I currently have the following verilog expression...

wire [15:0] address_delta = (rx_address_in * 8 + (rx_eof_in ? rx_len_in : 8)) - (seek_address + OUT_BYTES);

rx_address_in is 13 bits and OUT_BYTES is a parameter which is an integer constant (so I guess 32 bits). I am assuming the 8x multiplication will get optimized into a left shift of 3. If this is a bad assumption to make please let me know. I am using the latest Vivado.

Right now my design is having trouble meeting timing. Although my address_delta is only 16 bits, since I am using 32 bit constants inside the expression which I think might cause the operations to be carried out at a higher precision than necessary. How would you rewrite this expression for efficiency or is it fine the way it is?

I am thinking I could do something like this...

wire [15:0] seek_end = (seek_address + OUT_BYTES);
wire [15:0] address_delta = ((rx_address_in << 2'd3) + (rx_eof_in ? rx_len_in : 4'd8)) - seek_end;
  • 1
    \$\begingroup\$ Have you looked in the timing report to identify the critical path? Synthesis tools are pretty good at discarding logic that isn't needed, so I doubt that you really have identified the problem. \$\endgroup\$ – Elliot Alderson Jul 11 '18 at 19:09

No, using an intermediate assignment does not solve your problem.

HDL languages do not work like C. There still is a continuous timing path from seek_address to address_delta.

To break the path you must e.g. insert a register. But that may stop your design from working as seek_end will arrive one clock cycle later.

Your calculation will not be done in 32 bits. Maybe in a simulator but the synthesis tool will remove all superfluous bits.

You do not tell us how fast your design must be, but as Elliot Alderson already suggest in the comment: three 16 bit adders are rarely critical. (The rx_address_in << 2'd3 does not cause any delay.) Unless this expression is part of a logic cone.

  • \$\begingroup\$ I think the OP is saying that using wire[15:0] will force the intermediate result to be truncated to 16 bits, which might help meet timing since less stuff is computed. \$\endgroup\$ – immibis Jul 12 '18 at 2:06
  • \$\begingroup\$ @immibis As I said: the synthesis tool will remove all superfluous bits. \$\endgroup\$ – Oldfart Jul 12 '18 at 2:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.