View source for Profiling (computer programming)

{{Short description|Measuring the time or resources used by a section of a computer program}}
{{more citations needed|date=January 2009}}
{{Software development process|Tools}}
In [[software engineering]], '''profiling''' ('''program profiling''', '''software profiling''') is a form of [[dynamic program analysis]] that measures, for example, the space (memory) or time [[Computational complexity theory|complexity of a program]], the [[instruction set simulator|usage of particular instructions]], or the frequency and duration of function calls.  Most commonly, profiling information serves to aid [[program optimization]], and more specifically, [[performance engineering]].

Profiling is achieved by [[Instrumentation (computer programming)|instrumenting]] either the program [[source code]] or its binary executable form using a tool called a ''profiler'' (or ''code profiler''). Profilers may use a number of different techniques, such as event-based, statistical, instrumented, and simulation methods.

== Gathering program events ==
Profilers use a wide variety of techniques to collect data, including [[hardware interrupt]]s, [[Instrumentation (computer programming)|code instrumentation]], [[instruction set simulator|instruction set simulation]], operating system [[hooking|hooks]], and [[Hardware performance counter|performance counter]]s.

== Use of profilers ==
[[File:CodeAnalyst3.png|thumb|Graphical output of the [[CodeAnalyst]] profiler]]
{{quotation|text=
Program analysis tools are extremely important for understanding program behavior. Computer architects need such tools to evaluate how well programs will perform on new [[computer architecture|architectures]]. Software writers need tools to analyze their programs and identify critical sections of code. [[Compiler]] writers often use such tools to find out how well their [[instruction scheduling]] or [[branch prediction]] algorithm is performing...|author=ATOM|source=[[Conference on Programming Language Design and Implementation|PLDI]]|'94}}

The output of a profiler may be:

* A statistical ''summary'' of the events observed (a '''profile''')
:Summary profile information is often shown annotated against the source code statements where the events occur, so the size of measurement data is linear to the code size of the program.

 /* ------------ source------------------------- count */             
 0001            IF X = "A"                      0055
 0002                THEN DO                       
 0003                    ADD 1 to XCOUNT         0032
 0004                ELSE
 0005            IF X = "B"                      0055

* A stream of recorded events (a '''trace''')
:For sequential programs, a summary profile is usually sufficient, but performance problems in parallel programs (waiting for messages or synchronization issues) often depend on the time relationship of events, thus requiring a full trace to get an understanding of what is happening.
: The size of a (full) trace is linear to the program's [[instruction path length]], making it somewhat impractical. A trace may therefore be initiated at one point in a program and terminated at another point to limit the output.
* An ongoing interaction with the [[hypervisor]] (continuous or periodic monitoring via on-screen display for instance)
: This provides the opportunity to switch a trace on or off at any desired point during execution in addition to viewing on-going metrics about the (still executing) program. It also provides the opportunity to suspend asynchronous processes at critical points to examine interactions with other parallel processes in more detail.

