Hi Konstantin, good point. but I my intention is to have something like the llva project. Basically I would like to define a machine that looks as if it has processors running natively llvm code. And when that runs enhance the llvm byte code with some specific instructions. Besides this if I make my measurements on the running of llvm bytecode I can test several platforms simultaneously because I can translate the enhancements done on llvm code in optimizations specific for each platform without having to redo the research. regards, dacian ________________________________ From: Konstantin Tokarev <annulen at yandex.ru> To: Herbei Dacian <dacian_herbei at yahoo.fr>; "llvmdev at cs.uiuc.edu" <llvmdev at cs.uiuc.edu> Sent: Thursday, 19 September 2013, 16:25 Subject: Re: [LLVMdev] LLVM virtual machine 19.09.2013, 18:20, "Herbei Dacian" <dacian_herbei at yahoo.fr>:> Hi, > Is there possible to run the llvm byte code purely interpreted with lli? > I know there is this option but in the documentation is stated also that the option is not working and is not maintained. > Did I understand wrong? Is it working? > Is anyone working on this option? > I would like to make something similar to q-emu or bochs.In this case you probably need to interpret machine code (or assembly language) instead of llvm bitcode. -- Regards, Konstantin -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130919/5bd80e94/attachment.html>
On 9/19/13 9:53 AM, Herbei Dacian wrote:> > Hi Konstantin, > good point. > but I my intention is to have something like the llva project.If you want something like the LLVA project for user-space applications, then you basically want to use LLVM as-is. The only things missing are the instructions that replace certain in-line assembly sequences that cannot be represented by regular LLVM instructions or LLVM intrinsics (e.g., context switching), and chances are good that you don't need them unless you're supporting kernel code or something like a user-space pthreads library (or you are really pedantic about not using inline assembly code). As an FYI, I'm building a new version of SVA (LLVA) that works on 64-bit x86 processors. It supports the FreeBSD 9.0 kernel. I probably won't replace all the inline assembly code in the FreeBSD kernel like I did for Linux 2.4.22 years ago, but most of the important features will be there (and are, in fact, already implemented). We plan to release this version as open-source software within the next year (I suppose we could release the old version, too, but nobody uses Linux 2.4 on x86 anymore), but I don't have a timeline for that yet. My first order of business is to use the new SVA system to graduate. :) -- John T.> Basically I would like to define a machine that looks as if it has > processors running natively llvm code. > And when that runs enhance the llvm byte code with some specific > instructions. > Besides this if I make my measurements on the running of llvm bytecode > I can test several platforms simultaneously because I can translate > the enhancements done on llvm code in optimizations specific for each > platform without having to redo the research. > regards, > dacian > > > ------------------------------------------------------------------------ > *From:* Konstantin Tokarev <annulen at yandex.ru> > *To:* Herbei Dacian <dacian_herbei at yahoo.fr>; "llvmdev at cs.uiuc.edu" > <llvmdev at cs.uiuc.edu> > *Sent:* Thursday, 19 September 2013, 16:25 > *Subject:* Re: [LLVMdev] LLVM virtual machine > > > 19.09.2013, 18:20, "Herbei Dacian" <dacian_herbei at yahoo.fr > <mailto:dacian_herbei at yahoo.fr>>: > > Hi, > > Is there possible to run the llvm byte code purely interpreted with lli? > > I know there is this option but in the documentation is stated also > that the option is not working and is not maintained. > > Did I understand wrong? Is it working? > > Is anyone working on this option? > > I would like to make something similar to q-emu or bochs. > > In this case you probably need to interpret machine code (or assembly > language) instead of llvm bitcode. > > > -- > Regards, > Konstantin > > > > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130919/15ffe1f6/attachment.html>
Hi John, maybe if I help you I can get faster into my own problem. So may I? Maybe in two we could replace the inline assembly too. regards, dacian ________________________________ From: John Criswell <criswell at illinois.edu> To: Herbei Dacian <dacian_herbei at yahoo.fr> Cc: Konstantin Tokarev <annulen at yandex.ru>; "llvmdev at cs.uiuc.edu" <llvmdev at cs.uiuc.edu> Sent: Thursday, 19 September 2013, 17:37 Subject: Re: [LLVMdev] LLVM virtual machine On 9/19/13 9:53 AM, Herbei Dacian wrote:>Hi Konstantin, >good point. >but I my intention is to have something like the llva project. >If you want something like the LLVA project for user-space applications, then you basically want to use LLVM as-is. The only things missing are the instructions that replace certain in-line assembly sequences that cannot be represented by regular LLVM instructions or LLVM intrinsics (e.g., context switching), and chances are good that you don't need them unless you're supporting kernel code or something like a user-space pthreads library (or you are really pedantic about not using inline assembly code). As an FYI, I'm building a new version of SVA (LLVA) that works on 64-bit x86 processors. It supports the FreeBSD 9.0 kernel. I probably won't replace all the inline assembly code in the FreeBSD kernel like I did for Linux 2.4.22 years ago, but most of the important features will be there (and are, in fact, already implemented). We plan to release this version as open-source software within the next year (I suppose we could release the old version, too, but nobody uses Linux 2.4 on x86 anymore), but I don't have a timeline for that yet. My first order of business is to use the new SVA system to graduate. :) -- John T. Basically I would like to define a machine that looks as if it has processors running natively llvm code.>And when that runs enhance the llvm byte code with some specificinstructions.>Besides this if I make my measurements on the running of llvmbytecode I can test several platforms simultaneously because I can translate the enhancements done on llvm code in optimizations specific for each platform without having to redo the research.>regards, >dacian > > > > > > >________________________________ > From: Konstantin Tokarev <annulen at yandex.ru> >To: Herbei Dacian <dacian_herbei at yahoo.fr>; "llvmdev at cs.uiuc.edu" <llvmdev at cs.uiuc.edu> >Sent: Thursday, 19 September 2013, 16:25 >Subject: Re: [LLVMdev] LLVM virtual machine > > > >19.09.2013, 18:20, "Herbei Dacian" <dacian_herbei at yahoo.fr>: >> Hi, >> Is there possible to run the llvm byte code purelyinterpreted with lli?>> I know there is this option but in the documentationis stated also that the option is not working and is not maintained.>> Did I understand wrong? Is it working? >> Is anyone working on this option? >> I would like to make something similar to q-emu orbochs.> >In this case you probably need to interpret machine code(or assembly language) instead of llvm bitcode.> > >-- >Regards, >Konstantin > > > > > >_______________________________________________LLVM Developers mailing list LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130919/2f2e00b8/attachment.html>