News

news@lists.squeakfoundation.org

109 discussions

[Draft Article] Optimizing Contexts
by Matthew Fulmer 11 Jul '07

11 Jul '07

Cross-posted to: strongtalk-general(a)googlegroups.com vm-dev(a)lists.squeakfoundation.org exupery(a)lists.squeakfoundation.org news(a)lists.squeakfoundation.org comp.lang.smalltalk Hello lists. This is a draft of an article I am writing for an upcoming series at the Weekly Squeak. It is about optimizing context creation, I would like comments on its technical correctness, as well as suggestions for related content I should include. I would especially like to know if this article covers the way this is handled in systems I did not specifically research, such as Strongtalk, and Exupery, and non-smalltalk systems like Common Lisp, Erlang, Haskell, and OCaml, since this topic is not really specific to object-oriented systems. This will not be the first article of the series; however, the first two paragraphs are the preliminary series introduction. I put them in so that you can get an idea of what I am writing. Sections enclosed in ... tags are parts I am especially unsure of; I would appreciate comments (I used i tags, because my editor, vim, highlights the enclosed text when in HTML mode). Look forward to this appearing in the Weekly Squeak: http://news.squeak.org/ Thank you for your help, and enjoy the article! === Speeding Up Smalltalk Series, part 42: Context Optimizations === Ever since object-oriented systems became popular, they have had the image of being very easy to program, but very slow in terms of execution speed. This image and reality has plagued every popular object-oriented system, starting with Smalltalk in the 80's and continuing on to Java and Ruby today. However, this does not have to be the case. Early-bound languages have gotten much faster thanks to lots of research on compilers and computer architecture. Similarly, much research has been done on making late-bound object-oriented systems fast, both with software and with hardware. This research has not seen as much exposure as processor research, and so is probably less familiar to you. This series, "Speeding Up Smalltalk" attempts to summarize a lot of that research and present it for your enjoyment. Computation in Smalltalk and Self is done by sending messages to objects. This seemingly simple operation of sending a message is done so often and is so flexible that it is always the execution bottleneck in any implementation of Smalltalk or similar object-oriented system. Why is that? ---- Description of Context Objects ---- Every time a message is sent, four things happen: - A search is done through the message receiver for a method that has been sent. This was the topic of the previous article, and will not be discussed here. - A context object is allocated on the heap. - The context object is initialized and filled with: - local objects, including: - pointers to the message arguments - pointers for each local variable - pointers for temporary storage needed for intermediate results - A way to reference variables defined in other scopes. This is a property blocks have, but methods don't - In Smalltalk, by a pointer to the "home context" - In Self, by delegate slots to the enclosing context(s) - A return address at which to resume execution - Squeak does not need this, since every context has it's own PC, so when returning, the PC of the returned-to context is already in the right place. This is equivalent to a return address, since a child's return address is the parents PC. - A return context that will become current upon return - The newly-created context is made the "current" context for this thread/process In Squeak, MethodContexts and BlockContexts are subclasses of ContextPart. ContextPart has three instance variables: - sender: The context that sent the message being executed. This context will become current when the current context returns - pc: A pointer to the currently executing instruction - stackp: The pointer to the execution stack. The smalltalk virtual machine is a stack processor. local variables are stored in the indexed fields of the context. stackp keeps track of which OOP is at the top of the stack. For MethodContexts: MethodContext extends it with 3 instance variables, and a variable-sized area for holding the temporaries: - method: the CompiledMethod holding the code being executed - receiverMap: Something to do with the names of the variables. Not really sure - receiver: a pointer to "self", the receiver of the message. ---- Squeak optimization: Static Contexts ---- In Squeak, BlockContexts are partial closures (sometimes called static blocks/closures [3]). Rather than hold their own code and environment, BlockContexts piggyback off of another context/code pair. Home is the environment that the block piggybacks off of. This usually works, but there are some cases where this scheme will not work: - like what? Why not? I don't understand why this doesn't always work. Something to do with re-entrancy or circular references. Not sure why either is a problem. For BlockContexts: - nargs: the number of arguments that the block must be called with. - startpc: The index into <code>self home method</code> where the code for the block starts. - home: the home environment Squeak's NewCompiler has a different representation of blocks. These are full closures, storing their own code and environment. They are sometimes called dynamic closures to distinguish them from the above "static closures" [3]. They contain: - method: The code of the block (CompiledMethod2) - environment: Captured temps (ClosureEnvironment) ---- Squeak optimization: Context Recycling ---- To avoid having to create a new MethodContext on every message send, Squeak keeps a pool of pre-allocated MethodContexts, and simply fills one up when a new one is needed. To support this optimization, MethodContexts are only created with two sizes, rather than any size that they could support. The small size has room for a few temporaries, and the large size has room for a bunch [4] ---- VisualWorks optimization: Context Caches ---- The most natural way to store contexts is as a linked list of objects, but that does not map well to the hardware of many processors. Many processors (including all the Intel varieties) have several built-in instructions to mechanize the creation of stack frames, which are similar to context objects, but have several important differences. VisualWorks uses these native stack frames for most operations, meaning that context creation is identical to a C function send in many cases. Thus, the hardware stack acts as a stand-in for the context objects; this optimization is called a context cache. However, stack frames cannot handle context responsibilities beyond a single call and return; when more is needed, a context object must be allocated on the heap. The context object can either be a context in it's own right (a stable context), or act as a proxy to the hardware stack frame (hybrid context). There are several complications to this dual scheme; curious readers should refer to [5]. VW and Squeak does not use a stack of context frames; instead it uses a linked list of real context objects. This is easier, but perhaps slower. To save on allocation costs, contexts are cached and recycled for various method invocations. Not sure where to incorporate this There are three types of blocks: - open/clean blocks (pure function with no closure) - copying blocks - full blocks ---- Or you can cheat... ---- The most popular object-oriented languages (C++, Java) chose to dump the flexibility of blocks in exchange for runtime speed. These languages use a native stack frame without providing the escape mechanism of a context cache, so that blocks and other context behavior that does not follow stack principles is not supported. This is a tradeoff that eliminates a good deal of flexibility, and so will not be explored further. ---- Final words ---- Inlining solves many of the same problems as a Context cache does. If the method body is inlined into the caller, no new context need be created. However, inlining solves many more problems than just context elimination; I will dedicate an entire article to it. ---- References ---- - context recycling may be discussed in tims nublue book: http://www.rowledge.org/tim/squeak/OE-Tour.html - context caches are discussed in section 5.2.3 of http://www.kampjes.demon.co.uk/articles/exuperyDesign.pdf [3] add reference to squeak mailing list or Ian's something to discuss the difference between static and dynamic closures [4] Squeak Class comments of <code>InstructionStream</code>, <code>ContextPart</code>, <code> MethodContext</code>. and <code>BlockContext</code>. [5]: Context Management in VisualWorks 5i: http://wiki.cs.uiuc.edu/VisualWorks/DOWNLOAD/papers/oopsla99-contexts.pdf -- Matthew Fulmer -- http://mtfulmer.wordpress.com/ Help improve Squeak Documentation: http://wiki.squeak.org/squeak/808

