Check out my first novel, midnight's simulacra!
Performance Counters: Difference between revisions
From dankwiki
No edit summary |
No edit summary |
||
| Line 5: | Line 5: | ||
* This is the preferred way we get at [[x86 timing|TSC]] (TSC has fallen out of favor, though) | * This is the preferred way we get at [[x86 timing|TSC]] (TSC has fallen out of favor, though) | ||
* Collectors include [[oprofile]] and [[perf]]. | * Collectors include [[oprofile]] and [[perf]]. | ||
{| border="1" | |||
! Counter | |||
! Introduced | |||
! Description | |||
|- | |||
| MACHINE_CLEARS.ASSIST_FUSION | |||
| Core i7 | |||
| Number of macro-fused ops retired | |||
|- | |||
| LSD.UOPS | |||
| Core i7 | |||
| Number of μops delivered by the [[daytripper|Loop Stream Decoder]] | |||
|- | |||
|} | |||
Revision as of 01:00, 17 March 2010
- Normally driven through MSRs
- RDPMC instruction: http://www.x86.org/secrets/opcodes/rdpmc.htm
- FreeBSD includes libpmc as part of the base system
- On Linux, we have libpfm from perfmon2
- This is the preferred way we get at TSC (TSC has fallen out of favor, though)
- Collectors include oprofile and perf.
| Counter | Introduced | Description |
|---|---|---|
| MACHINE_CLEARS.ASSIST_FUSION | Core i7 | Number of macro-fused ops retired |
| LSD.UOPS | Core i7 | Number of μops delivered by the Loop Stream Decoder |
