[PowerPC] Support constrained fp operation for setcc

[steven.zhang]

Summary

The constrained fp operation fcmp was added by https://reviews.llvm.org/D69281. This patch is trying to add the support for PowerPC backend. For now, only scalar type is supported.

[steven.zhang]

Gentle ping

[steven.zhang]

Gentle ping...

[uweigand]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

409

Shouldn't all f128 support be guarded by EnableQuadPrecision?

llvm/lib/Target/PowerPC/PPCInstrInfo.td

3896

This looks incorrect; you cannot use the same FCMPUS operation for both signaling and quiet compares. If I read the PowerISA correctly, the "fcmpo" instruction (not "fcmpu") is the correct one to use to implement signaling compares.

[steven.zhang]

Address reviewer's great comments.

[steven.zhang]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

409

Yes, it should be. But as this option has been removed, we don't need it anymore.

llvm/lib/Target/PowerPC/PPCInstrInfo.td

3896

You are completely right. Thank you for point out this.

[uweigand]

The one issue I still see is that when falling back to library calls on older platforms, signaling and non-signaling compares both are treated the same. But that's really a pre-existing problem because this is not fully supported by libgcc (see the comment in TargetLowering::softenSetCCOperands), so it's not really a problem with this patch.

The patch itself LGTM.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

409

Ah, I missed that. OK, that's fine then.

llvm/include/llvm/Target/TargetSelectionDAG.td

 def strict_fp_to_sint : SDNode<"ISD::STRICT_FP_TO_SINT",
                                SDTFPToIntOp, [SDNPHasChain]>;
 def strict_fp_to_uint : SDNode<"ISD::STRICT_FP_TO_UINT",
                                SDTFPToIntOp, [SDNPHasChain]>;
 def strict_sint_to_fp : SDNode<"ISD::STRICT_SINT_TO_FP",
                                SDTIntToFPOp, [SDNPHasChain]>;
 def strict_uint_to_fp : SDNode<"ISD::STRICT_UINT_TO_FP",
                                SDTIntToFPOp, [SDNPHasChain]>;
+def strict_fsetcc  : SDNode<"ISD::STRICT_FSETCC",  SDTSetCC, [SDNPHasChain]>;
+def strict_fsetccs : SDNode<"ISD::STRICT_FSETCCS", SDTSetCC, [SDNPHasChain]>;
 
 def setcc      : SDNode<"ISD::SETCC"      , SDTSetCC>;
 def select     : SDNode<"ISD::SELECT"     , SDTSelect>;
 def vselect    : SDNode<"ISD::VSELECT"    , SDTVSelect>;
 def selectcc   : SDNode<"ISD::SELECT_CC"  , SDTSelectCC>;
 
 def brcc       : SDNode<"ISD::BR_CC"      , SDTBrCC,   [SDNPHasChain]>;
 def brcond     : SDNode<"ISD::BRCOND"     , SDTBrcond, [SDNPHasChain]>;
def any_fp_to_uint : PatFrags<(ops node:$src),
                               [(strict_fp_to_uint node:$src),
                                (fp_to_uint node:$src)]>;
 def any_sint_to_fp : PatFrags<(ops node:$src),
                               [(strict_sint_to_fp node:$src),
                                (sint_to_fp node:$src)]>;
 def any_uint_to_fp : PatFrags<(ops node:$src),
                               [(strict_uint_to_fp node:$src),
                                (uint_to_fp node:$src)]>;
+def any_fsetcc : PatFrags<(ops node:$lhs, node:$rhs, node:$pred),
+                          [(strict_fsetcc node:$lhs, node:$rhs, node:$pred),
+                           (setcc node:$lhs, node:$rhs, node:$pred)]>;
+def any_fsetccs : PatFrags<(ops node:$lhs, node:$rhs, node:$pred),
+                          [(strict_fsetccs node:$lhs, node:$rhs, node:$pred),
+                           (setcc node:$lhs, node:$rhs, node:$pred)]>;
 
 multiclass binary_atomic_op_ord<SDNode atomic_op> {
   def NAME#_monotonic : PatFrag<(ops node:$ptr, node:$val),
       (!cast<SDPatternOperator>(NAME) node:$ptr, node:$val)> {
     let IsAtomic = 1;
     let IsAtomicOrderingMonotonic = 1;
   }
   def NAME#_acquire : PatFrag<(ops node:$ptr, node:$val),