1 0

Re: News Team Monthly Report for June 2007
by Giovanni Corriga 04 Jul '07

04 Jul '07

Il giorno mar, 03/07/2007 alle 15.03 -0700, Blake ha scritto: > > We were listed in the list of top blog posts on wordpress a few times. > > We really could use some new members. If you have an interest in writing > > articles for weeklysqueak please contact me. > > Mailing back to the list in general to ask: What sort of articles? Do you > have some broad, general guidelines for material you'd like to see? The > two you mentioned in your e-mail seem to be an interview and a...well, > sort of another interview(s). Blake, what kind of articles are you interested in? Giovanni

1 0

Beginner's guide to developing in Squeak
by Michael Davies 02 Jul '07

02 Jul '07

Steve Wessels has published an excellent and comprehensive beginner's guide to development in Squeak, using the example of a simple game to give advice on use of projects, testing, use of morphs, tracking down bugs, and managing and publishing your source code. See it here: http://squeak.preeminent.org/tut2007/html/index.html Cheers, Michael

1 0

Scratch on the wishlist of openSUSE's add-on for educators "SLEDucator"
by Klaus D. Witzel 26 Jun '07

26 Jun '07

Perhaps in competition with Edubuntu? Here's the news: - http://education.zdnet.com/?p=1079 - http://linux.slashdot.org/linux/07/06/26/1225220.shtml /Klaus

