[PowerPC] Fix the none tail call in scalar MASS conversion

[masoud.ataei]

Summary

This patch is proposing a fix for patch https://reviews.llvm.org/D101759
on none tail call math function conversion to MASS call.

[nemanjai]

LGTM other than the test cases that need to be improved.

llvm/test/CodeGen/PowerPC/lower-intrinsics-afn-mass_notail.ll

6

This check is not adequate. You have added code to ensure that the function isn't tail called when it isn't in a tail call position, yet your test doesn't check anything along those lines.
It is probably best to simply use update_llc_test_checks.py to produce the checks.

[jsji]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

17760

What is the reason that we don't want to use TLI.makeLibCall here?

[masoud.ataei]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

17760

At first, I thought this function will be way simpler than TLI.makeLibCall but now that they are too similar, I will try to call makeLibCall function. There are still some implementation details that needs to be take care of, like RTLIB::Libcall and CallOptions.

Since I am fixing a bug here, I will commit this patch as is and will work on creating a new patch to use TLI.makeLibCall.

[jsji]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

17760

Sounds good. Thanks!

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

if (PartVT == MVT::f64 &&
     Parts[0] = Val;
     return true;
   }
   return false;
 }
 
 SDValue PPCTargetLowering::lowerToLibCall(const char *LibCallName, SDValue Op,
                                           SelectionDAG &DAG) const {
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   TargetLowering::CallLoweringInfo CLI(DAG);
   EVT RetVT = Op.getValueType();
+  Type *RetTy = RetVT.getTypeForEVT(*DAG.getContext());
   SDValue Callee =
       DAG.getExternalSymbol(LibCallName, TLI.getPointerTy(DAG.getDataLayout()));
   bool SignExtend = TLI.shouldSignExtendTypeInLibCall(RetVT, false);
   TargetLowering::ArgListTy Args;
   TargetLowering::ArgListEntry Entry;
   for (const SDValue &N : Op->op_values()) {
     EVT ArgVT = N.getValueType();
     Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
     Entry.Node = N;
     Entry.Ty = ArgTy;
     Entry.IsSExt = TLI.shouldSignExtendTypeInLibCall(ArgVT, SignExtend);
     Entry.IsZExt = !Entry.IsSExt;
     Args.push_back(Entry);
   }
+
+  SDValue InChain = DAG.getEntryNode();
+  SDValue TCChain = InChain;
+  const Function &F = DAG.getMachineFunction().getFunction();
+  bool isTailCall =
+      TLI.isInTailCallPosition(DAG, Op.getNode(), TCChain) &&
+      (RetTy == F.getReturnType() || F.getReturnType()->isVoidTy());
+  if (isTailCall)
+    InChain = TCChain;
   CLI.setDebugLoc(SDLoc(Op))
-      .setChain(DAG.getEntryNode())
-      .setLibCallee(CallingConv::C, RetVT.getTypeForEVT(*DAG.getContext()),
-                    Callee, std::move(Args))
-      .setTailCall(true)
+      .setChain(InChain)
+      .setLibCallee(CallingConv::C, RetTy, Callee, std::move(Args))
+      .setTailCall(isTailCall)
       .setSExtResult(SignExtend)
       .setZExtResult(!SignExtend)
       .setIsPostTypeLegalization(true);
   return TLI.LowerCallTo(CLI).first;
 }
 
 SDValue PPCTargetLowering::lowerLibCallBasedOnType(
     const char *LibCallFloatName, const char *LibCallDoubleName, SDValue Op,