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
   // PowerPC wants to turn select_cc of FP into fsel when possible.
   setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
   setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
 
   // PowerPC wants to optimize integer setcc a bit
   if (!Subtarget.useCRBits())
     setOperationAction(ISD::SETCC, MVT::i32, Custom);
 
+  if (Subtarget.hasFPU()) {
+    // The quiet comparison operation performed by
+    // llvm.experimental.constrained.fcmp will only raise an exception if
+    // either operand is a SNAN, which is matched with hw instruction behavior.
+    setOperationAction(ISD::STRICT_FSETCC, MVT::f32, Legal);
+    setOperationAction(ISD::STRICT_FSETCC, MVT::f64, Legal);
+    setOperationAction(ISD::STRICT_FSETCC, MVT::f128, Legal);
+
+    setOperationAction(ISD::STRICT_FSETCCS, MVT::f32, Legal);
+    setOperationAction(ISD::STRICT_FSETCCS, MVT::f64, Legal);
+    setOperationAction(ISD::STRICT_FSETCCS, MVT::f128, Legal);
+  }
+
   // PowerPC does not have BRCOND which requires SetCC
   if (!Subtarget.useCRBits())
     setOperationAction(ISD::BRCOND, MVT::Other, Expand);
 
   setOperationAction(ISD::BR_JT,  MVT::Other, Expand);
 
   if (Subtarget.hasSPE()) {
     // SPE has built-in conversions

llvm/lib/Target/PowerPC/PPCInstrInfo.td

multiclass FSetCCPat<SDNode SetCC, ValueType Ty, PatLeaf FCmp> {
   def : Pat<(i1 (SetCC Ty:$s1, Ty:$s2, SETEQ)),
             (EXTRACT_SUBREG (FCmp $s1, $s2), sub_eq)>;
   def : Pat<(i1 (SetCC Ty:$s1, Ty:$s2, SETUO)),
             (EXTRACT_SUBREG (FCmp $s1, $s2), sub_un)>;
 }
 
 let Predicates = [HasFPU] in {
 // SETCC for f32.
-defm : FSetCCPat<setcc, f32, FCMPUS>;
+defm : FSetCCPat<any_fsetcc, f32, FCMPUS>;
 
 // SETCC for f64.
-defm : FSetCCPat<setcc, f64, FCMPUD>;
+defm : FSetCCPat<any_fsetcc, f64, FCMPUD>;
 
 // SETCC for f128.
-defm : FSetCCPat<setcc, f128, XSCMPUQP>;
-}
+defm : FSetCCPat<any_fsetcc, f128, XSCMPUQP>;
+
+// SETCCS for f32.
+defm : FSetCCPat<strict_fsetccs, f32, FCMPUS>;
 
+// SETCCS for f64.
+defm : FSetCCPat<strict_fsetccs, f64, FCMPUD>;
+
+// SETCCS for f128.
+defm : FSetCCPat<strict_fsetccs, f128, XSCMPUQP>;
+}
 
