pushq cost too many cpu times

[raidencheng]

hi all

when we use vtune to collect light-weight hotspots, we found this

Address Line Assembly CPU Time Instructions Retired

0xdd6d70 499 pushq %rbp 0.573s 1,184,000,000

how can a pushq cost such many cpu times? this make a function cpi = 2, and we want to fix it, but dont know how.

any suggestion is appretiate. Thanks.

[Hussam_Mousa__Intel_]

Hi,

In hotspot analysis, the instruction identified is the active Instruction Pointer value when the event counter overflew (in this case INSTR_RETIRED).

Typically, this points to the application function which was statistically responsible for generating lots of this event/time spent.

Typically, the latency inducing instructions are occuring prior to the identified event. if this event is in a loop with some loads for example, these maybe high latency loads.

Hope this helps,
Hussam

Note: there's a forum dedicated for Vtune Amplifier analysis and issues.