[llvm] [SPARC] Use hardware byteswapper when we have V9 (PR #191720)

[Sergei Barannikov via llvm-commits]

================
@@ -3218,13 +3255,109 @@ SDValue SparcTargetLowering::PerformBITCASTCombine(SDNode *N,
   return SDValue();
 }
 
+SDValue SparcTargetLowering::PerformBSWAPCombine(SDNode *N,
+                                                 DAGCombinerInfo &DCI) const {
+  SDLoc DL(N);
+  SelectionDAG &DAG = DCI.DAG;
+  SDValue Op0 = N->getOperand(0);
+  EVT Op0VT = N->getValueType(0);
+  bool IsLittleEndian = DAG.getDataLayout().isLittleEndian();
+
+  // Turn BSWAP (LOAD) -> ld*a #ASI_P(_L) on V9.
+  if (Subtarget->isV9() && ISD::isNormalLoad(Op0.getNode()) &&
+      Op0.getNode()->hasOneUse() &&
+      (Op0VT == MVT::i16 || Op0VT == MVT::i32 ||
+       (Subtarget->is64Bit() && Op0VT == MVT::i64))) {
+    SDValue Load = Op0;
+    LoadSDNode *LD = cast<LoadSDNode>(Load);
+
+    // Create the byte-swapping load.
+    SDValue Ops[] = {
+        LD->getChain(),         // Chain
+        LD->getBasePtr(),       // Ptr
+        DAG.getValueType(Op0VT) // VT
+    };
+
+    SDValue BSLoad = DAG.getMemIntrinsicNode(
+        IsLittleEndian ? SPISD::LOAD_BIG : SPISD::LOAD_LITTLE, DL,
+        DAG.getVTList(Op0VT == MVT::i64 ? MVT::i64 : MVT::i32, MVT::Other), Ops,
+        LD->getMemoryVT(), LD->getMemOperand());
+
+    // If this is an i16 load, insert the truncate.
+    SDValue ResVal = BSLoad;
+    if (Op0VT == MVT::i16)
+      ResVal = DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, BSLoad);
+
+    // First, combine the bswap away.  This makes the value produced by the
+    // load dead.
+    DCI.CombineTo(N, ResVal);
+
+    // Next, combine the load away, we give it a bogus result value but a real
+    // chain result.  The result value is dead because the bswap is dead.
+    DCI.CombineTo(Load.getNode(), ResVal, BSLoad.getValue(1));
+
+    // Return N so it doesn't get rechecked!
+    return SDValue(N, 0);
----------------
s-barannikov wrote:

Can this just be
```suggestion
    return DCI.CombineTo(N, ResVal);
```


https://github.com/llvm/llvm-project/pull/191720