[llvm] [AMDGPU] Remove `NoSignedZerosFPMath` uses (PR #178343)

[via llvm-commits]

================
@@ -2565,11 +2562,10 @@ SDValue AMDGPUTargetLowering::LowerFROUND(SDValue Op, SelectionDAG &DAG) const {
   SDLoc SL(Op);
   SDValue X = Op.getOperand(0);
   EVT VT = Op.getValueType();
+  SelectionDAG::FlagInserter Inserter(DAG, Op->getFlags());
 
   SDValue T = DAG.getNode(ISD::FTRUNC, SL, VT, X);
 
-  // TODO: Should this propagate fast-math-flags?
----------------
paperchalice wrote:

It now propagates fast-math-flags by flag inserter.

https://github.com/llvm/llvm-project/pull/178343