ISE5.2i strange behavior in PAR (command-line)

[Jo Pletinckx]

Hi all,

I have a 4-FPGA design with nearly identical VHDL source code, so in
order to speed development, I started using the command-line tools
instead of the Xilinx IDE (which basically also runs the cmd line tools
from the shell). I've encountered strange behavior since:

1. MPPR and PAR with the same cost table generate different results on
my PC (w2k, sp3, running 5.2i sp3). This is not such a big issue, as I
wrote my own batch to do an MPPR by calling PAR with different cost
tables values.

2. PAR run from the ISE and PAR from the command line generate different
results, with identical options and input files (I actually used the
..cmd_log file to write my own batch file). More specifically, ISE
returns result A, cmd line returns B.
Strangely enough, I ran the cmdline tools on another PC, with identical
settings, and this returned result A as well.

I browsed through the Xilinx answer database for clues, and I found no.
17134, describing a similar problem. A workaround is presented, but when
I implement it (it involves settings the PL_NODIROPT env. variable to
1), I get yet another result, C.

Has anyone encountered this problem before and was able to solve it?

Thanks in advance,

Jo Pletinckx
___________________________________________________________________________

ir. Jo Pletinckx

Ghent University
Department of Information technology / INTEC-design

Sint-Pietersnieuwstraat 41
B-9000 GENT
Belgium, Europe
___________________________________________________________________________

 

[Allan Herriman]

On Thu, 11 Dec 2003 09:39:14 +0100, Jo Pletinckx
<jo_pletinckx@yahoo.com> wrote:

Hi all,

I have a 4-FPGA design with nearly identical VHDL source code, so in
order to speed development, I started using the command-line tools
instead of the Xilinx IDE (which basically also runs the cmd line tools
from the shell). I've encountered strange behavior since:

1. MPPR and PAR with the same cost table generate different results on
my PC (w2k, sp3, running 5.2i sp3). This is not such a big issue, as I
wrote my own batch to do an MPPR by calling PAR with different cost
tables values.

2. PAR run from the ISE and PAR from the command line generate different
results, with identical options and input files (I actually used the
.cmd_log file to write my own batch file). More specifically, ISE
returns result A, cmd line returns B.
Strangely enough, I ran the cmdline tools on another PC, with identical
settings, and this returned result A as well.

I browsed through the Xilinx answer database for clues, and I found no.
17134, describing a similar problem. A workaround is presented, but when
I implement it (it involves settings the PL_NODIROPT env. variable to
1), I get yet another result, C.

Has anyone encountered this problem before and was able to solve it?

I've seen plenty of cases in which different PCs with supposedly
identical installations of Xilinx software produce different FPGA
images from identical source with identical options.

I've also seen examples of MPPR and SPPR producing different results
with the same cost table.

It sounds like you have just one PC, so there must be some difference
in the input to PAR (for your point #2), perhaps an environment
variable or command line option.


In most cases, the difference doesn't seem to matter. We did find
some examples of routes from one PC working while supposedly identical
routes from another PC didn't work.

The solution was to test everything in the lab, and avoid reusing the
PC configurations that are known to produce bad results.

(IIRC this was all reported to Xilinx some time ago. I think it was
happening with 4.x software.)

Regards,
Allan.

 

[Jo Pletinckx]

Allan Herriman wrote:

On Thu, 11 Dec 2003 09:39:14 +0100, Jo Pletinckx
jo_pletinckx@yahoo.com> wrote:


Hi all,

I have a 4-FPGA design with nearly identical VHDL source code, so in
order to speed development, I started using the command-line tools
instead of the Xilinx IDE (which basically also runs the cmd line tools

from the shell). I've encountered strange behavior since:

1. MPPR and PAR with the same cost table generate different results on
my PC (w2k, sp3, running 5.2i sp3). This is not such a big issue, as I
wrote my own batch to do an MPPR by calling PAR with different cost
tables values.

2. PAR run from the ISE and PAR from the command line generate different
results, with identical options and input files (I actually used the
.cmd_log file to write my own batch file). More specifically, ISE
returns result A, cmd line returns B.
Strangely enough, I ran the cmdline tools on another PC, with identical
settings, and this returned result A as well.

I browsed through the Xilinx answer database for clues, and I found no.
17134, describing a similar problem. A workaround is presented, but when
I implement it (it involves settings the PL_NODIROPT env. variable to
1), I get yet another result, C.

Has anyone encountered this problem before and was able to solve it?


I've seen plenty of cases in which different PCs with supposedly
identical installations of Xilinx software produce different FPGA
images from identical source with identical options.

I've also seen examples of MPPR and SPPR producing different results
with the same cost table.

It sounds like you have just one PC, so there must be some difference
in the input to PAR (for your point #2), perhaps an environment
variable or command line option.


In most cases, the difference doesn't seem to matter. We did find
some examples of routes from one PC working while supposedly identical
routes from another PC didn't work.

The solution was to test everything in the lab, and avoid reusing the
PC configurations that are known to produce bad results.

(IIRC this was all reported to Xilinx some time ago. I think it was
happening with 4.x software.)

Regards,
Allan.

I suspect the GUI will pass other unknown settings to the PAR tool, as I
use identical options and identical input files for both the GUI based
and cmd line based PAR run. I've opened a case with Xilinx to discuss
this further.

UPDATE: I set the environment variable PL_NODIROPT=1 on all PCs and now
the cmdline PAR on PC_A, PC_B as well as the GUI on PC_A return
identical results. Strangely enough, this result is different from the
previous ones. (i.e. I now have 3 different end results)

Best regards,

Jo