Which “Core Instruction” number data is correct?

[hisengberg]

Dear Professor，
I want to get the total number of instructions spent on each core to complete a program run.

Figure 1 is the data obtained through `dumpstats.py` in the python script that comes with HotSniper. Figure 2 is the `sim.out` file automatically generated in each result folder. 

I found that the final statistics on the number of instructions per core were not the same. The statistics in the two files showed a difference of 100 instructions per core. 

May I ask why this phenomenon occurs and which data is the real core instruction numbers?

Hope to get your reply kindly.

Regards,
Walter

[anujpathania]

Hi,

On closer inspection, it seems it is an accounting mismatch inherited from original Sniper.

-- Anuj

[hisengberg]

Hi anuj,

  Thank you for your reply. Which data do you think is correct  and which data should I use?

  Hope to get your suggestions.

-- Walter

[anujpathania]

I would go with sim.out if you are interested in only accumulated data.

Since difference is quite small, I think the last incomplete scheduling interval is not accounted for in periodic dumpstats.py but is accounted in sim.out. But I am not 100% sure.

-- Anuj

[hisengberg]

Thanks a lot.