[llvm] [LoongArch] Support parsing the `%le_{hi20,add,lo12}_r` modifiers (PR #99485)

[Lu Weining via llvm-commits]

================
@@ -346,6 +359,37 @@ void LoongArchMCCodeEmitter::expandToVectorLDI(
   support::endian::write(CB, Binary, llvm::endianness::little);
 }
 
+void LoongArchMCCodeEmitter::expandAddTPRel(const MCInst &MI,
+                                            SmallVectorImpl<char> &CB,
+                                            SmallVectorImpl<MCFixup> &Fixups,
+                                            const MCSubtargetInfo &STI) const {
+  MCOperand Rd = MI.getOperand(0);
+  MCOperand Rj = MI.getOperand(1);
+  MCOperand Rk = MI.getOperand(2);
+  MCOperand Symbol = MI.getOperand(3);
+  assert(Symbol.isExpr() &&
+         "Expected expression as third input to TP-relative add");
+
+  const LoongArchMCExpr *Expr = dyn_cast<LoongArchMCExpr>(Symbol.getExpr());
+  assert(Expr &&
+         Expr->getKind() == LoongArchMCExpr::VK_LoongArch_TLS_LE_ADD_R &&
+         "Expected %le_add_r relocation on TP-relative symbol");
+
+  // Emit the correct %le_add_r relocation for the symbol.
+  Fixups.push_back(MCFixup::create(
+      0, Expr, MCFixupKind(LoongArch::fixup_loongarch_tls_le_add_r),
+      MI.getLoc()));
+
----------------
SixWeining wrote:

Shall we leave a TODO for emitting fixup_loongarch_relax when linker relaxation is enabled?

https://github.com/llvm/llvm-project/pull/99485