 // This must be in this file because it relies on patterns defined in this file
 // after the inclusion of the instruction sets.
 let Predicates = [HasSPE] in {
 // SETCC for f32.
 def : Pat<(i1 (setcc f32:$s1, f32:$s2, SETOLT)),
           (EXTRACT_SUBREG (EFSCMPLT $s1, $s2), sub_gt)>;
 def : Pat<(i1 (setcc f32:$s1, f32:$s2, SETLT)),

llvm/test/CodeGen/PowerPC/fp-strict-fcmp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   < %s -mtriple=powerpc64-unknown-linux -mcpu=pwr8 | FileCheck %s \
+; RUN:   -check-prefix=P8
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   < %s -mtriple=powerpc64le-unknown-linux -mcpu=pwr9 -enable-ppc-quad-precision \
+; RUN:   | FileCheck %s -check-prefix=P9
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   < %s -mtriple=powerpc64le-unknown-linux -mcpu=pwr8 -mattr=-vsx | \
+; RUN:   FileCheck %s -check-prefix=NOVSX
+
+define i32 @test_f32_oeq_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_oeq_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_oeq_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_oeq_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"oeq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ogt_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ogt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ogt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ogt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ogt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_oge_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_oge_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, lt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_oge_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, lt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_oge_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, lt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"oge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_olt_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_olt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_olt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_olt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"olt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ole_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ole_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, gt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ole_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, gt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ole_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, gt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ole",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_one_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_one_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, eq
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_one_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, eq
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_one_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, eq
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"one",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ord_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ord_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r4, r3, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ord_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r4, r3, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ord_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r4, r3, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ord",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ueq_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ueq_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, eq, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ueq_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, eq, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ueq_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, eq, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ueq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ugt_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ugt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, gt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ugt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, gt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ugt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, gt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ugt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_uge_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_uge_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_uge_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_uge_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"uge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ult_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ult_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, lt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ult_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, lt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ult_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, lt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ult",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ule_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ule_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ule_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ule_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"ule",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_une_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_une_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_une_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_une_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"une",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_uno_q(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_uno_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r3, r4, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_uno_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r3, r4, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_uno_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r3, r4, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f32(
+                                               float %f1, float %f2, metadata !"uno",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_oeq_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_oeq_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_oeq_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_oeq_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"oeq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ogt_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ogt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ogt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ogt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ogt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_oge_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_oge_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, lt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_oge_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, lt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_oge_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, lt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"oge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_olt_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_olt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_olt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_olt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"olt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ole_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ole_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, gt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ole_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, gt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ole_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, gt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ole",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_one_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_one_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, eq
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_one_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, eq
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_one_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, eq
