Re: MACRO_INSTS double definition on Nehalems
Andi Kleen <[email protected]>
| Newsgroups | gmane.linux.oprofile |
|---|---|
| Message-ID | <[email protected]> |
> Andi, is http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-system-programming-manual-325384.pdf the official document that should be reference for this? Tables 19-1 and Table 19-17 look to be the table to refer to for Nehalem. I would use https://download.01.org/perfmon/NHM-EP/ > > > > > $ cat events/i386/nehalem/events | grep MACRO_INSTS > > event:0xA6 counters:0,1,2,3 um:one minimum:6000 name:MACRO_INSTS > > event:0xD0 counters:0,1,2,3 um:macro_insts minimum:6000 name:MACRO_INSTS > > According to the table they are counting different events the 0xa6 event is counting the number of seculatively decoded instructions that are macro-fused. The 0xd0 is just counting speculatively executed instructions. > > It looks like westmere also suffers this same naming issue. > > Maybe change the 0xa6 event in nehalem and westemere to MACRO_INSTS_FUSED. Would that be reasonable Andi? Sounds reasonable to me. -Andi ------------------------------------------------------------------------------