Hi,
For your information, I have had some discussion with my peer on this. If you refer to the "Variable Precision DSP Block Architecture for Cyclone V Devices" figure in CV device handbook, there seems to be no direct way to bypass the multiplier when you are using DSP hard block. This would explain why when you are implementing Adder, Quartus will use ALM for it.
One possible workaround that we can think of is to leveraging the pre-adder and then multiply the sum with "1". However, since this is not something validated or documented, I cannot guarantee if the design will work and how is the impact to fmax and latency. You would need to perform further evaluation to ensure the design is meeting your expectation. Hop you can understand this.
In addition to this, when you are implementing your RTL, you might need to use the LogicLock feature to force the code into DSP block just in case Quartus implement your logic into ALM.
Please let me know if there is any concern. Thank you.
Best regards,
Chee Pin