- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Greetings,
I am running i7 Haswell architecture and I don't see "call stack information" in analysis as I had for Core2 in VTune.
Am I doing something wrong, or this architecture simply doesn't support this in hardware counters?
My config: Win 8.1 x64, project c++0x x64, ICC 14 SP1 update 2, VTune the latest to date.
링크가 복사됨
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
iliyapolak wrote:
Maybe your code has inlined functions?
Yes, indeed, and IPO is AFAIK inline-ing functions where it is possible.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Though ICC will NOT inline functions if there are inline specifications, when the inlineing will be counter-productive.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
I mean the case when I use VTuune inside of MSVC 2013 integration. I get for my Haswell few analysis, but in GUI there are no "collect stacks",
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
My comments are:
1. If you are using general-exploration analysis and want to see call stack info, you have to add option "-knob enable-stack-collect=true". This is not a default enabled option.
2. You can use optimization, but read this article.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
re 1. ICC is complaining about "/Qknob enable-stack-collect=true" being obsolete and will be removed in future
re 2. I need IPO optimization
The problem is, when I was on Core2 platform, in IDE's VTune GUI I could hit check box "collect stacks". Now I am on i7 Haswell, and I cannot do such thing. I worked around the problem by copying "advanced hostspots" to "custom analysis", and then I have option to collect stack. Inline'd functions are okay for me (when collecting stack informations).
But this is a bit tricky, I wish I could do it in IDE easily, without creating new custom analysis.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
nevermind, I have just noticed it collects stack by default.
Another problem is, "advanced hotspots" analysis stops collecting hardware events after exactly 70 seconds. In the GUI the time counter stops increasing and button to mark timeline is inactive. When the process exits, there is only first 70 seconds to examine.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Collecting stack is not default option, when you start a new analysis, for example, advanced-hotspots.
Collector will stop if you set Duration. Otherwise Collector will stop only the process exits.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
But I do think the option is "sticky." Once you select stacks in Advanced Hotspots and run a collection, the next time you open up the Advanced Hotspots analysis type, stacks will be selected. FWIW ;)
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
MrAnderson (Intel) wrote:
But I do think the option is "sticky." Once you select stacks in Advanced Hotspots and run a collection, the next time you open up the Advanced Hotspots analysis type, stacks will be selected. FWIW ;)
But where I can enable such option? I always do "Advanced Hotspots", whether I do or not optimization/ilining, I get an error without stack information.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Using option "-knob collection-detail=stack-sampling" for advanced-hotspots analysis.
Using option "-g -O0" to disable inline function, or using "-O2 -g -inline-debug-info" enables inline function and keep symbol info (function name in report).
