[llvm] [X86][GlobalISel] Enable scalar versions of G_UITOFP and G_FPTOUI (PR #100079)

[Matt Arsenault via llvm-commits]

================
@@ -644,6 +691,77 @@ bool X86LegalizerInfo::legalizeBuildVector(MachineInstr &MI,
   return true;
 }
 
+bool X86LegalizerInfo::legalizeFPTOUI(MachineInstr &MI,
+                                      MachineRegisterInfo &MRI,
+                                      LegalizerHelper &Helper) const {
+  MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+  auto [Dst, DstTy, Src, SrcTy] = MI.getFirst2RegLLTs();
+  unsigned DstSizeInBits = DstTy.getScalarSizeInBits();
+  const LLT s32 = LLT::scalar(32);
+  const LLT s64 = LLT::scalar(64);
+
+  // Simply reuse FPTOSI when it is possible to widen the type
+  if (DstSizeInBits <= 32) {
+    auto Casted = MIRBuilder.buildFPTOSI(DstTy == s32 ? s64 : s32, Src);
+    MIRBuilder.buildTrunc(Dst, Casted);
+    MI.eraseFromParent();
+    return true;
+  }
+
+  if (DstTy == s64) {
----------------
arsenm wrote:

You can't really directly rely on known target instruction behavior and still use the regular G_* opcodes. Other optimizations could do something assuming the poison behavior 

Also add that as a comment somewhere? 

https://github.com/llvm/llvm-project/pull/100079