Difference between revisions of "Intel VTune Tutorial/Analysis Types"

From HPC Wiki
Intel VTune Tutorial/Analysis Types
Jump to navigation Jump to search
 
(2 intermediate revisions by 2 users not shown)
Line 10: Line 10:
 
=== Video === <!--T:5-->
 
=== Video === <!--T:5-->
  
−
<youtube width="600" height="340" right>5hQcd3MQpBw</youtube>
+
<youtube width="600" height="340" right>ghFn5IBzjrc</youtube>
  
 
([[Media:VTune_analysis_types_compressed.pdf |Slides as pdf]])
 
([[Media:VTune_analysis_types_compressed.pdf |Slides as pdf]])
Line 51: Line 51:
 
|| Not correct. Please make sure to not melt your CPU.
 
|| Not correct. Please make sure to not melt your CPU.
 
- A directed graph of a hotspots call stack. Functions are nodes and weighted edges encode the execution time
 
- A directed graph of a hotspots call stack. Functions are nodes and weighted edges encode the execution time
−
|| Not correct. This sounds like something you would want to use gprof2dot for, though!
+
|| Not correct. This sounds like something you would want to use [[Gprof Tutorial | gprof2dot]] for, though!
 
</quiz>
 
</quiz>
 
{{hidden end}}
 
{{hidden end}}
Line 62: Line 62:
 
|| Correct.
 
|| Correct.
 
- Yes, by calling <source enclose="none">vtune /path/to/source.cxx</source> from the commandline
 
- Yes, by calling <source enclose="none">vtune /path/to/source.cxx</source> from the commandline
−
|| Not correct. The
+
|| Not correct.
 
- No, VTune only analyses binaries
 
- No, VTune only analyses binaries
 
|| Not correct. The association between binaries and the source code is possible and very useful.
 
|| Not correct. The association between binaries and the source code is possible and very useful.

Latest revision as of 12:09, 15 July 2022

Tutorial
Title: Intel VTune Tutorial
Provider: HPC.NRW

Contact: tutorials@hpc.nrw
Type: Multi-part video
Topic Area: Performance analysis
License: CC-BY-SA
Syllabus

1. Introduction
2. CPU Architecture
3. Analysis Types
4. Useful Tips

The third Intel VTune tutorial covers a couple of important analysis types and shows their results. The hotspots analysis is discussed in detail and can tell you where your application spends most of its time. You can go into more detail with the threading analysis, microarchitecture exploration, or HPC performance characterization, each focussing of a specific topic.

Video

(Slides as pdf)

Quiz

1. What is a "hotspot"?

A burnt spot on the CPU, caused by a badly fitted cooler
A code segment that is very inefficient
A code segment where the program spends most of its time

2. What does the Bottom-Up tab show?

Low level performance results that are close to the hardware ("bottom")
Time spent in each code section, with the quickest sections at the top
List of code sections (functions, loops) with their attributed measurements

3. What is a Flame Graph?

A graphical presentation of the call stack on a timeline
A statistic of dead CPUs from ill fitted coolers
A directed graph of a hotspots call stack. Functions are nodes and weighted edges encode the execution time

4. Can VTune present source code with the performance of each line?

Yes, by clicking on the function name in any of the tabs
Yes, by calling vtune /path/to/source.cxx from the commandline
No, VTune only analyses binaries

5. How do analysis types differ?

All analysis types are exactly the same
There is exactly one analysis type for each high level metric (Memory, Back-end, Front-end, etc.)
Analysis types may collect similar data, but the presentation focuses on a certain topic

Retrieved from "https://hpc-wiki.info/hpc/index.php?title=Intel_VTune_Tutorial/Analysis_Types&oldid=4894"