When coding for inference and targeting the DSP block, it is recommended to use signed arithmetic and it is a requirement to have one extra bit of width for the pre-adder result so that it can be packed into the DSP block.
When coding for inference and targeting the DSP block, it is recommended to use signed arithmetic and it is a requirement to have one extra bit of width for the pre-adder result so that it can be packed into the DSP block.