- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I experienced an interesting problem while using vtune 2016. In top-down analysis method (Gerneal Exploration), vtune uses UOPS_ISSUED.ANY, UOPS_RETIRE.RETIRE_SLOTS, INT_MISC.RECOVERY_CYCLES and CPU_CLK_UNHALTED.THREAD to evaluate the effect of Bad Speculation. I created a custom analysis in vtune including above events.
What makes me suprising is that, the results of General Exploration and my custom analysis is significantly different while measuring same program. In the result of my custom analysis, UOPS_ISSUED.ANY and UOPS_RETIRE.RETIRE_SLOTS are doubled while CPU_CLK_UNHALTED.THREAD stays the same. Thus, the effect of Bad Speculation is just doubled in the result of my custom analysis. Could anyone kindly give some help and make some detailed explanations to me?
PS : The program I measured is just a single-thread application. And the events included in my custom analysis is as follows:
CPU_CLK_UNHALTED.THREAD, CPU_CLK_UNHALTED.THREAD_P, IDQ_UOPS_NOT_DELIVERD.CORE, INST_RETIRED.ANY, INST_RETIRED.ANY_P, INT_MISC.RECOVERY_CYCLES, UOPS_ISSUED.ANY, UOPS_RETIRED.RETIRE_SLOTS
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
BTW, the number of UOPS_ISSUED.ANY is half of that of INST_RETIRED.ANY in the result of Genernal Exploration. Is this a reasonable result?

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page