[FPEnv][AArch64] Add lowering of f128 STRICT_FSETCC

[john.brawn]

Summary

These get lowered to function calls, like the non-strict versions.

[pengfei]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

5293–5294

Does integer comparison have strict version?

[john.brawn]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

5293–5294

No. What's happening here for f128 is that just above we have a call to softenSetCCOperands that has turned the LHS into a function call that returns an integer result.

[pengfei]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

5293–5294

Oh I see, thanks!

[dmgreen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

5293–5294

And in that case the chain is updated to the call, so we use the new one here?

And in other integer cases we do not ever care about the chain?

[john.brawn]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

5293–5294

That's correct.

[dmgreen]

Sounds good then. LGTM

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::FREM, MVT::f128, Expand);
   setOperationAction(ISD::FRINT, MVT::f128, Expand);
   setOperationAction(ISD::FSIN, MVT::f128, Expand);
   setOperationAction(ISD::FSINCOS, MVT::f128, Expand);
   setOperationAction(ISD::FSQRT, MVT::f128, Expand);
   setOperationAction(ISD::FSUB, MVT::f128, Custom);
   setOperationAction(ISD::FTRUNC, MVT::f128, Expand);
   setOperationAction(ISD::SETCC, MVT::f128, Custom);
+  setOperationAction(ISD::STRICT_FSETCC, MVT::f128, Custom);
+  setOperationAction(ISD::STRICT_FSETCCS, MVT::f128, Custom);
   setOperationAction(ISD::BR_CC, MVT::f128, Custom);
   setOperationAction(ISD::SELECT, MVT::f128, Custom);
   setOperationAction(ISD::SELECT_CC, MVT::f128, Custom);
   setOperationAction(ISD::FP_EXTEND, MVT::f128, Custom);
 
   // Lowering for many of the conversions is actually specified by the non-f128
   // type. The LowerXXX function will be trivial when f128 isn't involved.
   setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
if (LHS.getValueType() == MVT::f128) {
     // If softenSetCCOperands returned a scalar, use it.
     if (!RHS.getNode()) {
       assert(LHS.getValueType() == Op.getValueType() &&
              "Unexpected setcc expansion!");
       return IsStrict ? DAG.getMergeValues({LHS, Chain}, dl) : LHS;
     }
   }
 
   if (LHS.getValueType().isInteger()) {
-    assert(!IsStrict && "Unexpected integer in strict fp comparison!");
     SDValue CCVal;
     SDValue Cmp = getAArch64Cmp(
         LHS, RHS, ISD::getSetCCInverse(CC, LHS.getValueType()), CCVal, DAG, dl);
 
     // Note that we inverted the condition above, so we reverse the order of
     // the true and false operands here.  This will allow the setcc to be
     // matched to a single CSINC instruction.
-    return DAG.getNode(AArch64ISD::CSEL, dl, VT, FVal, TVal, CCVal, Cmp);
+    SDValue Res = DAG.getNode(AArch64ISD::CSEL, dl, VT, FVal, TVal, CCVal, Cmp);
+    return IsStrict ? DAG.getMergeValues({Res, Chain}, dl) : Res;
   }
 
   // Now we know we're dealing with FP values.
   assert(LHS.getValueType() == MVT::f16 || LHS.getValueType() == MVT::f32 ||
          LHS.getValueType() == MVT::f64);
 
   // If that fails, we'll need to perform an FCMP + CSEL sequence.  Go ahead
   // and do the comparison.

llvm/test/CodeGen/AArch64/fp-intrinsics.ll

 
 ; CHECK-LABEL: trunc_f128:
 ; CHECK: bl truncl
 define fp128 @trunc_f128(fp128 %x) #0 {
   %val = call fp128 @llvm.experimental.constrained.trunc.f128(fp128 %x, metadata !"fpexcept.strict") #0
   ret fp128 %val
 }
 
-; TODO: fcmp (missing STRICT_FSETCC handling)
+; CHECK-LABEL: fcmp_olt_f128:
+; CHECK: bl __lttf2
+define i32 @fcmp_olt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ole_f128:
+; CHECK: bl __letf2
+define i32 @fcmp_ole_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ogt_f128:
+; CHECK: bl __gttf2
+define i32 @fcmp_ogt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_oge_f128:
+; CHECK: bl __getf2
+define i32 @fcmp_oge_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_oeq_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmp_oeq_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_one_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmp_one_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ult_f128:
+; CHECK: bl __getf2
+define i32 @fcmp_ult_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ule_f128:
+; CHECK: bl __gttf2
+define i32 @fcmp_ule_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ugt_f128:
+; CHECK: bl __letf2
+define i32 @fcmp_ugt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_uge_f128:
+; CHECK: bl __lttf2
+define i32 @fcmp_uge_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_ueq_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmp_ueq_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmp_une_f128:
+; CHECK: bl __netf2
+define i32 @fcmp_une_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_olt_f128:
+; CHECK: bl __lttf2
+define i32 @fcmps_olt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ole_f128:
+; CHECK: bl __letf2
+define i32 @fcmps_ole_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ogt_f128:
+; CHECK: bl __gttf2
+define i32 @fcmps_ogt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_oge_f128:
+; CHECK: bl __getf2
+define i32 @fcmps_oge_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_oeq_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmps_oeq_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_one_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmps_one_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ult_f128:
+; CHECK: bl __getf2
+define i32 @fcmps_ult_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ule_f128:
+; CHECK: bl __gttf2
+define i32 @fcmps_ule_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ugt_f128:
+; CHECK: bl __letf2
+define i32 @fcmps_ugt_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_uge_f128:
+; CHECK: bl __lttf2
+define i32 @fcmps_uge_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_ueq_f128:
+; CHECK: bl __eqtf2
+define i32 @fcmps_ueq_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
+
+; CHECK-LABEL: fcmps_une_f128:
+; CHECK: bl __netf2
+define i32 @fcmps_une_f128(fp128 %a, fp128 %b) #0 {
+  %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
+}
 
 
 ; Intrinsics to convert between floating-point types
 
 ; CHECK-LABEL: fptrunc_f32_f64:
 ; CHECK: fcvt s0, d0
 define float @fptrunc_f32_f64(double %x) #0 {
   %val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %x, metadata !"round.tonearest", metadata !"fpexcept.strict") #0