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"one",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ord_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ord_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r4, r3, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ord_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r4, r3, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ord_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r4, r3, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ord",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ueq_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ueq_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, eq, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ueq_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, eq, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ueq_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, eq, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ueq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ugt_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ugt_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, gt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ugt_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, gt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ugt_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, gt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ugt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_uge_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_uge_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_uge_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_uge_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"uge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ult_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ult_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, lt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ult_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, lt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ult_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, lt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ult",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ule_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ule_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ule_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ule_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"ule",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_une_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_une_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_une_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_une_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"une",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_uno_q(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_uno_q:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r3, r4, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_uno_q:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r3, r4, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_uno_q:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r3, r4, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmp.f64(
+                                               double %f1, double %f2, metadata !"uno",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_oeq_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_oeq_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_oeq_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_oeq_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"oeq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ogt_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ogt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ogt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ogt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ogt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_oge_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_oge_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, lt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_oge_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, lt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_oge_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, lt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"oge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_olt_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_olt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_olt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_olt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"olt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ole_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ole_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, gt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ole_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, gt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ole_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, gt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ole",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_one_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_one_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, eq
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_one_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, eq
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_one_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, eq
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"one",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ord_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ord_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r4, r3, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ord_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r4, r3, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ord_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r4, r3, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ord",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ueq_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ueq_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, eq, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ueq_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, eq, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ueq_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, eq, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ueq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ugt_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ugt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, gt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ugt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, gt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ugt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, gt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ugt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_uge_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_uge_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_uge_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_uge_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"uge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ult_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ult_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, lt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ult_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, lt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ult_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, lt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ult",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_ule_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_ule_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_ule_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_ule_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"ule",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_une_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_une_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_une_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_une_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"une",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f32_uno_s(i32 %a, i32 %b, float %f1, float %f2) #0 {
+; P8-LABEL: test_f32_uno_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r3, r4, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f32_uno_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r3, r4, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f32_uno_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r3, r4, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f32(
