llvm dev - Apr 2016 - ArrayBoundChecks in SafeCode-llvm37

Hi,

I am wondering if anyone could run ArrayBoundChecks located in
SafeCode-llvm37 (https://github.com/jtcriswell/safecode-llvm37) on
llvm-3.8?

Thanks.
Syed

-- 
Rafi

On 4/27/16 12:00 PM, Syed Rafiul Hussain via llvm-dev
wrote:
> Hi,
>
> I am wondering if anyone could run ArrayBoundChecks located in
> SafeCode-llvm37 (https://github.com/jtcriswell/safecode-llvm37) on
> llvm-3.8?
As the 3.7 port isn't complete, I would guess that the answer is no.
:)

Within that directory, which pass do you need?  The 
BreakConstantGEPs.cpp pass would be easy to get working with LLVM 3.8 
(note: in general, I don't recommend using this pass, but there are rare 
situations for which it is needed).

ArrayBoundsCheckLocal would probably be easy. ArrayBoundsCheckStruct 
should be easy if you have DSA updated to LLVM 3.8.

The omega-based static array bounds checking code has long bitroted, and 
it's design could use significant improvement.  That code executed the 
omega program for every array it checked.  A more efficient 
implementation would link the constraint solver in as a library and 
query it via a function call.

If you can explain in more detail what you need, I can provide advice on 
what I think your best path forward will be.

Regards,

John Criswell
>
> Thanks.
> Syed
>

-- 
John Criswell
Assistant Professor
Department of Computer Science, University of Rochester
http://www.cs.rochester.edu/u/criswell

On 4/28/16 8:43 PM, Syed Rafiul Hussain wrote:
> Thanks Professor Criswell for your reply.
You're welcome.  Please keep replies on the list so that others can 
track and add to the conversation.
>
> I just need the constraintGeneration pass (i.e., interprocedural
> constraint generation pass) of ArrayBoundChecks.
The constraint generation code is really old.  It will need some 
significant work to get it working again.  There is also the question of 
whether it is better to generate constraints for Omega (the linear 
solver from Bill Pugh's group, IIRC) or whether generating constraints 
for CVC4 or Z3 would be better (I suspect Omega uses a better algorithm, 
but CVC4 and Z3 are better maintained; one could also adapt Omega to 
CVC4 if desired).

>   I have the DSA
> updated to LLVM 3.8 already.
Cool.  Can you put that into a Github repo somewhere?  Other people are 
wanting DSA for LLVM 3.8.
>   More specifically, I want to find out the
> post dominance frontier (pdf) for a particular basic block. I can find
> the pdf for a function. But, I need the pdf up to the main.entry.
Constraint generation is not needed for computing the post-dominance 
frontier.  You simply need to use the call graph from either DSA or the 
LLVM Call Graph pass and compute the post dominance frontier on your 
own.  It shouldn't be too difficult to implement, though call graph 
precision for code with function pointers could be an issue.

The old constraint generation code generates constraints for values that 
are stored in memory.  It won't help compute an inter-procedural 
dominance frontier.

Regards,

John Criswell
>
> Sincerely,
> Syed
>
>
>
>
> On Thu, Apr 28, 2016 at 7:41 PM, John Criswell <jtcriswel at
gmail.com> wrote:
>> On 4/27/16 12:00 PM, Syed Rafiul Hussain via llvm-dev wrote:
>>> Hi,
>>>
>>> I am wondering if anyone could run ArrayBoundChecks located in
>>> SafeCode-llvm37 (https://github.com/jtcriswell/safecode-llvm37) on
>>> llvm-3.8?
>>
>> As the 3.7 port isn't complete, I would guess that the answer is
no.
>> :)
>>
>> Within that directory, which pass do you need?  The
BreakConstantGEPs.cpp
>> pass would be easy to get working with LLVM 3.8 (note: in general, I
don't
>> recommend using this pass, but there are rare situations for which it
is
>> needed).
>>
>> ArrayBoundsCheckLocal would probably be easy. ArrayBoundsCheckStruct
should
>> be easy if you have DSA updated to LLVM 3.8.
>>
>> The omega-based static array bounds checking code has long bitroted,
and
>> it's design could use significant improvement.  That code executed
the omega
>> program for every array it checked.  A more efficient implementation
would
>> link the constraint solver in as a library and query it via a function
call.
>>
>> If you can explain in more detail what you need, I can provide advice
on
>> what I think your best path forward will be.
>>
>> Regards,
>>
>> John Criswell
>>
>>> Thanks.
>>> Syed
>>>
>>
>> --
>> John Criswell
>> Assistant Professor
>> Department of Computer Science, University of Rochester
>> http://www.cs.rochester.edu/u/criswell
>>
>
>

-- 
John Criswell
Assistant Professor
Department of Computer Science, University of Rochester
http://www.cs.rochester.edu/u/criswell

I have fixed the BottomUpCallGrap pass (in ArrayBoundChecks) for
LLVM-3.8 to make the call graph.