A profiler can be applied to an individual method or at the scale of a module or program, to identify performance bottlenecks by making long-running code obvious.<ref>{{cite web| title=How to find the performance bottleneck in C# desktop application?| publisher=[[Stack Overflow]]| year=2012| url=https://stackoverflow.com/questions/13698674/how-to-find-the-performance-bottleneck-in-c-sharp-desktop-application}}</ref> A profiler can be used to understand code from a timing point of view, with the objective of optimizing it to handle various runtime conditions<ref>{{cite web| last=Krauss| first=Kirk J| title=Performance Profiling with a Focus| publisher=Develop for Performance| year=2017| url=http://www.developforperformance.com/PerformanceProfilingWithAFocus.html}}</ref> or various loads.<ref>{{cite web| work=Stackify Developer Tips, Tricks and Resources| title=What is code profiling? Learn the 3 Types of Code Profilers| publisher=Disqus| year=2016| url=https://stackify.com/what-is-code-profiling/}}</ref> Profiling results can be ingested by a compiler that provides [[profile-guided optimization]].<ref>{{cite web| last=Lawrence| first=Eric| work=testslashplain| title=Getting Started with Profile Guided Optimization| publisher=WordPress| year=2016| url=https://textslashplain.com/2016/01/10/getting-started-with-profile-guided-optimization/}}</ref> Profiling results can be used to guide the design and optimization of an individual algorithm; the [[Krauss matching wildcards algorithm]] is an example.<ref>{{cite web| last=Krauss| first=Kirk| title=Matching Wildcards: An Improved Algorithm for Big Data| publisher=Develop for Performance| year=2018| url=http://www.developforperformance.com/MatchingWildcards_AnImprovedAlgorithmForBigData.html}}</ref> Profilers are built into some [[application performance management]] systems that aggregate profiling data to provide insight into [[transaction processing|transaction]] workloads in [[distributed computing|distributed]] applications.<ref>{{cite web| work=Stackify Developer Tips, Tricks and Resources| title=List of .Net Profilers: 3 Different Types and Why You Need All of Them| publisher=Disqus| year=2016| url=https://stackify.com/three-types-of-net-profilers/}}</ref>

==History==
Performance-analysis tools existed on [[IBM/360]] and [[IBM/370]] platforms from the early 1970s, usually based on timer interrupts which recorded the [[program status word]] (PSW) at set timer-intervals to detect "hot spots" in executing code.{{citation needed|date=February 2014}} This was an early example of [[Sampling (statistics)|sampling]] (see below). In early 1974 [[Instruction Set Simulator|instruction-set simulator]]s permitted full trace and other performance-monitoring features.{{citation needed|date=February 2014}}

Profiler-driven program analysis on Unix dates back to 1973,<ref name="prof">[http://www.tuhs.org/Archive/Distributions/Research/Dennis_v4/v4man.tar.gz Unix Programmer's Manual, 4th Edition]</ref> when Unix systems included a basic tool, <code>prof</code>, which listed each function and how much of program execution time it used. In 1982 <code>gprof</code> extended the concept to a complete [[call graph]] analysis.<ref name="gprof">
S.L. Graham, P.B. Kessler, and M.K. McKusick, [http://docs.freebsd.org/44doc/psd/18.gprof/paper.pdf ''gprof: a Call Graph Execution Profiler''], Proceedings of the SIGPLAN '82 Symposium on Compiler Construction, ''[[SIGPLAN]] Notices'', Vol. 17, No 6, pp. 120-126; [[doi:10.1145/800230.806987]]</ref>

In 1994, Amitabh Srivastava and [[Alan Eustace]] of [[Digital Equipment Corporation]] published a paper describing ATOM<ref>
A. Srivastava and A. Eustace, [http://www.ece.cmu.edu/~ece548/tools/atom/man/wrl_94_2.pdf ''ATOM: A system for building customized program analysis tools''], Proceedings of the ACM SIGPLAN Conference on Programming language design and implementation (PLDI '94), pp. 196-205, 1994; ACM ''SIGPLAN Notices'' - Best of PLDI 1979-1999 Homepage archive, Vol. 39, No. 4, pp. 528-539; [[doi:10.1145/989393.989446]]
</ref> (Analysis Tools with OM). The ATOM platform converts a program into its own profiler: at [[compile time]], it inserts code into the program to be analyzed. That inserted code outputs analysis data. This technique - modifying a program to analyze itself - is known as "[[Instrumentation (computer programming)|instrumentation]]".

In 2004 both the <code>gprof</code> and ATOM papers appeared on the list of the 50 most influential [[Conference on Programming Language Design and Implementation|PLDI]] papers for the 20-year period ending in 1999.<ref>
[http://www.cs.utexas.edu/users/mckinley/20-years.html 20 Years of PLDI (1979–1999): A Selection], [[Kathryn S. McKinley]], Editor</ref>

==Profiler types based on output ==

===Flat profiler ===
Flat profilers compute the average call times, from the calls, and do not break down the call times based on the callee or the context.

