[clang-tools-extra] [llvm] [clang] [PowerPC] Implement llvm.set.rounding intrinsic (PR #67302)

[Qiu Chaofan via cfe-commits]

================
@@ -8900,6 +8900,82 @@ SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
   return FP;
 }
 
+SDValue PPCTargetLowering::LowerSET_ROUNDING(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  SDLoc Dl(Op);
+  MachineFunction &MF = DAG.getMachineFunction();
+  EVT PtrVT = getPointerTy(MF.getDataLayout());
+  SDValue Chain = Op.getOperand(0);
+
+  // If requested mode is constant, just use simpler mtfsb.
+  if (auto *CVal = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
+    uint64_t Mode = CVal->getZExtValue();
+    assert(Mode < 4 && "Unsupported rounding mode!");
+    unsigned InternalRnd = Mode ^ (~(Mode >> 1) & 1);
+    SDNode *SetHi = DAG.getMachineNode(
+        (InternalRnd & 2) ? PPC::MTFSB1 : PPC::MTFSB0, Dl, MVT::Other,
+        {DAG.getConstant(30, Dl, MVT::i32, true), Chain});
+    SDNode *SetLo = DAG.getMachineNode(
+        (InternalRnd & 1) ? PPC::MTFSB1 : PPC::MTFSB0, Dl, MVT::Other,
+        {DAG.getConstant(31, Dl, MVT::i32, true), SDValue(SetHi, 0)});
+    return SDValue(SetLo, 0);
+  }
+
+  // Use x ^ (~(x >> 1) & 1) to transform LLVM rounding mode to Power format.
----------------
ecnelises wrote:

I think we are using a at-best-effort approach. The meaning looks implementation-defined:

> The `llvm.set.rounding` intrinsic sets the current rounding mode. It is similar to C library function ‘fesetround’, however this intrinsic does not return any value and uses platform-independent representation of IEEE rounding modes.

https://github.com/llvm/llvm-project/pull/67302