HBM IPC numbers about ramulator

[kevincha]

Hi Kunal,

Thanks for pointing out the issue. There was a problem on how the Ramulator maps which bits from the physical address to the channel/rank IDs. I've pushed a patch that fixes the problem to the main branch. So please check it out.

Regarding your assert, if the purpose is to make sure the function is called, shouldn't it be "assert(req.addr_vec[0] >= 0);", instead of <= 0? req.addr_vec[0] is the channel ID, which is >= 0.

Thanks,
Kevin

[kkorgaon]

Hi Kevin,

Thanks Kevin. To validate (for now) I ran this, and I do see a difference in IPC nos for HBM.
./ramulator-cputrace cputraces/429.mcf;
DDR3: 1.00000
DDR4: 1.30493
SALP: 1.22766
LPDDR3: 0.71974
LPDDR4: 0.93880
GDDR5: 1.63193
HBM: 1.86421
WideIO: 0.95743
WideIO2: 1.44096
DSARP: 0.98984
ALDRAM: 1.04933
TLDRAM: 0.77555

The assert was for debugging purposes only. Now that the issue is fixed, it can be commented out.

Thanks,
Kunal