz
+; CHECK-32-NEXT:    fucompp
+; CHECK-32-NEXT:    fnstsw %ax
+; CHECK-32-NEXT:    # kill: def $ah killed $ah killed $ax
+; CHECK-32-NEXT:    sahf
+; CHECK-32-NEXT:    setne %al
 ; CHECK-32-NEXT:    retl
 ;
 ; CHECK-64-LABEL: not_issubnormal_or_zero_or_nan_f_daz:
 ; CHECK-64:       # %bb.0:
-; CHECK-64-NEXT:    movd %xmm0, %eax
-; CHECK-64-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
-; CHECK-64-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
-; CHECK-64-NEXT:    sete %cl
-; CHECK-64-NEXT:    addl $-8388608, %eax # imm = 0xFF800000
-; CHECK-64-NEXT:    cmpl $2130706432, %eax # imm = 0x7F000000
-; CHECK-64-NEXT:    setb %al
-; CHECK-64-NEXT:    orb %cl, %al
+; CHECK-64-NEXT:    xorps %xmm1, %xmm1
+; CHECK-64-NEXT:    ucomiss %xmm1, %xmm0
+; CHECK-64-NEXT:    setne %al
 ; CHECK-64-NEXT:    retq
   %class = tail call i1 @llvm.is.fpclass.f32(float %x, i32 780)  ; ~(0xf0|0x3) = ~"subnormal|zero|nan"
   ret i1 %class
 }
 
 define i1 @not_issubnormal_or_zero_or_snan_f(float %x) {
 ; CHECK-32-LABEL: not_issubnormal_or_zero_or_snan_f:
 ; CHECK-32:       # %bb.0: