[LLVMdev] llvm alloca dependencies

Alexandru Ionut Diaconescu alexandruionutdiaconescu at gmail.com
Thu Jan 24 05:54:55 PST 2013

Hello again,

I want to ask you some piece of advice if it is possible.

I tried methods related to point 1) suggested by you, but I still have
problems of finding dependencies. What exactly I want to do:

I have a chain like : Alloca -> Load(1) -> ... -> Computation where the
variable might be changed -> Store(new_var) -> ... -> Load(n)

Computation where the variable is changed = means that : I might have
Alloca(a),  c=a+7000*b[32], Load(c)...ICMP(c, ...)

I want to get the corresponding Alloca from every Load (corresponding from
the point of view of the variable used, meaning that the Load is using a
variables based/dependent on the Alloca or Allocas).

First, I tried the point 1) suggested by you. AliasSetTracker uses methods
from AliasAnalysis, so I tried to use directly those methods.
After I see all the must-aliases, I categorize them into 2 :
1. category A : aliases with allocas
2. category B: aliases without allocas
For category A, is straight ahead for what I need.
For category B, I check if there is an instruction where the variable is
used also from an instruction from the aliases from category A. If it is,
it is ok.

But the following given me no result at all (and they should, I have
dependencies - meaning Load j is used multiple times in my code, so the
pointers should be must-alias) I have all the Load instructions in an array
loadInstrArray. :

1. if( AA.isMustAlias(Loci,Locj) ) - no result
2. if( AA.alias(Loci,Locj) ) - wrong results, like "LOAD   %2 = load i32*
%j, align 4 IS DEPENDENT ON   %3 = load i32* %c, align 4"
                                         where j is totally independent
from c
3. if( AA.getModRefInfo(allocaInstrArray[j],Loci) ) - no result

Second, I tried to use methods from DependencyAnalysis class.

1. if (DA.depends(loadInstrArray[i],loadInstrArray[j],false)) - no result

Third, I tried methods from MemoryDependenceAnalysis class - the existing
pass http://llvm.org/docs/doxygen/html/MemDepPrinter_8cpp_source.html .

1. I have wrong results, like : %1 = load i32* %j, align 4    IS Clobber
FROM:      store i32 0, i32* %c, align 4
2. then I tried to get only DEF (not clobber), since I only need to see the
correspondent Alloca for every Load. I don't have any result for Loads. I
also checked with getDef() method, so Loads are dependent on themselves
3. I have to mention that I commented line 00131, since I had an unsolved
segfault and not the parameters are the problem.

Fourth, I took a look in to DSNodeEquivs pass suggested by you. I think
that getMemberForValue() method I can use. To be honest, I tried to focus
on the classes that are already existent on LLVM.

What do you think I should focus on, what approach is better to take into
account and what to eliminate?

Thank you a lot for your time !

On Tue, Jan 22, 2013 at 4:57 PM, John Criswell <criswell at illinois.edu>wrote:

>  On 1/21/13 5:22 AM, Alexandru Ionut Diaconescu wrote:
> Hello everyone !
> I am trying to determine for certain Load instructions from my pass their
> corresponding Alloca instructions (that can be in other previous blocks).
> The chain can be something like : `TargetLoad(var) -> other stores/loads
> that use var (or dependencies on var) -> alloca(var).` , linked on several
> basic blocks. Do you know how can I do it?
> This sounds like an alias analysis query that you're trying to do.  There
> are a couple of options:
> 1) Cycle through all the allocas and ask if they alias with the loaded
> pointer.  You can use any of the alias analysis implementations available
> for LLVM.  Alternatively, I think LLVM might have a class called
> AliasSetTracker which can do something similar for you.
> 2) Use DSA from the poolalloc project.  The DSNodeEquivs pass can be used
> to determine if a load and an alloca alias.  Depending on your situation,
> DSA may give better or worse results than LLVM's current alias analysis
> algorithms.  I would think that DSA can do a better job for pointers stored
> into memory locations.
> -- John T.

Best regards,
Alexandru Ionut Diaconescu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130124/90ab36ce/attachment.html>

More information about the llvm-dev mailing list