===Call-graph profiler===
[[Call graph]] profilers<ref name="gprof" /> show the call times, and frequencies of the functions, and also the call-chains involved based on the callee. In some tools full context is not preserved<!--, but others can save full call tree-->.

===Input-sensitive profiler===
Input-sensitive profilers<ref name="aprof">E. Coppa, C. Demetrescu, and I. Finocchi, [https://web.archive.org/web/20180611201601/https://ieeexplore.ieee.org/document/6858059/ ''Input-Sensitive Profiling''], IEEE Trans. Software Eng. 40(12): 1185-1205 (2014); [[doi:10.1109/TSE.2014.2339825]]</ref><ref>D. Zaparanuks and M. Hauswirth, ''Algorithmic Profiling'', Proceedings of the 33rd ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI 2012), ACM SIGPLAN Notices, Vol. 47, No. 6, pp. 67-76, 2012; [[doi:10.1145/2254064.2254074]]</ref><ref>T. Kustner, J. Weidendorfer, and T. Weinzierl, ''Argument Controlled Profiling'', Proceedings of Euro-Par 2009 – Parallel Processing Workshops, Lecture Notes in Computer Science, Vol. 6043, pp. 177-184, 2010; [[doi:10.1007/978-3-642-14122-5 22]]</ref> add a further dimension to flat or call-graph profilers by relating performance measures to features of the input workloads, such as input size or input values. They generate charts that characterize how an application's performance scales as a function of its input.

==Data granularity in profiler types==
Profilers, which are also programs themselves, analyze target programs by collecting information on the target program's execution. Based on their data granularity, which depends upon how profilers collect information, they are classified as ''event-based'' or ''statistical'' profilers. Profilers interrupt program execution to collect information.  Those interrupts can limit time measurement resolution, which implies that timing results should be taken with a grain of salt. [[Basic block]] profilers report a number of machine [[cycles per instruction|clock cycles]] devoted to executing each line of code, or timing based on adding those together; the timings reported per basic block may not reflect a difference between [[CPU cache|cache]] hits and misses.<ref>{{cite web| work=OpenStax CNX Archive| title=Timing and Profiling - Basic Block Profilers| url=https://archive.cnx.org/contents/d29c016a-2960-4fc9-b431-9eda881a28f5@3/timing-and-profiling-basic-block-profilers#id6897344}}</ref><ref>{{cite journal| last1=Ball| first1=Thomas| last2=Larus| first2=James R.| journal=ACM Transactions on Programming Languages and Systems| volume=16| issue=4| pages=1319–1360| title=Optimally profiling and tracing programs| publisher=ACM Digital Library| year=1994| url=https://www.classes.cs.uchicago.edu/current/32001-1/papers/ball-larus-profiling.pdf| doi=10.1145/183432.183527| s2cid=6897138| access-date=2018-05-18| archive-url=https://web.archive.org/web/20180518195918/https://www.classes.cs.uchicago.edu/current/32001-1/papers/ball-larus-profiling.pdf| archive-date=2018-05-18| url-status=dead}}</ref>

===Event-based profilers===
Event-based profilers are available for the following programming languages:
* [[Java (programming language)|Java]]: the [[Java Virtual Machine Tools Interface|JVMTI]] (JVM Tools Interface) API, formerly JVMPI (JVM Profiling Interface), provides hooks to profilers, for trapping events like calls, class-load, unload, thread enter leave.
* [[.NET Framework|.NET]]: Can attach a profiling agent as a ''COM'' server to the ''CLR'' using Profiling ''API''. Like Java, the runtime then provides various callbacks into the agent, for trapping events like method [[Interpreter|JIT]] / enter / leave, object creation, etc. Particularly powerful in that the profiling agent can rewrite the target application's bytecode in arbitrary ways.
* [[Python (programming language)|Python]]: Python profiling includes the profile module, hotshot (which is call-graph based), and using the 'sys.setprofile' function to trap events like c_{call,return,exception}, python_{call,return,exception}.
* [[Ruby (programming language)|Ruby]]: Ruby also uses a similar interface to Python for profiling. Flat-profiler in profile.rb, module, and ruby-prof a C-extension are present.

