Would pascal be able to keep up with the competition? lazarus / 8-Other ---------------------------------------------------------------------- pierre.se.pos · 2013-03-30 12:42 Hi Would pascal be able to keep up with the competition? We are at the end of the "languages war" and in the RIA period. http://www.codenewsfast.com/cnf/article/2055760141/permalink.art- ng1914q21709 The next period would probably be parallel (not threaded) processing on desktops, tablets  and smartphones. http://www.extremetech.com/computing/134760-pc-obsolescence-is- obsolete http://linuxgizmos.com/pondering-the-future-of-moores-law/ https://developer.nvidia.com/cuda-llvm-compiler http://www.phoronix.com/scan.php?page=news_item&px=MTIzOTE http://www.amd.com/us/products/desktop/processors/amdfx/Pages/amdfx.as px http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-August/052395.html http://www.gottabemobile.com/2013/03/26/samsung-quad-core-cpu- required-for-android-5-0/ Any computer language that cannot be auto translated to parallel code (where possible)  could be left behind and become a "dead" computer language (since it would not be able to  compete with the rest).  I could not find any indication that people are working on a pascal compilers with auto parallelism. The only compiler technology that I could find where people is spending a lot of money and also working on auto parallellisim is the (open source) LLVM compiler technology. https://sites.google.com/site/parallelizationforllvm/ http://polly.llvm.org/ Many languages have LLVM support http://en.wikipedia.org/wiki/LLVM http://www.phoronix.com/scan.php?page=news_item&px=MTIxMDM http://donsbot.wordpress.com/2010/02/21/smoking-fast-haskell-code- using-ghcs-new-llvm-codegen/ Embecadero seems to be working on LLVM for Delphi. http://www.codenewsfast.com/cnf/article//permalink.art-ng1914q119901 FPC have it on their todo list. http://www.freepascal.org/future.var There are people working on a pascal compiler for LLVM http://code.google.com/p/llvm-pascal/ Would pascal keep up with the competition and have a LLVM pascal compiler available in the next two years? ---------------------------------------------------------------------- Leledumbo · 2013-03-30 12:47 QuoteWould pascal keep up with the competition and have a LLVM pascal compiler available in the next two years?No, if you read a mailing list thread talking about this a few weeks ago. We don't need LLVM for that. And no, in the next two years parallel language won't yet conquer the world. Business people tend to speak for the future as if they know everything. They said the same thing about web application killing desktop application around five years ago. Has it come true? Hell NO! As for parallelism, you can already do that with existing libraries. I personally have tried OpenMP with FPC when I took parallel computing class during college time around two years ago. ---------------------------------------------------------------------- jwdietrich · 2013-03-30 13:12 Quote from: Leledumbo on March 30, 2013, 01:47:25 pmAs for parallelism, you can already do that with existing libraries. I personally have tried OpenMP with FPC when I took parallel computing class during college time around two years ago. Were you successful with your tries? Is there a recommended way to include OpenMP functionality in Free Pascal? The wiki page (http://wiki.freepascal.org/OpenMP_support) seems to consist of hypotheses and proposals only. ---------------------------------------------------------------------- Leledumbo · 2013-03-30 13:17 My, my. Shoot my head. I picked the wrong thing, what I mean is MPI. ---------------------------------------------------------------------- jwdietrich · 2013-03-30 13:27 Quote from: Leledumbo on March 30, 2013, 02:17:58 pmwhat I mean is MPI. Did you use the MPICH implementation? ---------------------------------------------------------------------- Leledumbo · 2013-03-30 13:56 QuoteDid you use the MPICH implementation?Yes, MPICH2 wrapper unit from here. ---------------------------------------------------------------------- CaptBill · 2013-03-30 19:15 Quote from: pierre.se.pos on March 30, 2013, 01:42:35 pm"Would pascal be able to keep up with the competition?" "We are at the end of the "languages war" and in the RIA period." "The next period would probably be parallel (not threaded) processing..." "Any computer language that cannot be auto translated to parallel code (where possible)  could be left behind and become a "dead" computer language (since it would not be able to  compete with the rest)." Well, they you expect these new concepts to happen, and become the new "be all" solution, they should at least understand