llvm/test/CodeGen/PowerPC/lower-intrinsics-afn-mass_notail.ll

+; RUN: llc -enable-ppc-gen-scalar-mass -O3 -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s
+; RUN: llc -enable-ppc-gen-scalar-mass -O3 -mtriple=powerpc-ibm-aix-xcoff < %s | FileCheck %s
+
+define void @cos_f64(double* %arg) {
+; CHECK-LABEL: cos_f64
+; CHECK: __xl_cos
+bb:
+  %i = bitcast double* %arg to i8*
+  %i1 = getelementptr i8, i8* %i, i64 undef
+  br label %bb2
+
+bb2:
+  %i3 = getelementptr inbounds i8, i8* %i1, i64 undef
+  %i4 = bitcast i8* %i3 to double*
+  store double undef, double* %i4, align 8
+  %i5 = getelementptr inbounds i8, i8* %i1, i64 0
+  %i6 = bitcast i8* %i5 to double*
+  %i7 = tail call afn double @llvm.sqrt.f64(double undef)
+  %i8 = fmul afn double undef, 0x401921FB54442D28
+  %i9 = tail call afn double @llvm.cos.f64(double %i8) #2
+  %i10 = fmul afn double %i7, %i9
+  store double %i10, double* %i6, align 8
+  br label %bb2
+}
+
+define void @log_f64(double* %arg) {
+; CHECK-LABEL: log_f64
+; CHECK: __xl_log
+bb:
+  %i = bitcast double* %arg to i8*
+  %i1 = getelementptr i8, i8* %i, i64 undef
+  br label %bb2
+
+bb2:
+  %i3 = getelementptr inbounds i8, i8* %i1, i64 undef
+  %i4 = bitcast i8* %i3 to double*
+  store double undef, double* %i4, align 8
+  %i5 = getelementptr inbounds i8, i8* %i1, i64 0
+  %i6 = bitcast i8* %i5 to double*
+  %i7 = tail call afn double @llvm.sqrt.f64(double undef)
+  %i8 = fmul afn double undef, 0x401921FB54442D28
+  %i9 = tail call afn double @llvm.log.f64(double %i8) #2
+  %i10 = fmul afn double %i7, %i9
+  store double %i10, double* %i6, align 8
+  br label %bb2
+}
+
+declare double @llvm.sqrt.f64(double)
+declare double @llvm.cos.f64(double)
+declare double @llvm.log.f64(double)

llvm/test/CodeGen/PowerPC/lower-intrinsics-fast-mass_notail.ll

+; RUN: llc -enable-ppc-gen-scalar-mass -O3 -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s
+; RUN: llc -enable-ppc-gen-scalar-mass -O3 -mtriple=powerpc-ibm-aix-xcoff < %s | FileCheck %s
+
+define void @cos_f64(double* %arg) {
+; CHECK-LABEL: cos_f64
+; CHECK: __xl_cos_finite
+bb:
+  %i = bitcast double* %arg to i8*
+  %i1 = getelementptr i8, i8* %i, i64 undef
+  br label %bb2
+
+bb2:
+  %i3 = getelementptr inbounds i8, i8* %i1, i64 undef
+  %i4 = bitcast i8* %i3 to double*
+  store double undef, double* %i4, align 8
+  %i5 = getelementptr inbounds i8, i8* %i1, i64 0
+  %i6 = bitcast i8* %i5 to double*
+  %i7 = tail call fast double @llvm.sqrt.f64(double undef)
+  %i8 = fmul fast double undef, 0x401921FB54442D28
+  %i9 = tail call fast double @llvm.cos.f64(double %i8) #2
+  %i10 = fmul fast double %i7, %i9
+  store double %i10, double* %i6, align 8
+  br label %bb2
+}
+
+define void @log_f64(double* %arg) {
+; CHECK-LABEL: log_f64
+; CHECK: __xl_log_finite
+bb:
+  %i = bitcast double* %arg to i8*
+  %i1 = getelementptr i8, i8* %i, i64 undef
+  br label %bb2
+
+bb2:
+  %i3 = getelementptr inbounds i8, i8* %i1, i64 undef
+  %i4 = bitcast i8* %i3 to double*
+  store double undef, double* %i4, align 8
+  %i5 = getelementptr inbounds i8, i8* %i1, i64 0
+  %i6 = bitcast i8* %i5 to double*
+  %i7 = tail call fast double @llvm.sqrt.f64(double undef)
+  %i8 = fmul fast double undef, 0x401921FB54442D28
+  %i9 = tail call fast double @llvm.log.f64(double %i8) #2
+  %i10 = fmul fast double %i7, %i9
+  store double %i10, double* %i6, align 8
+  br label %bb2
+}
+
+declare double @llvm.sqrt.f64(double)
+declare double @llvm.cos.f64(double)
+declare double @llvm.log.f64(double)