MOESI_CMP_directory is indeed the most robust of the protocols, but also
quite complex.
I'll see about fixing up MESI_SCMP_tiledL2bank_directory. Looks like I
may have released this a bit premature.
In a tiled CMP, MOESI_CMP_directory will always replace L1 data back to
the home L2 bank even if the home doesn't map to the "private" L2 bank.
This is fine if you either a) want to limit replication, or b) are
simulating a dancehall layout where the processors are one side of the CMP
with L2 banks on the other side.
--Mike
> Hi Calvin,
> WOW !! I got the same errors yesterday :)
>
> I can suggest 2 things:
>
> 1. If you are using the TILED protocol, use the FILE_SPECIFIED topology
> and the corresponding TILED_*** network file. I am not sure, but the
> protocol might be assuming something about the network before hand.
> Again, I am not sure, but I did have the some errors.
>
> 2. What I finally got running for my CMP setup was the
> MOESI_CMP_directory protocol. I chose this because, I do not understand
> the MESI_SCMP** protocol. I used my own network file with 8 processors
> on a chip and a particular topology.
>
> One of the stupid mistakes that I found that I was making was that the
> SIMICS_EXTRA_LIB was not pointing to the correct modules.
>
> You might want to check that also.
>
> Cheers,
> Niket.
> _______________________________________________
> 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.
>
|