[Gems-users] Ruby and TM system failing on 64 core serengethi

[BYONG WU CHONG <bernard.chong@xxxxxxxx>]

Hi,

 

I am researching about transactional memory and I had this error.

 

--------------- CLI Error Start ---------------

Ruby Timing Mode

Creating event queue...

Creating event queue done

Creating system...

  Processors: 64

Creating system done

Ruby initialization complete

Runtime Error at ../protocols/MESI_CMP_filter_directory-L2cache.sm:255, Ruby Time: 9: assert failure, PID: 11341

press return to continue.

--------------- CLI Error End ---------------

 

I am using abisko or serengethi 64 core checkpoint and checkpoints with lesser than 64 cores (2, 4, 8, 16, 32) work fine.

 

I tracked ../protocols/MESI_CMP_filter_directory-L2cache.sm:255 and found out that

 

--------------- $GEMS/protocols/MESI_CMP_filter_directory-L2cache.sm Start ---------------

  void addSharer(Address addr, MachineID requestor) {

    DEBUG_EXPR(machineID);

    DEBUG_EXPR(requestor);

    DEBUG_EXPR(addr);

    assert(map_L1CacheMachId_to_L2Cache(addr, requestor) == machineID);

    L2cacheMemory[addr].Sharers.add(requestor);

  }

--------------- $GEMS/protocols/MESI_CMP_filter_directory-L2cache.sm End ---------------

 

Above bold line is causing the trouble.

I tried to debug it, but it wasn’t easy because this is SLICC code.

I couldn’t printf or save map_L1CacheMachId_to_L2Cache(addr, requestor)’s result to temp variable.

 

 

Did any of you face this type of error?

 

Thank you.

 

- Byong Wu "Bernard" Chong

bernard.chong@xxxxxxxx

bchong@xxxxxxxxxxx