Quartus II Fitter advice

Hi,

This is very common problem when approaching the end of a project and appears as migratory warnings.

There are some options:

- Try design doctor from tools menu (I have tried it many times but is not efficient).

- your idea of code redesign is best. Try move as much as you can to a lower frequency clock if that is possible. Create this clock if feasible.

move as much as you can of logic to dedicated blocks(DSP or memory).

- look for any multicycle assignments. This could be very useful.

- The use of seed is good but has to be done at the very final end.

The best way is to run DSE(design Space Explorer) from tools menu and let it scan a range of seeds overnight or weekends. Remember seeds can lead

to 10% difference

Kaz

The Design Space Explorer might be useful as Kaz suggests. Also:

Tools>>Advisors>> Timing Optimisation Advisor might help

My two cents:

- Use TimeQuest with proper SDC constraints instead of the Classic Timing Analyzer

- Set Fitter effort to 'Auto Fit'

--- Quote Start ---

My two cents:

- Use TimeQuest with proper SDC constraints instead of the Classic Timing Analyzer

- Set Fitter effort to 'Auto Fit'

--- Quote End ---

Hi AVTX30,

sorry, but as far as I know FLEX10Ke Devices are not support by TimeQuest.

oops, sorry my bad. I did not read your question carefully.

I've just checked the Timing Optimisation Advisor, it looked very helpful.

Hi again,

Many thanks for all the responses. I have been studying the Design Space Explorer for the last few hours and it seems quite useful. I had read about it before but I don´t know why I hadn´t experimented with it before now.

I have read a little about the Timing Optimisation Advisor also but I haven´t quite gotten around to doing much with it yet.

Anyway, my plan is to specify a range of seeds and let the DSE run overnight. I ran it for two seeds this morning and it took 40 mins, so i am thinking of specifying the seeds as 1 - 25 or so. It should take about 7 or 8 hours I think. I´m not sure if this is even reasonable. From what I understand, the effect that a specified seed has is completely random, so specifying more seeds should give me a better chance of getting a better result? Should I specify as many seeds as possible for the time I have, or once a certain number of seeds is used, does the result not improve significantly?

Also another question I have reagrding the fitter. I noticed that when compiling my design, sometimes the fitter may take ~5 mins while after a very minor change it can take almost 7 mins. When it takes 5 mins the results are always worse (i.e. more timing warnings), while when it takes 7 mins the results are always much better. Does anybody know what the fitter is actually doing in the 2 extra minutes that it sometimes takes?

I presumed that if it didn´t meet it certain requirments that it would try some extra effort or something to that effect. But why does it not always do this if it is consistently not meeting the requirments?

I also see it prints to the screen the following:

"Info: Started fitting attempt 1 on Mon Nov 10 2008 at 15:49:57"

for 3 different attempts. Why does it do the 3 different attempts. Judging by the times specified, it does the first 2 in a matter of seconds and the the 3rd takes much longer.

Thanks for your help.

Hi,

Your estimate for seeds time and range is ok. Each seed is treated as separate compilation. There would be no programming files for which you have to choose the right seed and recompile as usual back in quartus.

In one case I got success 1 in 40 seeds

You may repeat that on any other range and results are independant.

As to the filtter having different times, this is normal. Quartus defaults to 3 attempts I believe. I wouldn't give the difference much weight. Be aware of any changes that may not be visible e.g. time stamp register value or version register values. These should not be changed once you get a good seed.

I will reconfirm that multicycle assignment can be very useful. If you use clock enable signal just assign it the appropriate number and you will see a lot of difference but don't assign multicycle on anything else as it gets out of control.

Kaz

--- Quote Start ---

...i am thinking of specifying the seeds as 1 - 25 or so. It should take about 7 or 8 hours I think. I´m not sure if this is even reasonable. From what I understand, the effect that a specified seed has is completely random, so specifying more seeds should give me a better chance of getting a better result? Should I specify as many seeds as possible for the time I have, or once a certain number of seeds is used, does the result not improve significantly?

--- Quote End ---

You never know when 1 more seed will give you a better result, so sometimes I'll run a huge number of seeds. However, it's very unlikely that more than 10 would be useful. That's the number I often use. Some Altera people will say that more like 5 seeds is plenty, but I often see the best result happen after more than 5 seeds. The variation really is random, but the amount of variation depends on the design. If your design tends to have only a small variation from seed to seed, you might find that a small number of seeds is enough to have a good chance of getting a result that is about as good as any seed would give with an infinite number of seeds.

DSE has (or at least used to have) a setting that would make it stop automatically when you get a good result. I've never used that setting. I monitor the dse\results.csv file (updated as the DSE seed sweep runs for most Quartus versions) while the seed sweep is running to see whether I have the results I want before the seed sweep is finished. Then I manually stop DSE early if I don't need any more seeds to compile. (I never open the .dse.rpt or results.csv directly while DSE is running. I make a copy and open the copy. I think I've had DSE have a problem when I had one of these files open while DSE tried to update it.)

--- Quote Start ---

There would be no programming files for which you have to choose the right seed and recompile as usual back in quartus.

--- Quote End ---

There is a DSE setting to archive the results of all compilations. I think something changed between Quartus versions for whether the Assembler will run, but you will probably have programming files in the DSE output .qar files if you have the Assembler enabled in your project settings. If DSE in your Quartus version skips the Assembler, then you could still use the DSE output .qar file to run the Assembler without recompiling.

Hi all,

Thanks for all the replies. Last night I let the DSE run and this morning found that it had meet timing on 1 of 20 seeds. So good news. I gotta change at least one thing in the design so I´ll run this again when the changes have been completed.

The DSE really is a very useful tool, so thanks to all for informing me about it.