that OO Pascal is the obvious choice,and why. OOPascal was and is the 'prime mover' in object oriented languages. C is and was and will be the 'overall prime mover' as it should be, I would imagine. It's not going anywhere. That said, even C must admit that C++ was created to 'catch up to Delphi'. So in the OO world Lazarus/Fpc/Delphi has the deepest strongest roots. Question is, are all those other languages really ready to challenge Lazarus head on? I think not. So Leledumbo, What is the status of FPos? What would it take to target GPUs with your os? Something like that would sure end this supposed 'competition'. ---------------------------------------------------------------------- taazz · 2013-03-30 21:17 The future for me is divided on 2 fronts 1 The server side where google has already  proven that a farm of low cost low energy computers is better than a single high end server add to that some raid nats on gigabit networks and you have a big market for parallelism with easy node replacement, addition, subtraction etc. This front is still not clear as a single high end server is easier to handle at the moment but as the compilers start to support parallelism more sufficient it will tip the scale to the low end high volume part of the things if not already there. 2) client side (desktops, handhelds, phones etc) This front has already been tiped in favor of scripting especialy javascript on the browser this can easily be overcome as the frameworks compilers change to allow easier multi threading and GPU support making things a lot faster to handle. there is a huge spreed for 3d and hd computing at the moment treating the 2D as a second class citizen although 90% of the current computing needs are 2D. It is a shame after all those years of 2D computing there is no real evolution, especially on the existing OO frameworks which most of them (QT excluded for luck of research at this point) are monolithic designed. there are a lot of things that can be improved including the combination of 2d and 3d programming in a single roof instead of having two frameworks. Of course all that need to come under a single umbrella that will eliminate the language wars something that MS tried to accomplish with the .NET framework but made a small mess of things. In any way to answer the OP question no pascal will not be able to keep up not of luck of talent but from luck of work force. My estimation is that pascal has around 5% of the developer power that C/C++ has so it requires at least 20 or 30 times more work from that force to just keep up, the good thing on all that is that it does not have to keep up it just have to shine brightly in a single item to start attracting developers and work its way up to a critical mass that will allow faster evolution. ---------------------------------------------------------------------- marcov · 2013-03-30 22:02 Quote from: pierre.se.pos on March 30, 2013, 01:42:35 pm Any computer language that cannot be auto translated to parallel code (where possible)  could be left behind and become a "dead" computer language (since it would not be able to  compete with the rest). This is all wishful thinking. All the current efforts only convert very small parts to autoparallel, and the overall speed improvements (of general code auto-parallellized, so not manually parallellized code which happens outside of compilers) are negiable. Yes, people want to go there, and people are announcing plans, and special cases. But in reality, if you buy a big name development toolchain (Delphi, MSVCRT, whatever), it won't autoparallellize code.  So this is all "we want". It is about the same as using GPGPU where two years ago everybody was yacking about, and except for photoshop and a few codecs it is not that much used in programs for the general audience. And of course the compesitors that were using it back then too. There is a big difference in trends that are predicted, and trends that make it. Quote The only compiler technology that I could find where people is spending a lot of money and also working on auto parallellisim is the (open source) LLVM compiler technology So what speedups have you seen and found? You are just posting a very broad and general statement on LLVM, and even more general LLVM links. It doesn't really paint a convincing picture. And not all parallel work is suitable for general usage. Some still need specially crafted code, or code using a new framework (but the general code is then relatively independent of the number of cores) I think you are taking rumours and intentions as facts. But another hard fact is that there is no broadly applicable algorithm to autoparallellize arbitrary