cache miss ratio for intel core i3

[tanvis]

I have intel core i3 processor. I want to know the formulae for calculating cache miss rate for different level of caches i.e L1 L2 and L3. I have referred to the previous threads but since there are variants of formulae for the same i'm a little confused.

[Peter_W_Intel]

If your i3 is Nehalem processor, please refer to this article to know formulas that you want.

[tanvis]

Hi Peter
Thanks for your reply.
Yes my i3 330M is Nehalem Processor.But the link that you gave mentions that we can estimate the % of cycles due to long latency data access.
However i would like to know formulae for calculating L1/L2/L3 cache misses.
Are the formulae mentioned below correct?
1. L1: L1D_CACHE_LD.I_STATE / L1D_CACHE_LD. MESI
2. L2: (MEM_LOAD_RETIRED.LLC_UNSHARED_HIT + MEM_LOAD_RETIRED.OTHER_CORE_L2_HIT_HITM) / L2_RQSTS.LOADS
3. L3: MEM_LOAD_RETIRED.LLC_MISS / (MEM_LOAD_RETIRED.LLC_UNSHARED_HIT + MEM_LOAD_RETIRED.OTHER_CORE_L2_HIT_HITM)

[Peter_W_Intel]

Quoting tanvis

1. L1: L1D_CACHE_LD.I_STATE / L1D_CACHE_LD. MESI
2. L2: (MEM_LOAD_RETIRED.LLC_UNSHARED_HIT + MEM_LOAD_RETIRED.OTHER_CORE_L2_HIT_HITM) / L2_RQSTS.LOADS
3. L3: MEM_LOAD_RETIRED.LLC_MISS / (MEM_LOAD_RETIRED.LLC_UNSHARED_HIT + MEM_LOAD_RETIRED.OTHER_CORE_L2_HIT_HITM)

The optimization guideline I mentioned last time, is to estimate the performanceimpact on program. I guess that you need the formulas to know "miss rates".

Above formulas are good in my view, you can use anythreshold - for example, .2 tojudge the result is good or bad.

Regard, Peter

[tanvis]

Referring to the threadhttp://software.intel.com/en-us/forums/showthread.php?t=71832i found

L1 data cache miss rate= MEM_LOAD_RETIRED.L1D_MISS/

INST_RETIRED.ANY

For L2 data cache miss = MEM_LOAD_RETIRED.L2_MISS event/

INST_RETIRED.ANY

Are these for core 2 duo processors?

Please also mention what is the difference between MEM_LOAD_RETIRED.L1D_MISS event, and L1D_CACHE_LD.I_STATE event.

[Peter_W_Intel]

L1D_CACHE_LD counts L1D read and store, but MEM_LOAD_RETIRED.L1D counts L1D read only.

The ratio is defined by user,Misses can be divided by L1/L2/L3 access, INST_RETIRED, (with penalty) CPU_CLK_UNHALED, MEM_INST_RETIRED.LOADS, etc.

It depends onyour needs.

Regards, peter

[tanvis]

Thanks for reply again.
Further i would like to know if i can measure L1/L2/L3 cache bandwidth using vtune?

[Peter_W_Intel]

You are welcome:-)

There is no predefined L1/L2/L3 bandwidth analysis in VTune.

Thereispredefined "memory bandwidth" analysis in VTune.

Regards, Peter