Luke,
yes, I see. Still if that was the case the same thing would be happening for
the trace file produced by tester.exec and not only for thise that are
created within simics. Isn't that correct?
Irene.
-------------------------------------------------------------------
......in accordance with the prophecy.
----- Original Message -----
From: "Luke Yen" <lyen@xxxxxxxxxxx>
To: "Gems Users" <gems-users@xxxxxxxxxxx>
Sent: Wednesday, February 06, 2008 4:52 PM
Subject: Re: [Gems-users] problem with CMP and tester.exec
Irene,
If the threads are not bound to processors all the created threads may
be executing on a single processor (proc 0?), especially for a
short micro-benchmark like deque.
Luke
On Wed, 6 Feb 2008, Irene-Aggeliki Chounta wrote:
Hello Luke and thanks for the answer.
Can you elaborate on these traces? How are they gathered, and what
relevant config settings are you using? Are these transactional memory
runs? Are the threads getting created and bound to processors without
problems?
Luke
In order to gather the traces I use the ruby0.tracer-output-file command
from simics. I run the deque-tm.simics script for a MOSI_SMP_bcast
protocol,
where I have defined g_NUM_PROCESSORS 4. I haven't checked how threads
are
bound to processors though....
irene.
-------------------------------------------------------------------
......in accordance with the prophecy.
----- Original Message -----
From: "Luke Yen" <lyen@xxxxxxxxxxx>
To: "Gems Users" <gems-users@xxxxxxxxxxx>
Sent: Monday, February 04, 2008 6:00 PM
Subject: Re: [Gems-users] problem with CMP and tester.exec
On Mon, 4 Feb 2008, Irene-Aggeliki Chounta wrote:
oh, I see. Still, do you think I can use it through simics without
major
problems?
Yes, by default we don't check data values when running with
Simics, and we haven't experimentally run into any races with PUTX until
now.
and another thing, I tried to get some traces from MOSI_SMP_bcast and
MESI_CMP_filter_directory while executing barnes,deque and other
benchmarks
but I get traces only from CPU 0, although I use g_NUM_PROCESSORS=16.
Following the same methodology in gems 1.4 everything was fine. Do you
have
any ideas on this one?
Can you elaborate on these traces? How are they gathered, and what
relevant config settings are you using? Are these transactional memory
runs? Are the threads getting created and bound to processors without
problems?
Luke
-
irene.
-------------------------------------------------------------------
......in accordance with the prophecy.
----- Original Message -----
From: "Luke Yen" <lyen@xxxxxxxxxxx>
To: "Gems Users" <gems-users@xxxxxxxxxxx>
Sent: Monday, February 04, 2008 5:19 PM
Subject: Re: [Gems-users] problem with CMP and tester.exec
There are some missing u_writeDataToL1Cache actions in the
current MESI_CMP_filter_directory-L1cache.sm file. They should be
added
whenever the L1 receives data from the L2 or L1. Here's a list of
transitions that need this action:
transition(IS, L2_Data_all_Acks, S) {
u_writeDataToL1Cache;
...
transition(IS_I, L2_Data_all_Acks, S) {
u_writeDataToL1Cache;
...
transition({IS, IS_I}, L2_Data, IS_S) {
u_writeDataToL1Cache;
...
transition({IS, IS_I}, L2_Exclusive_Data, IS_E) {
u_writeDataToL1Cache;
...
transition({IS, IS_I}, L2_Exclusive_Data_all_Acks, E){
u_writeDataToL1Cache;
...
transition(IS, DataS_fromL1, S) {
u_writeDataToL1Cache;
...
transition(IS_I, DataS_fromL1, S) {
u_writeDataToL1Cache;
...
transition(IM, Data_all_Acks, M) {
u_writeDataToL1Cache;
...
transition(IM, L2_Data, IM_M) {
u_writeDataToL1Cache;
...
transition(IM, L2_Data_all_Acks, M){
u_writeDataToL1Cache;
...
Even with these fixes, this protocol can still fail the tester due
to
complex PUTX races that we are currently debugging & fixing. The next
version of this protocol will be coming out with GEMS 2.1, once we get
all
its features tested & debugged.
Luke
On Mon, 4 Feb 2008, Irene-Aggeliki Chounta wrote:
hello all.
I've recently upgraded to GEMS 2.0 and working on
MESI_CMP_filter_directory. The making of ruby for this protocol is
quite
succesful but while running the tester.exec (./tester.exec -p 16 -l
10000), I get several warning and a fatal error:
"Fatal Error: in fn void Check::performCallback(NodeID, SubBlock&) in
tester/Check.C:208: Action/check failure"
In rubyconfig.defautls I have used NUMBER_OF_VIRTUAL_NETWORK=5 and 6
but
with no luck. I also tried to test the MESI_CMP_directory which also
doesn't work and produces the same error.
Do you have any idea of what's wrong?
thank you.
-------------------------------------------------------------------
Irene.
......in accordance with the prophecy.
_______________________________________________
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.
_______________________________________________
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.
_______________________________________________
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.
_______________________________________________
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.
_______________________________________________
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.
|