I suggest you ask in the Simics forum.
Dave Z. wrote:
I'm sorry about that. I didn't think it would be the
problem. I checked the simics reference manual, and
v9_memory_transaction_t still uses priv:1 for
supervisor mode.
Regards,
Dave
--- Dan Gibson <degibson@xxxxxxxx> wrote:
You have been working on a modified version of GEMS
from the onset? That
would have been useful to know somewhat earlier.
Simics 3.0 has probably changed the way it marks
accesses as supervisor
mode. See the Simics Reference Manual...
Dave Z. wrote:
may_stall value is 0, so it's not the problem. I
guess
the problem is in the operate memory function,
where I
check for mem_trans->priv bit and return 0 if it
is 1.
This used to work just fine with the older Simics.
The
strange thing is that when I comment it out, none
of
the user misses get recorded, but when I leave it
as
is, none of the misses, coherence events are
recorded
(all results are 0). What else can I do to
separate
user and supervisor execution?
Thanks.
--- Dan Gibson <degibson@xxxxxxxx> wrote:
---------------------------------
That is the expected functionality. By
default,
the snoop interface(aka the observe memory
function)
is not installed.
This is getting stranger and stranger.
You can try adding more printf's throughout Ruby's
hierarchy to seewhere the operate memory function
is
returning from. You might want tocheck the
memory_transaction_t's->s.may_stall value -- If
may_stlalis consistently set to 0, then that would
be
the root of the problem.
Regards,
Dan
Dave Z. wrote:
Operate memory function was called many times,
butobserve memory function was not called at
all.---
Dan Gibson <degibson@xxxxxxxx>
wrote:--------------------------------- Add a
printf("foo\n") to ruby's simics/commands.Cin the
operate memoryfunction and the observe
memoryfunction
and recompile. If Ruby isreceiving requests,there
should be a lot of foo's printed.Dave Z. wrote:
simics> phys_mem.infoInformation about phys_mem
[classmemory-space]===============================================
Snoop device : none
Timing model : ruby0Dave.--- Dan
Gibson<degibson@xxxxxxxx> wrote: Try the
following
afterruby0.init:phys_mem0.infophys_mem.infoWhat
does
Simicsprint?Dave Z. wrote: Yes, the
SIMICS_VERSION is set to 3.0 and theGEMS_ROOTis
"..",
which is fine. After loading Ruby,
Simics runs very fast as if Ruby is not
working
(which is true because
the
results are zero).Thanks,Dave---Dan Gibson
<degibson@xxxxxxxx> wrote:
Did
you change the SIMICS_VERSION
settingsinMakefile.simics_version and did you
change
theGEMS_ROOT setting in the
two
module makefiles? After
changing,
you must perform afull
recompile.
Dave Z. wrote:
I'm using the -stall option when I'm
starting Simics.
But the results are all zero.
In
my benchmarks, I'm
using processor_bind to
bind a
thread to a
processor.
For example, the layout
of
my benchmarks areasfollows:Thread0 (assigned
tocpu0)MAGIC_BREAKPOINT;...Thread1 (assigned
tocpu1)...MAGIC_BREAKPOINT;So, when Thread0
reaches
thebreakpoint, I load
ruby and when Thread1
reaches the breakpoint, I dump
the ruby stats.
But
the results I get are all zero.
The Simics
ouput
is as follows:simics> c[cpu0]v:0x0000000000011184
p:0x001fdefd184
magic
(sethi 0x40000,
%g0)Setting new inspection cpu:cpu0simics>
load-module
rubysuccessful installation ofthe ruby timing
model.
simics> ruby0.initRuby Timing
ModeWarning:optimizations not enabled.Creating
eventqueue...Creating event queue doneCreating
system... Processors: 2Creating system doneRuby
initializationcompletesimics> c[cpu1]
v:0x00000000000130c0p:0x001d68130c0
magic
(sethi 0x40000, %g0)Setting new inspection
cpu:cpu1simics> cThis used to work with GEMS 1.3
andSimics 2.2.19.
Is there any ideas
why
it doesn't work withSimics3.0.22?Thanks a lot!---
Dan
Gibson<degibson@xxxxxxxx> wrote:
Use the -stall option on the
simics
command
line.
Dave
Z. wrote:
Hi, Is the following problem not a
GEMSissue?Thanks,Dave----- Original Message
----From:
DaveZ. <zhu_dave@xxxxxxxxx>To: gems
Users<gems-users@xxxxxxxxxxx>Sent: Tuesday,
November
21,2006 11:57:21 AMSubject: [Gems-users]
rubyresultsHi,I've upgraded my Simics version to
3.0.22.
Everything
seems to work fine except for the Ruby
results.
I'm
not sure if this is a GEMS
problem, but the
results
I
get regarding the memory
requests (misses,
coherence
events, etc.) are all zero.
After
readingmycheckpoint, I do
thefollowing:instruction-fetch-modeinstruction-fetch-traceistc-disabledstc-disablecpu-switch-time1magic-break-enableload-module
=== message truncated ===
____________________________________________________________________________________
Yahoo! Music Unlimited
Access over 1 million songs.
http://music.yahoo.com/unlimited
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
--
http://www.cs.wisc.edu/~gibson [esc]:wq!
|
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}