[LLVMdev] Extracting the global variables accessed by individual function with function pass

[Yan Lin Aung]

Hi Duncan and John,

Thank your very much for the advice and help.

Yes, I have got the first part working following Duncan's advice.

I think I can also incorporate Alias Analysis (AA) in my code.

A bit of guidance on how the LLVM provided methods in AA have to be invoked so as to get the info on 
the global being passed into the function as an argument or loaded from memory via an LLVM load instruction as John pointed out?

Thanks again.

Regards,
Yan Lin Aung



________________________________
 From: John Criswell <criswell at illinois.edu>
To: Duncan Sands <baldrick at free.fr> 
Cc: llvmdev at cs.uiuc.edu 
Sent: Friday, September 6, 2013 11:41 PM
Subject: Re: [LLVMdev] Extracting the global variables accessed by individual function with function pass
 

On 9/2/13 6:15 AM, Duncan Sands wrote:
> Hi,
>
>> Is there a way to dump the global variables, which are accessed 
>> (read/written
>> to) by each function within the LLVM function pass?
>> Any pointers on how that info can be extracted in LLVM for each 
>> function in the
>> application?
>
> you will have to write your own pass to do this.  For each global, 
> visit each
> of its uses (via use_begin, use_end).  If the use is a constant 
> (isa<Constant>),
> visit each of its uses too (do this recursively until you aren't visiting
> constants any more).  If a use isn't a constant, check whether it's 
> parent
> (getParent) is the function you are interested in.  The point of 
> recursing on
> constant uses is that the global may be used in a constant expression 
> (eg a
> constant bitcast) before actually being used by an instruction.

Note that this only finds globals that are directly accessed by the 
function.  It won't find globals that are accessed by the function due 
to the global being passed into the function as an argument or loaded 
from memory via an LLVM load instruction.  To find those, you'll need to 
use alias analysis or a points-to analysis.

-- John T.

>
> Ciao, Duncan.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu        http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

_______________________________________________
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/20130908/5b94a714/attachment.html>