03-28-2021 11:59 PM
Dear support team,
When I ran P&R for VU19Ps which have LUT utilization about 41% and 58% by MPF in route_design with Vivado 2020.2 AR159267, Vivado spent about 20 hours to finish the tasks.
The implementation is targeting HAPS-100. Below shows the relevant statistics of 3 runs:
|FB1.uD (LUT 58%)||22h31m40s||AltSpreadLogic_high|
WNS=-5.903, TNS=-21472.145, WHS=-12.184, THS=-589.401
|FB1.uD (LUT 58%)||21h59m50s||AltSpreadLogic_medium||WNS=-5.787, TNS=-14659.709, WHS=-12.170, THS=-561.918|
|FB2.uA (LUT 41%)||18h39m26s||AltSpreadLogic_high||WNS=-1.679, TNS=-3.462, WHS=-3,590, THS=-54.571|
Please help investigate whether the runtime with MPF can be reduced further.
The P&R databases are in a customer's secure domain. If you want to check them, I need to arrange a remote session for the access.
Thanks a lot, Cyril
04-19-2021 04:49 PM
I am not the xilinx support team. And I am not sure what you mean by MPF.
But check your timing constraints, look for missing constraints, and look for especially missing multi-cycle paths. Look also for asynchronous CDCs that you do not have properly constrained. Missing multi-cycle paths and improperly constrained asynchronous CDCs are bad enough on their own, but people don't realize it can have a big impact on build times. You're using a large part but that is still a long build time.
Also, what is your build machine? How many cores? How much memory does it have? Does it have m.2 pcie nVME ssds? If windows, did you turn off virus checkers?
If that doesn't bear fruit, look at doing hierarchical design. Are you already scripting your build? Project mode or non-project mode?
05-06-2021 03:54 AM