code. ---------------------------------------------------------------------- Leledumbo · 2013-03-30 22:04 QuoteSo Leledumbo, What is the status of FPos? What would it take to target GPUs with your os? I give up, I need to be a lot younger to play with such a project. As a fulltime employee + more useful (possibly open source) projects programmer, I don't have enough time. ---------------------------------------------------------------------- felipemdc · 2013-04-04 14:04 Parallelization as the maximum priority? I have never been in a real world project which had this requirement. Most projects for me need to be done as easily for me as possible and with as good debugging capabilities as possible, so I use Lazarus/FPC Where I work the projects often need very high performance for large web servers and fast graphics in telephone/tablet native applications and those use mostly C++, being that python and java are excluded due to their bad performance. Here paralelization is not a magical solution at all, by the way. The web server is already paralelized by other means and phone apps already use multiple threads. So which real world project really desperately needs more paralelization then multiple threads/processes can give? never seen one.... ---------------------------------------------------------------------- felipemdc · 2013-04-04 14:42 Webification on the other hand is indeed a threat .... at least for some areas. ---------------------------------------------------------------------- Leledumbo · 2013-04-04 15:22 QuoteWebification on the other hand is indeed a threat .... at least for some areas.I just had such a request yesterday. Thank God I made my applications as modular as possible, I just need to write the web front ends with brook and 2 apps have got online today ---------------------------------------------------------------------- BigChimp · 2013-04-04 15:33 Pierre my ou, are you still there? You kicked off quite a discussion and disappeared into the sunset, it seems... ---------------------------------------------------------------------- mica · 2013-04-04 16:56 Parallelization? ok 1993: https://en.wikipedia.org/wiki/SuperPascal http://brinch-hansen.net/READTHIS.html ---------------------------------------------------------------------- pierre.se.pos · 2013-05-08 07:16 Quote from: BigChimp on April 04, 2013, 05:33:29 pmPierre my ou, are you still there? You kicked off quite a discussion and disappeared into the sunset, it seems... Hi, yes, I am still here. Seems as if there is not enough resources in pascal if compared to the competition. I really like pascal, so why not join the competition (where a lot of time and money is spent) in such a way so as to preserve pascal? I have been following the competition: http://www.nasa.gov/directorates/spacetech/strg/creech.html http://aesop.ece.umd.edu/ http://www.sambamba.org/ http://www.phoronix.com/scan.php?page=news_item&px=MTM2NjU http://www.phoronix.com/scan.php?page=news_item&px=MTIzOTE http://www.phoronix.com/scan.php?page=news_item&px=OTc5NQ http://www.phoronix.com/scan.php?page=news_item&px=MTM2NjE http://www.phoronix.com/scan.php?page=news_item&px=MTMzNzY ---------------------------------------------------------------------- pierre.se.pos · 2013-05-09 07:01 Auto-Parallelization (Vectorization) already implemented by the competition: Visual Studio 2012 http://msdn.microsoft.com/en-us/library/vstudio/hh409293.aspx http://blogs.msdn.com/b/nativeconcurrency/archive/2012/04/12/auto- vectorizer-in-visual-studio-11-overview.aspx http://www.florian-rappl.de/News/Page/139/vs2012simd LLVM http://llvm.org/docs/Vectorizers.html GCC http://gcc.gnu.org/projects/tree-ssa/vectorization.html ---------------------------------------------------------------------- fabienwang · 2013-05-09 07:04 Come on kid, the parallelization is only used while compiling... so what's the gain in it? very small ! ---------------------------------------------------------------------- Leledumbo · 2013-05-09 08:43 Quotethe parallelization is only used while compiling... so what's the gain in it? very small !No, the vectorization is done to compilation result (i.e. object/executable code). pierre, I know your intention, but without any effort, it worths nothing. You want something in the compiler? Implement it! At least go find some help. I don't think auto vectorization is in any of the core developer's to do list yet. This is a very advanced optimization technique which requires a lot of effort (no wonder why VS, LLVM, GCC can do it, they have A LOT of money support and full time engineers). ---------------------------------------------------------------------- marcov · 2013-05-09 09:11 Quote