What's wrong with CoreGen 7.0: dynamic coef multiplier (port B = constant /w reloadable)

Jun 07, 2007 5 Replies
M

Hoping that the dynamic constant multiplier could save numbers of CLBs, but it's contrary. So, what's the use of it if it takes more resource than the true parallel multiplier?



Here's the settings for both:



Architect = virtex-e Inputs A=10bits, B=8bits, both unsigned and registered RPM rectangular shape Max. pipe line Output width = 10 bits Register output no CE/reset Latency = 4 For the dyn coef, don't care (stop) output during load, matter of fact the coefs just loaded once a while



Here's the footprint results for the dyn coef multiplier



110 LUT + 131 registers It speads over 9 CLBs wide, 7 CLBs tall Uses 53 CLBs

Here's the footprint results for true parallel



87LUT + 116 registers It speads over 5 CLBs wide, 10 CLBs tall Uses 37 CLBs
Vote
Twitter Facebook LinkedIn
J

The Xilinx Multiplier v9.0 from the architecture wizard has a constant-coefficient multipler. What are you using to generate the dynamic constant multipler?

Vote
Twitter Facebook LinkedIn
J

I missed that the subject line specified CoreGen 7.0. I don't know that there's an equivalent multiplier in CoreGen 9.1.01i. The same "Multiplier v9.0" came up in CoreGen that was in the Architecture Wizard. There was also no "Virtex-E" family listed in the default menu of this newer tool version.

Vote
Twitter Facebook LinkedIn
J

It looks like the basic tool setting doesn't have Virtex-E in the pick-list. Xilinx (in the form of an FAE or Austin) can tell what the last version is that natively supports the Virtex-E or if there's a way to enable it in current tools.

Have you considered building your own dynamic coefficient multiplier?

If you have a large number of multipliers, it's helpful. If you have one or two, it's probably not worth the effort. The task ends up pretty simple since you're using three memories with a total of about 37 bits to generate

3 values you add together. It's just a matter of programming those 37 LUTs through (ideally) shift registers. The intelligence to generate the coefficient stream is a few LUTs and a lot of thought.

Vote
Twitter Facebook LinkedIn

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required