[llvm] [AMDGPU] Legalize and select raw/struct_buffer_load with tfe (PR #93310)

[Matt Arsenault via llvm-commits]

================
@@ -10172,11 +10167,30 @@ SDValue SITargetLowering::lowerPointerAsRsrcIntrin(SDNode *Op,
 }
 
 // Handle 8 bit and 16 bit buffer loads
-SDValue
-SITargetLowering::handleByteShortBufferLoads(SelectionDAG &DAG, EVT LoadVT,
-                                             SDLoc DL, ArrayRef<SDValue> Ops,
-                                             MachineMemOperand *MMO) const {
+SDValue SITargetLowering::handleByteShortBufferLoads(SelectionDAG &DAG,
+                                                     EVT LoadVT, SDLoc DL,
+                                                     ArrayRef<SDValue> Ops,
+                                                     MachineMemOperand *MMO,
+                                                     bool IsTFE) const {
   EVT IntVT = LoadVT.changeTypeToInteger();
+
+  if (IsTFE) {
+    unsigned Opc = (LoadVT.getScalarType() == MVT::i8)
+                       ? AMDGPUISD::BUFFER_LOAD_UBYTE_TFE
+                       : AMDGPUISD::BUFFER_LOAD_USHORT_TFE;
+    MachineFunction &MF = DAG.getMachineFunction();
+    MachineMemOperand *OpMMO = MF.getMachineMemOperand(MMO, 0, 8);
+    SDVTList VTs = DAG.getVTList(MVT::v2i32, MVT::Other);
+    SDValue Op = getMemIntrinsicNode(Opc, DL, VTs, Ops, MVT::v2i32, OpMMO, DAG);
+    SDValue Status = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, Op,
+                                 DAG.getVectorIdxConstant(1, DL));
+    SDValue Data = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, Op,
+                               DAG.getVectorIdxConstant(0, DL));
----------------
arsenm wrote:

Can just hardcode these to getConstant with i32 instead of using getVectorIdxConstant

https://github.com/llvm/llvm-project/pull/93310