1 0

Squeak!
by antonio 22 Jun '07

22 Jun '07

http://www.youtube.com/watch?v=jVz76Tsi9so regards

1 0

Baked etoys
by Bert Freudenberg 16 Jun '07

16 Jun '07

Not sure if that is newsworthy, but I found point 3 below funny - OLPC is using Alan's self-running demo project for testing the laptop's function at extreme climate conditions (as in the desert with 30 degrees Celsius difference between night and day). I guess they use etoys because there is no other pre-installed software that has a similar self-running demo, yet. - Bert - Begin forwarded message: > From: "Walter Bender" <walter.bender(a)gmail.com> > Date: June 16, 2007 21:03:53 GMT+02:00 > To: community-news(a)laptop.org > Subject: [Community-news] OLPC News 2007-06-16 > Reply-To: walter(a)laptop.org > > 1. Montevideo: On Friday, the Technology Laboratory of Uruguay (LATU) > released a bid for Project Ceibal (Conectividad Educativa de > Informática Básica para el Aprendizaje en Línea)—one laptop per child > in Uruguay. > > 2. Olin College hosted the first OLPC Game Jam (See > http://wiki.laptop.org/go/Game_Jam) last weekend, bringing together > ten teams of game developers and some freelance artists, musicians, > and programmers, to make games for the XO. Organizers Mel Chua and SJ > Klein are working on general notes re: organizing game jams and other > local community events to develop materials for the XO. Most of the > teams chose to work in Python, though a few developed in Flash. (A > Flash developer who had rather vehemently against Python at the start > of the weekend, wouldn't stop talking about how nice Python was by > Sunday.) Teams collaborated with one another, in addition to competing > to make the best game; they shared music and artistic expertise, and > code snippets and coding advice. (The Flash developers uniformly > wanted to write things that would work in Gnash on our platform, not > standard Flash 9; they spent part of Friday and Saturday working with > the Gnash team to help improve its utility for game development.) > > The two best reviewed games both used PyGame; they were a version of > 3D Pong and a version of the old Crossfire game called Spray Play (See > http://wiki.laptop.org/images/c/c6/3dpong.activity.zip and > http://sprayplay.googlecode.com/svn/). > > 3. Taking the heat: We have decided to see how much heat XO can take. > Mary Lou Jepsen has instructed UL to test our laptop for a 50C (122F) > operating temperature. Typical laptops are only tested to 35C (95F) or > 40C (104F), which is unacceptable for the children who will be using > our laptops in hot temperatures (e.g., in direct sunlight and of > course without air conditioning). Mary Lou and Tracy Price are also > running a simple bake test at the OLPC office, pictured above. The > laptop is running days at 52C (125F), and nights at 22C (72F). UL and > Quanta are doing more extensive testing, but shown is a laptop, > running the eToys demo that sits in the oven night and day. Try that > with a conventional laptop! > > 4. Green: Mary Lou and Robert Fadel have started the application > process for EPEAT Gold—the highest award given to laptops; one no > other laptop has yet received. Also, late last week Google's Ethan > Beard and Megan Smith, and Red Hat's Mike Evans invited OLPC to join > with Google, Intel, Quanta, Red Hat, AMD, HP and others in the IT > industry to launch Climate Savers, an organization dedicated to > lowering the power consumption of computers through better power > management systems, and more efficient AC adaptors. Climate Savers > picked lower power as the single thing on which to concentrate in > order to have the biggest positive impact on the environment. OLPC > concurs with this believe. At first those that join Climate Savers > agree to meet the Energy Star goals—OLPC is already 14× better than > Energy Star. > > 5. $1 video microscope: Inspired by SJ Klein and EO Smith, Mary Lou > made a 100× video microscope for her XO for $1 (three plastic lenses > in plastic housing). She made videos of the XO screen compared with a > standard LCD screen, where the details of the pixel structure can be > clearly seen. She will be compiling a video for youtube.com in the > coming days. > > 6. Sugar: Eben Eliason has continued to refine a series of mock-ups > for rollovers, invitations, and notifications. He has created a new > series of Activity mockups, including Browse, Read, Write, Memorize, > Calculate, Photograph/Capture/Record, and TamTam that feature tagging > and tabs. He also created a preliminary specification for keyboard > shortcut design, now open for discussion. Also he worked with Jim > Gettys to figure out some logic for the hand-held buttons in terms of > desired functionality and semantic meaning. Marco Gritti has been > making changes to the GTK theme to incorporate many of these > improvements. > > 7. Marc Maurer continues work on the Write activity, with his focus > mostly around collaboration. He has been working on a new algorithm to > handle collisions in documents when people are editing the same part > of a document. He also spent a lot of time fixing bugs in Abiword to > close a blocker bug in the 406 Build. > > 8. Muriel de Souza Godoi updated the Memory Activity to the new sugar > API; now all the memory games were unified in one activity. He also > worked Eben designed a new Memorize Game UI; the new scoreboard was > developed as a component, with methods such as: set fill color, set > stroke color, increase score, set_current_player, etc. The new card > table was also developed as a component and can be controlled using > the hand-held-mode buttons. These UI components are designed to be as > flexible as possible, focusing on reusing components. > > 9. Journal: Tomeu Vizoso has been working on the Journal; he has added > the ability to do screen capture by typing Alt-1; the image is saved > to the Journal. He also has been working to make it possible to launch > downloaded activities directly from the Journal. He has been updating > the web browser in order making it work with the new Journal code as > well as the new code to interface with Python. Ben Saller has been > working on how to get the Journal to support alternate media