===Statistical profilers===
These profilers operate by [[Sampling (statistics)|sampling]]. A sampling profiler probes the target program's [[call stack]] at regular intervals using [[operating system]] [[interrupt]]s. Sampling profiles are typically less numerically accurate and specific, providing only a statistical approximation, but allow the target program to run at near full speed. "The actual amount of error is usually more than one sampling period. In fact, if a value is n times the sampling period, the expected error in it is the square-root of n sampling periods."<ref>[http://www.cs.utah.edu/dept/old/texinfo/as/gprof.html#SEC12 Statistical Inaccuracy of <code>gprof</code> Output] {{webarchive|url=https://web.archive.org/web/20120529075000/http://www.cs.utah.edu/dept/old/texinfo/as/gprof.html |date=2012-05-29 }}</ref>

In practice, sampling profilers can often provide a more accurate picture of the target program's execution than other approaches, as they are not as intrusive to the target program and thus don't have as many side effects (such as on memory caches or instruction decoding pipelines). Also since they don't affect the execution speed as much, they can detect issues that would otherwise be hidden. They are also relatively immune to over-evaluating the cost of small, frequently called routines or 'tight' loops. They can show the relative amount of time spent in user mode versus interruptible kernel mode such as [[system call]] processing.

Unfortunately, running kernel code to handle the interrupts incurs a minor loss of CPU cycles from the target program, diverts cache usage, and cannot distinguish the various tasks occurring in uninterruptible kernel code (microsecond-range activity) from user code. Dedicated hardware can do better: ARM Cortex-M3 and some recent MIPS processors' JTAG interfaces have a PCSAMPLE register, which samples the [[program counter]] in a truly undetectable manner, allowing non-intrusive collection of a flat profile.

Some commonly used<ref>{{cite web| title=Popular C# Profilers| publisher=Gingtage| year=2014| url=http://www.ginktage.com/2014/10/popular-c-profilers/}}</ref> statistical profilers for Java/managed code are [[SmartBear Software]]'s [[AQtime]]<ref>{{cite web| work=AQTime 8 Reference| title=Sampling Profiler - Overview| publisher=SmartBear Software| year=2018| url=https://support.smartbear.com/viewarticle/54581/}}</ref> and [[Microsoft]]'s [[CLR Profiler]].<ref>{{cite web| work=Microsoft .NET Framework Unmanaged API Reference| last=Wenzal| first=Maira|display-authors=etal| title=Profiling Overview| publisher=Microsoft| year=2017| url=https://docs.microsoft.com/en-us/dotnet/framework/unmanaged-api/profiling/profiling-overview#supported-features}}</ref> Those profilers also support native code profiling, along with [[Apple Inc.]]'s [[Apple Developer Tools#Shark|Shark]] (OSX),<ref>{{cite web| work=[[Apple Developer Tools]]| title=Performance Tools| publisher=Apple, Inc.| year=2013| url=https://developer.apple.com/library/content/documentation/Performance/Conceptual/PerformanceOverview/PerformanceTools/PerformanceTools.html}}</ref> [[OProfile]] (Linux),<ref>{{cite web| work=IBM DeveloperWorks| last1=Netto| first1=Zanella| last2=Arnold| first2=Ryan S.| title=Evaluate performance for Linux on Power| year=2012| url=https://www.ibm.com/developerworks/linux/library/l-evaluatelinuxonpower/}}</ref> [[Intel]] [[VTune]] and Parallel Amplifier (part of [[Intel Parallel Studio]]), and [[Oracle Corporation|Oracle]] [[Performance Analyzer]],<ref>{{cite conference |last1=Schmidl |first1=Dirk |first2=Christian |last2=Terboven |first3=Dieter |last3=an Mey |first4=Matthias S. |last4=Müller |title=Suitability of Performance Tools for OpenMP Task-Parallel Programs |conference=Proc. 7th Int'l Workshop on Parallel Tools for High Performance Computing |year=2013 |pages=25–37 |isbn=9783319081441 |url=https://books.google.com/books?id=-I64BAAAQBAJ&pg=PA27}}</ref> among others.

