[llvm] [AIX][TLS] Optimize the -maix-small-local-exec-tls local-exec access sequence for non-zero offsets (PR #71485)

[Amy Kwan via llvm-commits]

================
@@ -7567,8 +7567,63 @@ static void reduceVSXSwap(SDNode *N, SelectionDAG *DAG) {
   DAG->ReplaceAllUsesOfValueWith(SDValue(N, 0), N->getOperand(0));
 }
 
+// For non-TOC-based local-exec access where an addi is feeding into another
+// addi, fold this sequence into a single addi if possible.
+static void foldADDIForLocalExecAccesses(SDNode *N, SelectionDAG *DAG) {
+  const PPCSubtarget &Subtarget =
+      DAG->getMachineFunction().getSubtarget<PPCSubtarget>();
+  // This optimization is only performed for non-TOC-based local-exec accesses.
+  if (!Subtarget.hasAIXSmallLocalExecTLS())
+    return;
+
+  if (N->getMachineOpcode() != PPC::ADDI8)
+    return;
+
+  // InitialADDI is the addi feeding into N (also an addi), and the addi that
+  // we want optimized out.
+  SDValue InitialADDI = N->getOperand(0);
+  if (!InitialADDI.isMachineOpcode() ||
+      (InitialADDI.getMachineOpcode() != PPC::ADDI8))
+    return;
+
+  // The first operand of the InitialADDI should be the thread pointer.
+  // This transformation is only performed if the first operand of the
+  // addi is the thread pointer.
+  SDValue TPRegNode = InitialADDI.getOperand(0);
+  RegisterSDNode *TPReg = dyn_cast_or_null<RegisterSDNode>(TPRegNode.getNode());
+  if (!TPReg || (TPReg->getReg() != Subtarget.getThreadPointerRegister()))
+    return;
+
+  // The second operand of the InitialADDI should be the global TLS address
+  // (the local-exec TLS variable). We only perform the folding if the TLS
+  // variable is the second operand.
+  SDValue TLSVarNode = InitialADDI.getOperand(1);
+  GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(TLSVarNode);
+  if (!GA)
+    return;
+
+  // The local-exec TLS variable should only have the MO_TPREL_FLAG target flag,
+  // so this optimization is not performed otherwise if the flag is not set.
+  unsigned TargetFlags = GA->getTargetFlags();
+  if ((TargetFlags & PPCII::MO_TPREL_FLAG) == 0)
+    return;
+
+  // The second operand of the addi that we want to preserve will be an
+  // immediate. We add this immediate, together with the address of the TLS
+  // variable found in InitialADDI, in order to preserve the correct TLS address
+  // information during assembly printing.
+  int Offset = N->getConstantOperandVal(1);
----------------
amy-kwan wrote:

Do we need to check if the offset is non-zero?
If not, state in the comments.

https://github.com/llvm/llvm-project/pull/71485