Method AddLarge

cpp/src/gandiva/decimal_ir.cc:246–267

This is pretty complex, so use CPP fns.

// This is pretty complex, so use CPP fns.
llvm::Value* DecimalIR::AddLarge(const ValueFull& x, const ValueFull& y,
                                 const ValueFull& out) {
  auto block = ir_builder()->GetInsertBlock();
  auto out_high_ptr = new llvm::AllocaInst(types()->i64_type(), 0, "out_hi", block);
  auto out_low_ptr = new llvm::AllocaInst(types()->i64_type(), 0, "out_low", block);
  auto x_split = ValueSplit::MakeFromInt128(this, x.value());
  auto y_split = ValueSplit::MakeFromInt128(this, y.value());

  std::vector<llvm::Value*> args = {
      x_split.high(),  x_split.low(), x.precision(), x.scale(),
      y_split.high(),  y_split.low(), y.precision(), y.scale(),
      out.precision(), out.scale(),   out_high_ptr,  out_low_ptr,
  };
  ir_builder()->CreateCall(module()->getFunction("add_large_decimal128_decimal128"),
                           args);

  auto out_high = ir_builder()->CreateLoad(types()->i64_type(), out_high_ptr);
  auto out_low = ir_builder()->CreateLoad(types()->i64_type(), out_low_ptr);
  auto sum = ValueSplit(out_high, out_low).AsInt128(this);
  ADD_TRACE_128("AddLarge : sum", sum);
  return sum;
}

/// The output scale/precision cannot be arbitrary values. The algo here depends on them
/// to be the same as computed in DecimalTypeSql.