===Instrumentation ===
This technique effectively adds instructions to the target program to collect the required information. Note that [[instrumenting]] a program can cause performance changes, and may in some cases lead to inaccurate results and/or [[heisenbug]]s.  The effect will depend on what information is being collected, on the level of timing details reported, and on whether basic block profiling is used in conjunction with instrumentation.<ref>{{cite magazine| last1=Carleton| first1=Gary| last2=Kirkegaard| first2=Knud| last3=Sehr| first3=David| title=Profile-Guided Optimizations| magazine=[[Dr. Dobb's Journal]]| year=1998| url=http://www.drdobbs.com/profile-guided-optimizations/184410561}}</ref>  For example, adding code to count every procedure/routine call will probably have less effect than counting how many times each statement is obeyed.  A few computers have special hardware to collect information; in this case the impact on the program is minimal.

Instrumentation is key to determining the level of control and amount of time resolution available to the profilers. 
* '''Manual''': Performed by the programmer, e.g. by adding instructions to explicitly calculate runtimes, simply count events or calls to measurement [[API]]s such as the [[Application Response Measurement]] standard.
* '''Automatic source level''': instrumentation added to the source code by an automatic tool according to an instrumentation policy. 
* '''Intermediate language''': instrumentation added to [[Assembly language|assembly]] or decompiled [[bytecode]]s giving support for multiple higher-level source languages and avoiding (non-symbolic) binary offset re-writing issues.
* '''Compiler assisted'''
* '''Binary translation''': The tool adds instrumentation to a compiled [[executable]].
* '''Runtime instrumentation''': Directly before execution the code is instrumented. The program run is fully supervised and controlled by the tool. 
* '''Runtime injection''': More lightweight than runtime instrumentation. Code is modified at runtime to have jumps to helper functions.

===Interpreter instrumentation===
* '''Interpreter debug''' options can enable the collection of performance metrics as the interpreter encounters each target statement. A [[bytecode]], [[control table]] or [[Just-in-time compilation|JIT]] interpreters are three examples that usually have complete control over execution of the target code, thus enabling extremely comprehensive data collection opportunities.

===Hypervisor/simulator===
* '''Hypervisor''': Data are collected by running the (usually) unmodified program under a [[hypervisor]]. Example: [[SIMMON]]
* '''Simulator''' and '''Hypervisor''': Data collected interactively and selectively by running the unmodified program under an [[instruction set simulator]].

==See also==

<!-- Please keep entries in alphabetical order & add a short description {{annotated link|WP:SEEALSO}} -->
{{div col|small=yes|colwidth=20em}}
* {{annotated link|Algorithmic efficiency}}
* {{annotated link|Benchmark (computing)|Benchmark}}
* {{annotated link|Java performance}}
* {{annotated link|List of performance analysis tools}}
* {{annotated link|Performance Application Programming Interface|PAPI}}
* {{annotated link|Performance engineering}}
* {{annotated link|Performance prediction}}
* {{annotated link|Performance tuning}}
* {{annotated link|Runtime verification}}
* {{annotated link|Profile-guided optimization}}
* {{annotated link|Static code analysis}}
* {{annotated link|Software archaeology}}
* {{annotated link|Worst-case execution time}} (WCET)
{{div col end}}
<!-- please keep entries in alphabetical order -->

== References==
{{reflist|30em}}

==External links==
* Article "[http://www.ibm.com/developerworks/rational/library/05/1004_gupta/ Need for speed &mdash; Eliminating performance bottlenecks]" on doing execution time analysis of Java applications using [[IBM Rational Application Developer]].
*[http://software.intel.com/sites/products/documentation/hpc/vtune/windows/jit_profiling.pdf Profiling Runtime Generated and Interpreted Code using the VTune Performance Analyzer]

{{DEFAULTSORT:Software Performance Analysis}}
[[Category:Software optimization]]
[[Category:Profilers|*]]