[NVPTX] support f32x2 instructions for sm_100+ (#126337)

Lower `fadd`, `fsub`, `fmul`, and `fma` to f32x2 variants introduced in
PTX 8.6 for sm_100+. Adds a new register class for v2f32 as a b64
register in PTX. This causes other vector operations like loads and
stores to lower as .b64 instead of .v2.b32 as appropriate.

Also update test cases to use the autogenerator.

Author: Prince781

llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp

@@ -446,11 +446,18 @@ bool NVPTXDAGToDAGISel::tryUNPACK_VECTOR(SDNode *N) {
 bool NVPTXDAGToDAGISel::tryEXTRACT_VECTOR_ELEMENT(SDNode *N) {
   SDValue Vector = N->getOperand(0);
 
-  // We only care about 16x2 as it's the only real vector type we
-  // need to deal with.
   MVT VT = Vector.getSimpleValueType();
-  if (!Isv2x16VT(VT))
+  if (!(NVPTX::isPackedVectorTy(VT) && VT.getVectorNumElements() == 2))
     return false;
+
+  unsigned Opcode;
+  if (VT.is32BitVector())
+    Opcode = NVPTX::I32toV2I16;
+  else if (VT.is64BitVector())
+    Opcode = NVPTX::I64toV2I32;
+  else
+    llvm_unreachable("Unhandled packed type");
+
   // Find and record all uses of this vector that extract element 0 or 1.
   SmallVector<SDNode *, 4> E0, E1;
   for (auto *U : Vector.getNode()->users()) {
@@ -474,11 +481,11 @@ bool NVPTXDAGToDAGISel::tryEXTRACT_VECTOR_ELEMENT(SDNode *N) {
   if (E0.empty() || E1.empty())
     return false;
 
-  // Merge (f16 extractelt(V, 0), f16 extractelt(V,1))
-  // into f16,f16 SplitF16x2(V)
+  // Merge (EltTy extractelt(V, 0), EltTy extractelt(V,1))
+  // into EltTy,EltTy Split[EltTy]x2(V)
   MVT EltVT = VT.getVectorElementType();
   SDNode *ScatterOp =
-      CurDAG->getMachineNode(NVPTX::I32toV2I16, SDLoc(N), EltVT, EltVT, Vector);
+      CurDAG->getMachineNode(Opcode, SDLoc(N), EltVT, EltVT, Vector);
   for (auto *Node : E0)
     ReplaceUses(SDValue(Node, 0), SDValue(ScatterOp, 0));
   for (auto *Node : E1)
@@ -994,6 +1001,7 @@ pickOpcodeForVT(MVT::SimpleValueType VT, std::optional<unsigned> Opcode_i8,
   case MVT::i32:
   case MVT::f32:
     return Opcode_i32;
+  case MVT::v2f32:
   case MVT::i64:
   case MVT::f64:
     return Opcode_i64;