I will clean up my mess in the DSA and upload it to somewhere soon.

Thanks,
Syed

On Thu, Apr 28, 2016 at 10:18 PM, John Criswell <jtcriswel at gmail.com>
wrote:
> On 4/28/16 8:43 PM, Syed Rafiul Hussain wrote:
>>
>> Thanks Professor Criswell for your reply.
>
>
> You're welcome.  Please keep replies on the list so that others can
track
> and add to the conversation.
>
>>
>> I just need the constraintGeneration pass (i.e., interprocedural
>> constraint generation pass) of ArrayBoundChecks.
>
>
> The constraint generation code is really old.  It will need some
significant
> work to get it working again.  There is also the question of whether it is
> better to generate constraints for Omega (the linear solver from Bill
Pugh's
> group, IIRC) or whether generating constraints for CVC4 or Z3 would be
> better (I suspect Omega uses a better algorithm, but CVC4 and Z3 are better
> maintained; one could also adapt Omega to CVC4 if desired).
>
>
>>   I have the DSA
>> updated to LLVM 3.8 already.
>
>
> Cool.  Can you put that into a Github repo somewhere?  Other people are
> wanting DSA for LLVM 3.8.
>
>>   More specifically, I want to find out the
>> post dominance frontier (pdf) for a particular basic block. I can find
>> the pdf for a function. But, I need the pdf up to the main.entry.
>
>
> Constraint generation is not needed for computing the post-dominance
> frontier.  You simply need to use the call graph from either DSA or the
LLVM
> Call Graph pass and compute the post dominance frontier on your own.  It
> shouldn't be too difficult to implement, though call graph precision
for
> code with function pointers could be an issue.
>
> The old constraint generation code generates constraints for values that
are
> stored in memory.  It won't help compute an inter-procedural dominance
> frontier.
>
> Regards,
>
> John Criswell
>
>>
>> Sincerely,
>> Syed
>>
>>
>>
>>
>> On Thu, Apr 28, 2016 at 7:41 PM, John Criswell <jtcriswel at
gmail.com>
>> wrote:
>>>
>>> On 4/27/16 12:00 PM, Syed Rafiul Hussain via llvm-dev wrote:
>>>>
>>>> Hi,
>>>>
>>>> I am wondering if anyone could run ArrayBoundChecks located in
>>>> SafeCode-llvm37 (https://github.com/jtcriswell/safecode-llvm37)
on
>>>> llvm-3.8?
>>>
>>>
>>> As the 3.7 port isn't complete, I would guess that the answer
is no.
>>> :)
>>>
>>> Within that directory, which pass do you need?  The
BreakConstantGEPs.cpp
>>> pass would be easy to get working with LLVM 3.8 (note: in general,
I
>>> don't
>>> recommend using this pass, but there are rare situations for which
it is
>>> needed).
>>>
>>> ArrayBoundsCheckLocal would probably be easy.
ArrayBoundsCheckStruct
>>> should
>>> be easy if you have DSA updated to LLVM 3.8.
>>>
>>> The omega-based static array bounds checking code has long
bitroted, and
>>> it's design could use significant improvement.  That code
executed the
>>> omega
>>> program for every array it checked.  A more efficient
implementation
>>> would
>>> link the constraint solver in as a library and query it via a
function
>>> call.
>>>
>>> If you can explain in more detail what you need, I can provide
advice on
>>> what I think your best path forward will be.
>>>
>>> Regards,
>>>
>>> John Criswell
>>>
>>>> Thanks.
>>>> Syed
>>>>
>>>
>>> --
>>> John Criswell
>>> Assistant Professor
>>> Department of Computer Science, University of Rochester
>>> http://www.cs.rochester.edu/u/criswell
>>>
>>
>>
>
>
> --
> John Criswell
> Assistant Professor
> Department of Computer Science, University of Rochester
> http://www.cs.rochester.edu/u/criswell
>


-- 
Rafi