Inferring DSP pre-adders and post-adders in VHDL

[roberto_udc]

Hello, I have isolated my problem in a very simple piece of code:

entity pruebas is

   port( a, b : in unsigned(26 downto 0);

   b : in unsigned(7 downto 0);     

   res : out unsigned(53 downto 0));

end pruebas;

architecture RTL of pruebas is

 begin

   res <= a * (b + c);

end RTL;

I expected that Quartus Prime 16.1 would add “c” using the pre-adder before the multiplier in the DSP, but it’s using LUTs instead.

A very similar example (res <= (a*b) + c) also uses LUTs instead of a accumulator/adder at the end of the DPS. In this case, I made c larger, in the range of 30 bits. 

Is there any trick or condition to fulfill in order that Quartus implements those sums using the available resources inside the DSP block? All user guides I've checked so far explain the capabilities of the devices, but I cannot find any actual implementation example. 

I'm targeting an Arria 10 device, btw. 

Thanks in advance,

Roberto

[RichardTanSY_Intel]

Hi @roberto_udc 

You may checkout this similar post with a detailed answer. Let me know if it helps. 

https://electronics.stackexchange.com/questions/363915/avoid-using-dsps-in-quartus-prime

Best Regards,
Richard Tan

p/s: If any answer from the community or Intel support are helpful, please feel free to give Kudos. 

[roberto_udc]

Unfortunately, my problem is slightly more complex and, somehow, the opposite one. Quartus uses one DSP to implement the product, but it wastes the possibility of implementing the addition inside the DSP.

I’ve tried this:

    attribute multstyle of res : signal is "dsp";      

But Quartus uses 1 DSP plus 17 ALM. The latter could be saved, in my opinion.

I think there must be a way to do it. I’ve tried with Vivado and the addition is performed inside the DSP, that is, one DSP and no LUTs. There should be a way to suggest Quartus to do the same.

Best

[RichardTanSY_Intel]

Have you checkout the KDB below and see if the VHDL template works for you. 

https://www.intel.com/content/www/us/en/support/programmable/articles/000079100.html

Best Regards,
Richard Tan

p/s: If any answer from the community or Intel support are helpful, please feel free to give Kudos.