Hi David,

thanks for the detailed echo. I wasn't very precise about the illegal
values - they're way above 9, it really is garbage, and I haven't
figured out where it came from. Unfortunately my gdb can't seem to go
further into the GNAT internals the verbose way, I guess this would need
some extensive analysis (hopefully not down to assembly).
One thing I can tell though, that the variable is previously assigned to
something legal. Some memory trashing is happening on the way to that
specific procedure.

So the main question is: Can bad VHDL code cause internal garbage? Looks
like a buffer overflow effect, but I couldn't detect that anywhere.

More comments below:


> 
> I didn't follow your regime exactly and didn't get these warnings and errors.
> 
> 
> All the 32 bit buses were always zero, the clk was running the model was 
> introducing stimulus from the looks of things, but it was running.
> 
> david-koontz-VirtualBox:/mnt/share/fptest% !so
> source script
> fptop.vhd:120:1:warning: component instance "my_test" is not bound
> fptop.vhd:51:14:warning: (in default configuration of fptop(behavioral))
> ./fp_tb:info: simulation stopped by --stop-time
> 

When you're not binding my_test, the critical code won't be entered, so
can't crash. Right? I don't understand why my_test is not bound though.
Maybe put a --std=93c flag in the ghdl -m build rule.

> The warnings were due to the architecture not being specified.  The 
> simulation stopped after 10 usec.
> 
> I didn't have access to the ISE 11.1 you used, and with a bandwidth cap and 
> having just downloaded 13.2, I'm not inclined to download another version 
> just to get the XilinxCoreLib and Unisim libraries sources to match yours.  
> I'd consider the modifications I made to the iteration schemes to be less 
> than ideal, and I'd search for something that told me the changes were 
> necessary.  That the Xilinx source has a few of these makes you wonder who's 
> right.
> 

Don't bother with v11, I am soon going to have to upgrade anyhow. Maybe
the problem then disappears. If you still would like to try
reconstructing this crash, I can also pack the relevant libraries into a
.tgz and put them somewhere. Let me know.

> I can't help but wonder what's going on in your IEEE libraries (or ghdl) that 
> you were getting those warnings and error, but have no visibility into the 
> problem from here.  I also couldn't elaborate with an mcode version and 
> couldn't successfully use ieee=synopsys when analyzing 
> floating_point_v4_0_xst_comp.vhd:
> 

I guess it's not in the ieee libs, so much I can tell from the GDB
tracing. But then again, no proof that there are some unknown side effects..

Here my version info:

GHDL 0.29 (20100109) [Sokcho edition]
 Compiled with GNAT Version: 4.4.3

About the std_logic_arith: I am aware of those issues, but I don't get
around it when importing the Xilinx code. And there's no way to rewrite
all that with numeric_std functions (moreover, Xilinx would not like
their code to be altered, let aside redistributing it).

Greetings,

- Martin

_______________________________________________
Ghdl-discuss mailing list
[email protected]
https://mail.gna.org/listinfo/ghdl-discuss

Reply via email to