03-17-2014 01:38 PM
I am implementing and doing a place and route on a design that is about 79% utilized. Even thought it seems like that it is going through, but it is taking for ever. The MAP itself took about 3 hours, and the place and route is still going, but has been 1.5 hour already !!!
I have worked with the same version of ISE before, and I even worked on designs with slightly higher utilization (~85%). But I had never waited this long before. The waitest that I have ever waited for a MAP in the past never exceeded 1.5 hour !
Is this normal? Am I doing something wrong, or has anyone expereinced this much wait time before ?!
I also installed speed files, and I am targeting SPLX75 !
03-17-2014 02:43 PM
typically long runtimes are caused by congestion (too dense routing), high utilization, high performance or constraints problems.
To debug, I would recommend to run the following tests to rule out the possible causes:
1) for congestion, try without I/O LOC constraints
2) for high utilization, try targeting a larger device
3) for high performance, under-constrain
4) for constraints issues, remove all timing constraints
I'm looking forward to the results of these tests.
03-17-2014 03:09 PM
I beleive something else is happening.
Cause this is a project that my co-worker has worked on, and now I am taking it over. He put all the project on a network, and all I did, I installed 12.4 on my machine, and just took his project as is, and tried to run it.
When I look at his map report and timing report, he got his stuff done in less than 40 minutes for mapping and less than 45 minutes for routing. He used a tcl script to run the project. So, I used all the setting just how had it, and for some reason mine took lot longer, and at the end, I even got timing errors !!!
Could running Xilinx 12.4 on a different machine cause this behavior ?
03-17-2014 05:26 PM
I've seen similar symptoms when the computer doesn't have enough physical memory installed. If you're running Windows, you can open up the Task Manager and look at the "performance" tab to see memory usage. Even if you have a lot of physical memory, you can run into problems if you're running other programs at the same time.
03-17-2014 05:43 PM
I really don't know what is going on.
I reinstall all the speed files for 12.4 package, and in fact I was checking the memory usage while ISE was running. My CPU usage was fairly low (10%), and my ram usage was about (5.6G used out of 8G), when checked with Task Manager. And, I have an i7 core. Are these indicative of any sort of red flag?
I know that if I hadn't installed the speed files, I will run into problem. And, I went ahead and reinstall all the speed files again, and rebooted my computer, and still the same thing.
There is absoltely nothing different between the project that is run on my co-worker's machine and the one that is being run on my machine !!! He just kicked off another run on his machine, and the same project took him only 40 minutes to complete MAP.
Any clue what could be wrong on my machine ?!?
03-17-2014 06:20 PM
That's quite a bit of RAM usage. The low CPU usage may just be an indication of the number of processor cores. i.e. Windows reports 1 of 8 cores running 100 % as "12.5%". ISE 12.x did not have multithreading IIRC.
Are you seeing a lot of disk activity while the Map is running? That could indicate that you don't have enough physical memory and you are going to disk for virtual memory. That would also cause the low CPU usage number if you don't happen to have 8 or more cores.
Also I've noted in the past that since ISE is only running on a single CPU, there are different qualities to run ISE fast than to run general software fast. Mostly you need a lot of cache per processor and that often comes with a CPU that has fewer cores like my Core2 duo which outperforms an 8-core Xeon running Map in ISE 12.x And you need a lot of physical RAM freferably with a very high speed front-side bus. Map especially uses a lot of memory and its access is very random, causing a lot of cache misses when you don't have a fairly large cache.
You didn't describe your co-worker's machine. What was it?
03-17-2014 06:49 PM
Well he has an older computer that is why he doesn't directly run his project on his local machine, and he runs them on a Server.
I see a lot of Disk activity, but not used by MAP, mostly system. Memory Usage for MAP is about ~ 1.6GB. And the memory Cached is about 2.3GB. Do you still think this might be related to the strange behavior that I see ?!
03-17-2014 10:31 PM - edited 03-17-2014 11:08 PM
I installed the following patch:
My MAP time went down from 2 hours 45 minutes to 1 hour and 50 minutes.. (Almost one hour improvement).
My Place & Route time process went from 1 hour 45 minutes to ONLY 6 minutes .
The PAR time looks okay, and it is very close to what co-worder got on his machine (~ 7 minutes).
However, my machine is still off on the MAP time. He performs the MAP in 45 minutes, and I am still an hour over !!!
I have never seen something like this before. Do I still need to update something else ?
My Map report is VERY close to his, and we both have the exact set of flags set:
Slice Logic Distribution:
Number of occupied Slices: 9,196 out of 11,662 78%
Number of LUT Flip Flop pairs used: 30,519
Number with an unused Flip Flop: 11,966 out of 30,519 39%
Number with an unused LUT: 3,019 out of 30,519 9%
Number of fully used LUT-FF pairs: 15,534 out of 30,519 50%
Number of unique control sets: 952
Number of slice register sites lost
to control set restrictions: 2,953 out of 93,296 3%
Slice Logic Distribution:
Number of occupied Slices: 9,160 out of 11,662 78%
Number of LUT Flip Flop pairs used: 30,454
Number with an unused Flip Flop: 11,958 out of 30,454 39%
Number with an unused LUT: 2,865 out of 30,454 9%
Number of fully used LUT-FF pairs: 15,631 out of 30,454 51%
Number of unique control sets: 953
Number of slice register sites lost
to control set restrictions: 2,958 out of 93,296 3%
I am still not sure, how the same project having the same MAP and SYNT flags, will result in slightly different result, and about an hour difference in MAPing time.
By the way it is the "Global Placement" that takes that takes up almost the extra hour.
Please provide some hint.
03-18-2014 07:11 AM
It's possible that there is something still different. You can use the "Project Browser" to compare the two projects, assuming you have access to your co-worker's project on the network.
03-18-2014 07:33 AM - edited 03-18-2014 07:46 AM
I used WinMerge, on the entire Directory/Sub-Directories/Files in the project that he uses vs mine. There is not a single difference.
Plus, the fact that purely by installing the patch and with absotlue no change in the source code, my MAP run time went down with about an hour that tells me something has not completely been installed yet.
03-18-2014 07:53 AM
We both use 0x64; but what happend to me after installing the patch tells me that still something is miyssing. If I just put him (co-worder) out of picture for a second, I know that by installing a patch on my machine, and without changin anything else I improved the MAP time by about an hour. This suggests I can still bring down the MAP by another installation, cause I know he has it around 45 minutes (about an hour less than mine).