+                                               float %f1, float %f2, metadata !"uno",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_oeq_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_oeq_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_oeq_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_oeq_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"oeq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ogt_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ogt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ogt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ogt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ogt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_oge_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_oge_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, lt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_oge_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, lt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_oge_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, lt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"oge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_olt_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_olt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r3, r4
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_olt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r3, r4
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_olt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r3, r4
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"olt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ole_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ole_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, gt
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ole_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, gt
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ole_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, gt
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ole",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_one_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_one_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    crnor 4*cr5+lt, un, eq
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_one_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    crnor 4*cr5+lt, un, eq
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_one_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    crnor 4*cr5+lt, un, eq
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"one",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ord_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ord_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r4, r3, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ord_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r4, r3, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ord_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r4, r3, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ord",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ueq_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ueq_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, eq, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ueq_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, eq, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ueq_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, eq, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ueq",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ugt_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ugt_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, gt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ugt_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, gt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ugt_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, gt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ugt",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_uge_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_uge_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isellt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_uge_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_uge_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isellt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"uge",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ult_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ult_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    cror 4*cr5+lt, lt, un
+; P8-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ult_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    cror 4*cr5+lt, lt, un
+; P9-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ult_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    cror 4*cr5+lt, lt, un
+; NOVSX-NEXT:    isel r3, r3, r4, 4*cr5+lt
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ult",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_ule_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_ule_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iselgt r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_ule_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_ule_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iselgt r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"ule",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_une_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_une_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    iseleq r3, r4, r3
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_une_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_une_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    iseleq r3, r4, r3
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"une",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @test_f64_uno_s(i32 %a, i32 %b, double %f1, double %f2) #0 {
+; P8-LABEL: test_f64_uno_s:
+; P8:       # %bb.0:
+; P8-NEXT:    fcmpu cr0, f1, f2
+; P8-NEXT:    isel r3, r3, r4, un
+; P8-NEXT:    blr
+;
+; P9-LABEL: test_f64_uno_s:
+; P9:       # %bb.0:
+; P9-NEXT:    fcmpu cr0, f1, f2
+; P9-NEXT:    isel r3, r3, r4, un
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: test_f64_uno_s:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    fcmpu cr0, f1, f2
+; NOVSX-NEXT:    isel r3, r3, r4, un
+; NOVSX-NEXT:    blr
+  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
+                                               double %f1, double %f2, metadata !"uno",
+                                               metadata !"fpexcept.strict") #0
+  %res = select i1 %cond, i32 %a, i32 %b
+  ret i32 %res
+}
+
+define i32 @fcmp_olt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_olt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __ltkf2
+; P8-NEXT:    nop
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_olt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_olt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __ltkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"olt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ole_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ole_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __lekf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ole_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, gt, 4*cr1+gt
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ole_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __lekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ole", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ogt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ogt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gtkf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ogt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ogt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gtkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ogt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_oge_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_oge_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gekf2
+; P8-NEXT:    nop
+; P8-NEXT:    not r3, r3
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_oge_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, lt, 4*cr1+lt
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_oge_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    not r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"oge", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_oeq_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_oeq_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_oeq_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_oeq_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"oeq", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_one_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_one_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -160(r1)
+; P8-NEXT:    std r26, 112(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r27, 120(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r28, 128(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r29, 136(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r29, r5
+; P8-NEXT:    mr r28, r4
+; P8-NEXT:    mr r27, r3
+; P8-NEXT:    std r30, 144(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r30, r6
+; P8-NEXT:    bl __unordkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    mr r4, r28