such as > USB drives. Eben created a new series of Journal mock-ups that > incorporate tabbed toolbars, address support for "sort by, then by," > and for versioning. > > 10. Mesh Activities: Dan Williams made progress with Network Manager > (NM) and the mesh. NM will now automatically scan and get an address > on the mesh network. The Collabora folks continue down the path of > making the peer-to-peer presence-discovery code and tubes code work. > They also added a "Hellomesh" Activity that shows how to build a > tubes-enabled activity. (Please note that the activity will change > over time as the tubes API stabilizes.) Eben worked extensively back > and forth with Pentagram on an updated UI design for the mesh view. > > 11. Fedora Core 7: John Palmieri has been moving our builds to a > Fedora 7 base. Once that is done we will have a lot more opportunity > to collaborate with the community and also get more direct help from > the 1200 or so Fedora contributors. Moving to Fedora 7 also means that > many of our modified packages are rolled up into the main repository. > > 12. Build 406.14: Firmware and a stable kernel were released to Quanta > for the Btest-4 build, derived from Build 406. Suspend and resume are > working in a full build for the first time, including autonomous mesh > networking, a first for any system anywhere! It is almost, but not > quite stable enough for widespread use; a few remaining bugs need to > be squashed before deployment to a large audience. > > 13. Firmware: This week, Mitch Bradley worked on stabilizing software > and firmware for the B4 build. Mitch also merged ECC checking code > (written by Segher Boessenkool) into CAFE NAND driver and worked out a > plan for storage of the public key that secures firmware updates. > > 14. X Window System: Richard Smith worked with Adam Jackson of Red Hat > to figure out why his DCON mode patches to the X driver were causing > the DCON to flicker and glitch on the switch from DCON mode to GPU > mode. This will enable the window system to disable the video unit and > allow the GPU to idle when not in use. > > Bernardo Innocenti has been enhancing our X keyboard definitions to > include all the missing keyboard symbols and working with upstream to > cleanup and merge our changes into the official repository. Miles > Grimshaw has designed two new keyboards for the XO: Turkish and > Ethiopic. > > Daniel Stone of Nokia suggested to Jim that our slider keys be > represented in the X input extension in a better way: we're going to > have three "analog" sliders on the first row of the keyboard, which > will look like absolute axes to programs. This requires some kernel > work that Bernie has not yet started. > > Generally, we are in a much better shape this week. The new input > framework in X works already, EXA rendering pretty much works too. > Next week Bernie will look into packaging issues with Adam. Jordan > Crouse has fixed many bugs in the X driver, and the he number of bugs > blocking #1604 is quickly shrinking, so we may be able to push this > upgrade just in time for the Fedora Core 7 migration. > > 15. Kernel: Andres Salomon merged the device-tree patch, giving access > to hardware and manufacturing information. The wireless-driver version > supporting suspend/resume was also merged. The EC protocol was > debugged, and debugged some more, and is now mostly fixed. We have a > kernel/firmware combination that suspends/resumes in about two > seconds. The delay is mostly from libertas and USB; Marcelo Tosatti > and the Cozybit team are actively working on these drivers. > > Chris Ball did a lot of stable-build debugging. He found that our > camera's colormap becomes strange after resume and that the > "camera-active" LED comes on at resume even when the camera isn't > being used. Chris wrote a kernel patch to only power up the camera > when a user wants it; Jon Corbet is reviewing the patch. > > 16. IPV6: Scott Ananian began the week by trying to cram the entirety > of "Essential IPv6 Networking" into his head. He set up some IPv6 > tunnels and IPv6-enabled his home site to: (A) make sure he knew how > things worked; and (B) serve as a testbed for the school server > environment, which will likely be behind similar NATs. He took over as > the liaison to SIXXS, which is going to be providing our IPv6 > connectivity via tunnels for the short term, at least until we set up > infrastructure (and possibly write some code) to terminate > NAT-tunneling IPv6 tunnels ourselves here in Cambridge. Scott also > confirmed that private IPv4 addresses are properly assigned to the > laptops if a DHCP server cannot be found. > > Scott's second network-manager-related task was to get it to > understand DNS information sent via Router Advertisement messages as > part of IPv6 > autoconfiguration, so that the machines "just work" without requiring > round-trips to a DHCP server or other setup. Scott noticed that radvd > on our local (OLPC) network (tubes) was giving out "bogus" > information, and wrote a patch for radvdump and sent the patch > upstream in the process. As it turns out, radvd was still using a > stale config and just needed to be sent SIGHUP, which was simple > enough. Scott sent mail to a number of people (including the > appropriate kernel mailing list) outlining a plan to add support for > DNS-in-RA to the Linux kernel and to Network Manager. Scott hasn't > heard any objections yet, so will assume > the plan is good and code up a first-draft implementation next week. > > 17. Hardware: The asynchronous input/output (SPD) bus on the XOs has > problems when coming out of suspend/resume and was causing write to > the display controller (DCON) to fail. Mitch figured out the root > cause of a failure to resume that only shows up on some machines: a > DCON/system-management (SM) bus bug was found and a DCON hardware bug > discovered. Richard, Mitch, Andres, Chris, and Jordan Crouse worked > together to find and produce a fix. > > -walter > -- > Walter Bender > One Laptop per Child > http://laptop.org > _______________________________________________ > Community-news mailing list > Community-news(a)lists.laptop.org > http://lists.laptop.org/listinfo/community-news

