why a 17x16 multiplier (33bits out) needs two 18 bit DSP elements.

I'm not sure about the Arria family but the altera website had the following note about Stratix multipliers. Perhaps the same applies?

"In Altera® Stratix® III and IV devices, each DSP half block contains four 18x18 bit signed multipliers, but not all of them can be used during implementation. Each DSP half block has 144 input pins and 72 output pins. An independent 18x18 bit multipiler has an output width of 36 bits, so the DSP half block with 72 output pins can only fit two 18x18 bit multipliers. When creating an independent 18x18 bit multiplier, the tool will actually use two 18x18 bit elements."