[llvm] [X86][GlobalISel] Support fp80 for G_FPTRUNC and G_FPEXT (PR #141611)

[Evgenii Kudriashov via llvm-commits]

================
@@ -781,6 +789,33 @@ bool X86LegalizerInfo::legalizeNarrowingStore(MachineInstr &MI,
   return true;
 }
 
+bool X86LegalizerInfo::legalizeFPExtAndTrunc(MachineInstr &MI,
+                                             MachineRegisterInfo &MRI,
+                                             LegalizerHelper &Helper) const {
+  assert((MI.getOpcode() == TargetOpcode::G_FPEXT ||
+          MI.getOpcode() == TargetOpcode::G_FPTRUNC) &&
+         "Only G_FPEXT and G_FPTRUNC are expected");
+  auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
+  MachinePointerInfo PtrInfo;
+  LLT StackTy = MI.getOpcode() == TargetOpcode::G_FPEXT ? SrcTy : DstTy;
+  Align StackTyAlign = Helper.getStackTemporaryAlignment(StackTy);
+  auto StackTemp = Helper.createStackTemporary(StackTy.getSizeInBytes(),
+                                               StackTyAlign, PtrInfo);
+
+  MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+  MachineFunction &MF = MIRBuilder.getMF();
+  auto *StoreMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
+                                           StackTy, StackTyAlign);
+  MIRBuilder.buildStore(SrcReg, StackTemp, *StoreMMO);
+
+  auto *LoadMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
+                                          StackTy, StackTyAlign);
+  MIRBuilder.buildLoad(DstReg, StackTemp, *LoadMMO);
----------------
e-kud wrote:

@arsenm we have conflicting FP64->FP16 lowering

https://github.com/llvm/llvm-project/blob/59b7b5b6b5c032ed21049d631eb5d67091f3a21c/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp#L8010-L8019

It doesn't hurt X86 but is it ok for the generic lowering to have FP64->FP16 conversion through scalars but for others using memory?

https://github.com/llvm/llvm-project/pull/141611