+; P8-NEXT:    mr r5, r29
+; P8-NEXT:    mr r6, r30
+; P8-NEXT:    srwi r26, r3, 5
+; P8-NEXT:    mr r3, r27
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    ld r30, 144(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r29, 136(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r28, 128(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r27, 120(r1) # 8-byte Folded Reload
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    and r3, r26, r3
+; P8-NEXT:    ld r26, 112(r1) # 8-byte Folded Reload
+; P8-NEXT:    addi r1, r1, 160
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_one_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, eq, 4*cr1+eq
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_one_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r26, -48(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r27, -40(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r28, -32(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -80(r1)
+; NOVSX-NEXT:    mr r30, r6
+; NOVSX-NEXT:    mr r29, r5
+; NOVSX-NEXT:    mr r28, r4
+; NOVSX-NEXT:    mr r27, r3
+; NOVSX-NEXT:    bl __unordkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    mr r4, r28
+; NOVSX-NEXT:    mr r5, r29
+; NOVSX-NEXT:    mr r6, r30
+; NOVSX-NEXT:    srwi r26, r3, 5
+; NOVSX-NEXT:    mr r3, r27
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    and r3, r26, r3
+; NOVSX-NEXT:    addi r1, r1, 80
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r28, -32(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r27, -40(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r26, -48(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"one", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ult_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ult_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gekf2
+; P8-NEXT:    nop
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ult_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, lt, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ult_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ult", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ule_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ule_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gtkf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ule_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    iselgt r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ule_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gtkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ule", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ugt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ugt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __lekf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ugt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, gt, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ugt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __lekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ugt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_uge_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_uge_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __ltkf2
+; P8-NEXT:    nop
+; P8-NEXT:    not r3, r3
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_uge_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isellt r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_uge_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __ltkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    not r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"uge", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_ueq_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_ueq_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -160(r1)
+; P8-NEXT:    std r26, 112(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r27, 120(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r28, 128(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r29, 136(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r29, r5
+; P8-NEXT:    mr r28, r4
+; P8-NEXT:    mr r27, r3
+; P8-NEXT:    std r30, 144(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r30, r6
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    mr r4, r28
+; P8-NEXT:    mr r5, r29
+; P8-NEXT:    mr r6, r30
+; P8-NEXT:    srwi r26, r3, 5
+; P8-NEXT:    mr r3, r27
+; P8-NEXT:    bl __unordkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    ld r30, 144(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r29, 136(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r28, 128(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r27, 120(r1) # 8-byte Folded Reload
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    or r3, r3, r26
+; P8-NEXT:    ld r26, 112(r1) # 8-byte Folded Reload
+; P8-NEXT:    addi r1, r1, 160
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_ueq_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, eq, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_ueq_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r26, -48(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r27, -40(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r28, -32(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -80(r1)
+; NOVSX-NEXT:    mr r30, r6
+; NOVSX-NEXT:    mr r29, r5
+; NOVSX-NEXT:    mr r28, r4
+; NOVSX-NEXT:    mr r27, r3
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    mr r4, r28
+; NOVSX-NEXT:    mr r5, r29
+; NOVSX-NEXT:    mr r6, r30
+; NOVSX-NEXT:    srwi r26, r3, 5
+; NOVSX-NEXT:    mr r3, r27
+; NOVSX-NEXT:    bl __unordkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    or r3, r3, r26
+; NOVSX-NEXT:    addi r1, r1, 80
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r28, -32(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r27, -40(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r26, -48(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"ueq", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmp_une_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmp_une_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __nekf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmp_une_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    iseleq r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmp_une_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __nekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmp.f128(fp128 %a, fp128 %b, metadata !"une", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_olt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_olt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __ltkf2
+; P8-NEXT:    nop
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_olt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    isellt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_olt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __ltkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"olt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ole_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ole_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __lekf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ole_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, gt, 4*cr1+gt
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ole_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __lekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ole", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ogt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ogt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gtkf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ogt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    iselgt r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ogt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gtkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ogt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_oge_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_oge_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gekf2
+; P8-NEXT:    nop
+; P8-NEXT:    not r3, r3
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_oge_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, lt, 4*cr1+lt
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_oge_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    not r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"oge", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_oeq_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_oeq_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_oeq_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 0
+; P9-NEXT:    li r4, 1
+; P9-NEXT:    iseleq r3, r4, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_oeq_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"oeq", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_one_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_one_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -160(r1)
