Implementation

[zhang_297]

Anybody has experience skipping trce while running smartxplorer?

My design takes about 8+ hours to go through map/par. Then smartxplorer always run trce after them before picking the best strategy, which takes another 5+ hours. This causes delay to the bit file generation. I currently put a 15 hour cap on the max_time (to produce bit file overnight). Then I'll often see at the end of 15 hour, there is a strategy with much better timing score being deleted by smartxplorer because it hasn't finised trce yet.

It seems to me smartxplorer determins the best strategy based on timing score and run time. Both are availabe after par. There's really no need to wait for timing to complete before start bitgen process.

Is there a way to skip the trce step before picking the best strategy and start bitgen? Or am I using smartxplorer wrong?

[siktapany]

TRCE is mandatory while running SX as it relies on trce to pick the timing score to decide the best strategy.

[shantesh]

You could use a manual script. Something like,

#!/bin/bash
# mppr <t init> <t final>
#

design=your_design_name
i=$1

while [ $i -le $2 ]
    do
         gnome-terminal -x bash -c "source ../../../setenv; echo | date;
         echo \"Running multi-pass MAP/PAR with cost table = $i ...\";
         map -pr b -mt 2 -cm speed -global_opt on -retiming on -register_duplication on -w -logic_opt on -xe n -t $i -o mppr_$i.ncd $design.ngd;
         trce -v 250 -u 1000 -o mppr_$i.twr mppr_$i.ncd mppr_$i.pcf;
         par -ol high -xe n -w -mt 4 mppr_$i.ncd mppr_r_$i.ncd mppr_$i.pcf;
         trce -v 250 -u 1000 -o mppr_r_$i.twr mppr_r_$i.ncd mppr_$i.pcf; bash"
	 i=$(( $i + 1 ))
    done
echo "Multi-pass MAP/PAR done!"

 This script will run multiple instances of different seed values in parallel. The script can be changed to be executed sequentially (hence, less cpu load).

This is assuming that you know which options generate good timing results. I've a large design and this helps me choose which tools I want to run.

[zhang_297]

Just realized I never came back and post a status here. I filed a web case on this. Xilinx agreed running trce is un-necessary for each run before bitgen. (par result already have the timing score needed to find the best strategy) A CR has been filed but I haven't yet seen any changes in smartxplorer behavior in recent release yet.

[bwade]

Yes, the CR was filed but the change was never scheduled and the CR was closed as "Never Fix". It was determined that the change would have been too much work compared to the benefit received.