Re: [Gems-users] Problems with MOESI_CMP_NUCA protocol

[Mike Marty <mikem@xxxxxxxxxxx>]

Daniele Bordes wrote:

> Hi all.  I have been trying to simulate an eight-processors single CMP 
> environment with MOESI_CMP_NUCA as cache management protocol over a 
> 256-banks L2 cache, using the file
> "NUCACOL_Procs-8_ProcsPerChip-8_L2Banks-256_Memories-8.txt" which 
> describes the cache network architecture;  the problem is that the 
> simulation seems to enter a deadlock after only two instructions.
> I have the same problem in other scenarios; the only scenario which 
> seems to work with MOESI_CMP_NUCA as cache management protocol is the 
> eight-processors single CMP environment over a 32-banks L2 cache.
>  
If you are compiling Ruby on a 32-bit machine, please ensure that the 
__32BITS__ #define in ruby/common/OptBitSet.C is defined by your 
compiler.  If this is not defined properly, it will result in errors 
with protocols that use more than 32 controllers of a single type (like 
a 256-bank L2)
--Mike