+; P8-NEXT:    std r26, 112(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r27, 120(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r28, 128(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r29, 136(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r29, r5
+; P8-NEXT:    mr r28, r4
+; P8-NEXT:    mr r27, r3
+; P8-NEXT:    std r30, 144(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r30, r6
+; P8-NEXT:    bl __unordkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    mr r4, r28
+; P8-NEXT:    mr r5, r29
+; P8-NEXT:    mr r6, r30
+; P8-NEXT:    srwi r26, r3, 5
+; P8-NEXT:    mr r3, r27
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    ld r30, 144(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r29, 136(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r28, 128(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r27, 120(r1) # 8-byte Folded Reload
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    and r3, r26, r3
+; P8-NEXT:    ld r26, 112(r1) # 8-byte Folded Reload
+; P8-NEXT:    addi r1, r1, 160
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_one_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    xscmpuqp cr1, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, un, 4*cr1+un
+; P9-NEXT:    crnor 4*cr5+gt, eq, 4*cr1+eq
+; P9-NEXT:    crnand 4*cr5+lt, 4*cr5+gt, 4*cr5+lt
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_one_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r26, -48(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r27, -40(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r28, -32(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -80(r1)
+; NOVSX-NEXT:    mr r30, r6
+; NOVSX-NEXT:    mr r29, r5
+; NOVSX-NEXT:    mr r28, r4
+; NOVSX-NEXT:    mr r27, r3
+; NOVSX-NEXT:    bl __unordkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    mr r4, r28
+; NOVSX-NEXT:    mr r5, r29
+; NOVSX-NEXT:    mr r6, r30
+; NOVSX-NEXT:    srwi r26, r3, 5
+; NOVSX-NEXT:    mr r3, r27
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    and r3, r26, r3
+; NOVSX-NEXT:    addi r1, r1, 80
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r28, -32(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r27, -40(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r26, -48(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"one", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ult_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ult_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gekf2
+; P8-NEXT:    nop
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ult_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, lt, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ult_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ult", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ule_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ule_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __gtkf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ule_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    iselgt r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ule_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __gtkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ule", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ugt_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ugt_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __lekf2
+; P8-NEXT:    nop
+; P8-NEXT:    extsw r3, r3
+; P8-NEXT:    neg r3, r3
+; P8-NEXT:    rldicl r3, r3, 1, 63
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ugt_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, gt, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ugt_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __lekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    extsw r3, r3
+; NOVSX-NEXT:    neg r3, r3
+; NOVSX-NEXT:    rldicl r3, r3, 1, 63
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ugt", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_uge_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_uge_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __ltkf2
+; P8-NEXT:    nop
+; P8-NEXT:    not r3, r3
+; P8-NEXT:    srwi r3, r3, 31
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_uge_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isellt r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_uge_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __ltkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    not r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 31
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"uge", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_ueq_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_ueq_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -160(r1)
+; P8-NEXT:    std r26, 112(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r27, 120(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r28, 128(r1) # 8-byte Folded Spill
+; P8-NEXT:    std r29, 136(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r29, r5
+; P8-NEXT:    mr r28, r4
+; P8-NEXT:    mr r27, r3
+; P8-NEXT:    std r30, 144(r1) # 8-byte Folded Spill
+; P8-NEXT:    mr r30, r6
+; P8-NEXT:    bl __eqkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    mr r4, r28
+; P8-NEXT:    mr r5, r29
+; P8-NEXT:    mr r6, r30
+; P8-NEXT:    srwi r26, r3, 5
+; P8-NEXT:    mr r3, r27
+; P8-NEXT:    bl __unordkf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    ld r30, 144(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r29, 136(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r28, 128(r1) # 8-byte Folded Reload
+; P8-NEXT:    ld r27, 120(r1) # 8-byte Folded Reload
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    or r3, r3, r26
+; P8-NEXT:    ld r26, 112(r1) # 8-byte Folded Reload
+; P8-NEXT:    addi r1, r1, 160
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_ueq_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    crnor 4*cr5+lt, eq, un
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    isel r3, 0, r3, 4*cr5+lt
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_ueq_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r26, -48(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r27, -40(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r28, -32(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -80(r1)
+; NOVSX-NEXT:    mr r30, r6
+; NOVSX-NEXT:    mr r29, r5
+; NOVSX-NEXT:    mr r28, r4
+; NOVSX-NEXT:    mr r27, r3
+; NOVSX-NEXT:    bl __eqkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    mr r4, r28
+; NOVSX-NEXT:    mr r5, r29
+; NOVSX-NEXT:    mr r6, r30
+; NOVSX-NEXT:    srwi r26, r3, 5
+; NOVSX-NEXT:    mr r3, r27
+; NOVSX-NEXT:    bl __unordkf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    or r3, r3, r26
+; NOVSX-NEXT:    addi r1, r1, 80
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r28, -32(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r27, -40(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    ld r26, -48(r1) # 8-byte Folded Reload
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"ueq", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+define i32 @fcmps_une_f128(fp128 %a, fp128 %b) #0 {
+; P8-LABEL: fcmps_une_f128:
+; P8:       # %bb.0:
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    bl __nekf2
+; P8-NEXT:    nop
+; P8-NEXT:    cntlzw r3, r3
+; P8-NEXT:    srwi r3, r3, 5
+; P8-NEXT:    xori r3, r3, 1
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
+; P8-NEXT:    blr
+;
+; P9-LABEL: fcmps_une_f128:
+; P9:       # %bb.0:
+; P9-NEXT:    xscmpuqp cr0, v2, v3
+; P9-NEXT:    li r3, 1
+; P9-NEXT:    iseleq r3, 0, r3
+; P9-NEXT:    blr
+;
+; NOVSX-LABEL: fcmps_une_f128:
+; NOVSX:       # %bb.0:
+; NOVSX-NEXT:    mflr r0
+; NOVSX-NEXT:    std r0, 16(r1)
+; NOVSX-NEXT:    stdu r1, -32(r1)
+; NOVSX-NEXT:    bl __nekf2
+; NOVSX-NEXT:    nop
+; NOVSX-NEXT:    cntlzw r3, r3
+; NOVSX-NEXT:    srwi r3, r3, 5
+; NOVSX-NEXT:    xori r3, r3, 1
+; NOVSX-NEXT:    addi r1, r1, 32
+; NOVSX-NEXT:    ld r0, 16(r1)
+; NOVSX-NEXT:    mtlr r0
+; NOVSX-NEXT:    blr
+  %cmp = call i1 @llvm.experimental.constrained.fcmps.f128(fp128 %a, fp128 %b, metadata !"une", metadata !"fpexcept.strict") #0
+  %conv = zext i1 %cmp to i32
+  ret i32 %conv
+}
+
+attributes #0 = { strictfp nounwind }
+
+declare i1 @llvm.experimental.constrained.fcmp.f32(float, float, metadata, metadata)
+declare i1 @llvm.experimental.constrained.fcmp.f64(double, double, metadata, metadata)
+declare i1 @llvm.experimental.constrained.fcmps.f32(float, float, metadata, metadata)
+declare i1 @llvm.experimental.constrained.fcmps.f64(double, double, metadata, metadata)
+declare i1 @llvm.experimental.constrained.fcmps.f128(fp128, fp128, metadata, metadata)
+declare i1 @llvm.experimental.constrained.fcmp.f128(fp128, fp128, metadata, metadata)