Hey Brad,
Thanks for the suggestions.

The Protocol Trace flag looks really promising. It'd be very nice to get
everything in , but I'm just a little worried about tweaking that if these
RubyCycle counts are used for anything other than annotation. I'll take a
strong look at starting to get that implemented though.

As for the DMA problem, I haven't encountered the issue in the new round of
patches you committed (so good job!). I have the issue that I had when I
rolled back patches (which is the deadlock).

On Wed, Mar 23, 2011 at 5:07 PM, Beckmann, Brad <[email protected]>wrote:

> Hi Korey,
>
> A few comments:
>
> - The difference in time is because the sequencer prints out the RubyCycle
> count for the issue time while the tick count is the global curTick value.
>  Now that Ruby uses DPRINTFs, I think it makes sense to move all those ruby
> print outs to ticks.  I actually have that on my long todo list, but I'm
> sure I won't get to it soon.  If you want to go ahead and make the
> conversion, please do.
>
> - I'm pretty sure that Invalid range error is completely unrelated.
>  Instead that sort of error is caused when you try to print out a
> MachineType variable, but the value has not been set.  Typically that
> happens to network messages where the enqueue operations don't fill in all
> the fields.
>
> - Overall, when tracking down a deadlock issue, start with the protocol
> trace and track down what is happening with the particular address in
> question.  From there, you can typically get an idea of what to zero in on.
>
> - By the way, have you had a chance to confirm that my patches from this
> weekend fixed your previous dma problem?
>
> Brad
>
>
>
> > -----Original Message-----
> > From: [email protected] [mailto:[email protected]]
> > On Behalf Of Korey Sewell
> > Sent: Wednesday, March 23, 2011 12:43 PM
> > To: M5 Developer List
> > Subject: Re: [m5-dev] Debugging Ruby "Deadlocks"...
> >
> > This problem may be related to my earlier post i sent today about
> > "Debugging Ruby Deadlocks", but when adding the "RubyQueue" traceflag
> > for 32 cores you get this error:
> > "build/ALPHA_FS_MOESI_CMP_directory/m5.opt -d ruby_opt/ --trace-
> > flags=RubyQueue configs/example/ruby_fs.py -b fft_64t_base -n 1 ...
> > panic: Invalid range for type MachineType  @ cycle 793500
> > [MachineType_to_string:build/ALPHA_FS_MOESI_CMP_directory/mem/pro
> > tocol/MachineType.cc,
> > line 42]
> > Memory Usage: 2312860 KBytes
> > For more information see: http://www.m5sim.org/panic/f419fb7
> > Program aborted at cycle 793500
> > Aborted
> > "
> >
> > I thought this was solved a while ago by a previous patch but it seems to
> be
> > an issue again. Is it something in the SLICC that isnt being defined
> properly
> > when the core count is large? If anybody has any thoughts , please let me
> > know so we can patch it and push the changeset in the tree.
> >
> > Note: I dont get this problem when running for just 1 core.
> > --
> > - Korey
> > _______________________________________________
> > m5-dev mailing list
> > [email protected]
> > http://m5sim.org/mailman/listinfo/m5-dev
>
>
> _______________________________________________
> m5-dev mailing list
> [email protected]
> http://m5sim.org/mailman/listinfo/m5-dev
>



-- 
- Korey
_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev

Reply via email to