similar to: [LLVMdev] Problems understanding alias analysis validation logic

I may have found a bug in the andersends pass, but before I try to strip the code down to something reasonable I'd like to make sure I'm not misunderstanding anything. This is against the current SVN head, on my Mac OS 10.5 box. 1) I'm compiling a bunch of C++ code with llvm-g++ (4.0.1). The compiled code includes the gcc 4.0.1 implementation for dynamic_cast (so it can be

On Aug 26, 2009, at 6:40 PM, Max Stonebraker wrote: > Hello, > > Does the LLVM Andersens alias analysis handle function pointers > precisely? I ran it and it looks like it says every function points > points to the universal set. Is this what I should expect? Hi Max, I don't know the answer offhand but please keep in mind that the current andersen's pass has many

Might I suggest the following patch, because this issue has bite me too? You're lured in by the initial good results from the Andersens alias analysis, only to discover the bugs later on. Robert Zeh On Aug 26, 2009, at 9:11 PM, Chris Lattner wrote: > On Aug 26, 2009, at 6:40 PM, Max Stonebraker wrote: >> Hello, >> >> Does the LLVM Andersens alias analysis handle

Hello, Does the LLVM Andersens alias analysis handle function pointers precisely? I ran it and it looks like it says every function points points to the universal set. Is this what I should expect? Max -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20090826/ff8d6c78/attachment.html>

Thanks for the replies, Daniel and Matthijs. I added some code to generate the copy constraint, which I think (in the present form) is wrong (or at best not field sensitive): +void Andersens::visitGetResultInst(GetResultInst &GR) { + if (isa<PointerType>(GR.getType())) + { + // P1 = getresult P2 --> <Copy/P1/P2> +

On Thursday 09 August 2007 17:27, Chris Lattner wrote: > > So what's the right way to do this? There's the regalloc way, which > > invents a whole new class just to register register allocators and > > provide an option for picking one. But alias analysis already uses > > AnalysisGroup so a new class to register alias analysis passes isn't > >

At one point it handled function pointers. As chris says, it's buggy and probably broken. Some function pointers also do point to the universal set (like global function pointers). On Wed, Aug 26, 2009 at 10:11 PM, Chris Lattner<clattner at apple.com> wrote: > On Aug 26, 2009, at 6:40 PM, Max Stonebraker wrote: >> Hello, >> >> Does the LLVM Andersens alias analysis

Thanks for the reply. Would you happen to know whether there a global alias analysis written for LLVM that is more robust? Is the Steensgaard pass any more reliable? Max On Thu, Aug 27, 2009 at 10:01 AM, Daniel Berlin <dberlin at dberlin.org> wrote: > At one point it handled function pointers. > As chris says, it's buggy and probably broken. > Some function pointers also do

Hello, I am researching a more efficient approach to subset-based alias analysis in the style of Andersen's algorithm, and I noticed that there used to be an implementation of that for LLVM back in 2.6 (-anders-aa) which was removed because it was "not being actively maintained and had substantial problems". I'd be interested in knowing what was wrong with it (other than

Hi Damien, I think the problem is that when the passmanager sees an Alias Analysis available, it will use that, but when none is available, it will always run the default implementation (BasicAA), regardless of any alias analysis's run before. This means you should insert Andersen's before any pass that needs it. I suspect (not sure though) that the passmanager will not rerun

Hi all, Did someone forget to check-in a patch? I'm getting this error during compilation on PPC: /Volumes/SandBox/Clean/llvm-9999-01.roots/llvm-9999-01~obj/src/llvm/lib/Analysis/IPA/Andersens.cpp: In function 'void dumpToDOUT(llvm::SparseBitVector<128u>*)': /Volumes/SandBox/Clean/llvm-9999-01.roots/llvm-9999-01~obj/src/llvm/lib/Analysis/IPA/Andersens.cpp:1189: error: no

On Thursday 09 August 2007 19:21, Chris Lattner wrote: > Interesting question, I don't have an answer to this. To make things more > complicated, you can have multiple instances of an analysis group and may > want different things at different times: > > -basicaa -licm -something_that_invalidates_aa -andersaa -licm -whatever Some questions about that: How does this

Hi Prakash, I think this is highly related to PR2527 [1] and PR2451 [2]. I assume you are using LLVM 2.3, since current svn automatically replaces getresult with extractvalue instructions (but doesn't help for Andersen's). In either case, it seems that Andersen's currently has no ready made way to indicate a part of a first-class aggregate (or vector), which is the essence of the

On Sep 24, 2007, at 2:29 PM, Bill Wendling wrote: > Hi all, > > Did someone forget to check-in a patch? I'm getting this error during > compilation on PPC: A recent checkout compiled fine for me (on x86). > /Volumes/SandBox/Clean/llvm-9999-01.roots/llvm-9999-01~obj/src/llvm/ > lib/Analysis/IPA/Andersens.cpp: > In function 'void

A debug or release build? -bw On Sep 24, 2007, at 2:36 PM, Dale Johannesen wrote: > > On Sep 24, 2007, at 2:29 PM, Bill Wendling wrote: > >> Hi all, >> >> Did someone forget to check-in a patch? I'm getting this error during >> compilation on PPC: > > A recent checkout compiled fine for me (on x86). > >>

On Thu, 9 Aug 2007, David Greene wrote: >> intentionally does not publish all of the llvm options through cc1. We >> can add new options, but that has to be an explicit design decision. We >> do allow you to use some options through -mllvm -foo, but that won't help >> in this case. > Alias analysis is just the closest example of what I really need to do: >

On Aug 10, 2007, at 10:15 AM, David Greene wrote: > However, what happens when alias analysis information gets > invalidated? > Is Andersen's still "available" in the sense analysis groups use it? > > If not, it seems it would be tough to make sure Andersen's is always > used everywhere that AliasAnalysis is asked for because there are > passes > llvm

Sean Silva via llvm-dev <llvm-dev at lists.llvm.org> writes: > No tests fail with the patch below, so I would say it's pretty useless. It > seems that the C bindings are the only user but we can probably just have them > return IPSCCP instead. I don't necessarily think your conclusion is wrong, but the patch isn't proving what you think it's proving. In fact, the

On Aug 9, 2007, at 1:52 PM, David Greene wrote: > So how do I get opt to use Andersens instead of basicaa, for example? Following uses basic alias analysis $ opt -licm foo.bc -disable-output -debug-pass=Structure where as $ opt -anders-aa -licm foo.bc -disable-output -debug-pass=Structure uses Andersens analysis algorithm implementation. - Devang

Ok, here's a potentialy stupid question: how do I choose the various flavors of alias analysis on the command line? I followed the alias analysis design in its use of AnalysisGroup when I did the register coalescing refactoring and alternative implentations. But I can't figure out how to actually invoke the different versions. There's no cl_opt for the various alias analysis flavors