1 0

News Team Monthly Report - May 2007
by Ron Teitelbaum 06 Jun '07

06 Jun '07

Hi All, News Team Report for May 2007. Our Highest read article for May: http://news.squeak.org/2007/05/17/minding-your-business-with-smalltalk-part- 3-of-4/ had 2414 readers. This is an interview with Bert Freudenberg. Check it out if you haven't had the time to read it yet. We are getting about 200-300 hits a day. Our best day in May was 1906 readers. We were listed in the top blogs and top post on wordpress twice. We even had a mention on a podcast: http://www.cincomsmalltalk.com/blog/blogView?showComments=true <http://www.cincomsmalltalk.com/blog/blogView?showComments=true&entry=335716 0734> &entry=3357160734 Our readership is still down significantly mostly because of lack of time to publish articles. We really could use some new members. If you have an interest in writing articles for weeklysqueak please contact me. Thanks, Ron Teitelbaum Squeak News Team Leader

1 0

Squeak project SCRATCH in The Economist magazine
by John Shand 24 May '07

24 May '07

Check this out ... http://www.economist.com/science/tq/displaystory.cfm?story_id=9215195 John Shand Squeaker noob

1 0

Re: Something for the News Team? 'UNCW, School Districts Using "Squeak"'
by Giovanni Corriga 24 May '07

24 May '07

Il giorno mar, 22/05/2007 alle 20.56 +0100, Michael Davies ha scritto: > Interesting press release here: > http://appserv02.uncw.edu/news/article.asp?ID=2148 > [...] Thanks Michael! I've published it on http://news.squeak.org/2007/05/23/squeak-in-north-carolina-schools/ . > By the way, should there be an official 'news' address on the site > that such proposed articles could be sent to? I've written a small page on the blog with all the information you need: http://news.squeak.org/write-us/ Ciao, Giovanni

2 3

[News] Re: UNCW, School Districts Using "Squeak" to Increase Student Interest in Information Technology
by Giovanni Corriga 23 May '07

23 May '07

Dear Professor Tagliarini, my name is Giovanni Corriga and I work on the Weekly Squeak blog (http://news.squeak.org). The Weekly Squeak is a blog that provides a news service for the whole Squeak community. I've found your press release regarding your Squeak project, and have blogged about it at http://news.squeak.org/2007/05/23/squeak-in-north-carolina-schools/ . I was wondering if you were available for a small email interview to be published on the blog. Thanks in advance, Giovanni Corriga

1 0

← Newer
1
2
3
4
5
6
7
...
11
Older →

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

News