from: Leledumbo on May 09, 2013, 10:43:13 ampierre, I know your intention, but without any effort, it worths nothing. You want something in the compiler? Implement it! At least go find some help. I don't think auto vectorization is in any of the core developer's to do list yet. This is a very advanced optimization technique which requires a lot of effort (no wonder why VS, LLVM, GCC can do it, they have A LOT of money support and full time engineers). One of the things that strikes me most is that it is a madless dash for some optimization without any reflection or discussion where it is applied in practice (kind of applications), and what magnitude of performance it will bring. It is just mindless spinning of some LLVM/gcc links, and IMHO better ignored. ---------------------------------------------------------------------- JuhaManninen · 2013-05-09 09:39 The video here shows impressive results  http://www.sambamba.org/ Ok, demos are always made to show impressive results, but this one was said to adjust to the length of data in the program and using parallel execution only when it really made sense. It will work with all languages compiled with LLVM. FPC with LLVM backend will be supported, too. That is what I understood. ---------------------------------------------------------------------- Leledumbo · 2013-05-09 11:10 QuoteFPC with LLVM backend will be supported, too. That is what I understood.I don't think so (based on a mailing list discussion a while ago), especially Florian. ---------------------------------------------------------------------- marcov · 2013-05-09 11:38 Quote from: JuhaManninen on May 09, 2013, 11:39:37 amThe video here shows impressive results  http://www.sambamba.org/ Ok, demos are always made to show impressive results, but this one was said to adjust to the length of data in the program and using parallel execution only when it really made sense. That kind of demonstrations is exactly what I meant.  You can see that the example program is in vector notation (All data is in arrays or vector types), and the rate determining step is a sequence of operations on such arrays. How many programs that you make have that bottleneck? QuoteIt will work with all languages compiled with LLVM. FPC with LLVM backend will be supported, too. That is what I understood. I haven't heard anything about the LLVM backend in an year. It does seem to be the most "live" of technologies. ---------------------------------------------------------------------- chrnobel · 2013-05-09 12:21 Quote from: marcov on May 09, 2013, 11:11:28 amOne of the things that strikes me most is that it is a madless dash for some optimization without any reflection or discussion where it is applied in practice (kind of applications), and what magnitude of performance it will bring.Must totally agree on that - helloooo, earth calling. Pascal/Lazarus is a wonderful tool for creating programs, and can solve tons of daily problems, without requiring a lot of processor power etc. So if we step on step back, and look at all the ordinary problems that have to be solved in this world, we will already see a zillion of more or less badly made program, which people uses each and every day. And if we use the power of Pascal to solve these ordinary problems, and write good and decent programs, there is a colossal future for Pascal. Unfortunately we are surrounded by buzzword mongers, who love to play bullshit bingo, and who see it as paramount to make things as complex as possible, so they by the end of the day can produce a "hello world" program which they by hard effort were able to optimize so "it only requires a Deep Blue" to execute - gone are all thoughts of KISS, and focus on making good programs that each and everybody will benefit from. Just my 5 cents. /Christian ---------------------------------------------------------------------- JD · 2013-05-09 16:06 Quote from: chrnobel on May 09, 2013, 02:21:18 pmUnfortunately we are surrounded by buzzword mongers, who love to play bullshit bingo, and who see it as paramount to make things as complex as possible, so they by the end of the day can produce a "hello world" program which they by hard effort were able to optimize so "it only requires a Deep Blue" to execute - gone are all thoughts of KISS, and focus on making good programs that each and everybody will benefit from. Blame it on consultants. That's how we get hired by the big corporations who love to hear the latest from the "spin" doctors believing that the latest is always the best. And if you don't give them what they want, you'll lose out to someone else. Unfortunately it will not change anytime soon. Smaller companies are more realistic.